Thinks 2147

FT: “This is an era of Davids, not Goliaths…Why smaller powers are thwarting bigger ones.”

Andy Kessler: “AI architecture is evolving. Instead of a trillion parameters, efficient small language models might use 70 billion and run on an Apple Mac. These local models have no token costs beyond the cost of hardware and electricity. But they must be less capable than the large language models, right? Not really. Stanford researchers show that a locally run small language model “can successfully answer 88.7% of single-turn chat and reasoning queries.” No question we’ll always have high-end models running in data centers. IBM still sells mainframes. But growth will happen elsewhere. Where? Another Silicon Valley saying I’ve lived by: Intelligence moves out to the edge of the network. The real breakthrough will happen when robust AI models run on smartphones. It’s already starting. In September, California-based PrismML released a compressed version of the popular open-source Qwen3.8 27B AI model. The company got it down to under 6 gigabytes. That means it fits not only on PCs but also on high-end smartphones. The memory size is reduced by roughly 9 to 10 times. There’s that factor of 10.”

NYTimes: “Strength training isn’t just for muscle and bone health — it’s also surprisingly good cardio.”

Mint: “Over the past two decades, Indian households have undergone an extraordinary material transformation. Our household data shows electricity access rising from 75% in 2004-05 to an estimated 99% in 2025-26. Toilet access has increased from 52.9% to 95%, Liquefied petroleum gas (LPG) or piped natural gas (PNG) from 32.3% to 85% and tap water from just 19.4% to 80%. Refrigerator ownership has risen from 18.5% to an estimated 65% while air- conditioner ownership, almost negligible at 1.2% two decades ago, has reached 21.1%. Mobile phones, present in only 16.1% of households in 2004-05, are now almost ubiquitous. These are not cosmetic changes in consumption. They represent a profound improvement in everyday life. Development has entered the Indian household, its kitchen and bathroom, communications, mobility and comfort. Yet, beneath this transformation lies a less visible change: the face of disadvantage has changed faster than the structure of disadvantage.”

The Publisher’s Second Business

From telling readers what happened to helping them decide what to do next

The argument in one paragraph. For a century, publishers built their businesses around one job: telling people what happened. Readers have two more jobs, deciding what it means for them and doing something about it, and most of the value in those went elsewhere. Publishers once held part of it, through classifieds, and lost it to better software. AI now makes it cheaper to turn editorial knowledge into tools and services, but the same assistants that keep the answer also want the decision and the action. A publisher can build a second business only where it holds something the assistant lacks, can run it reliably, and can find someone who will pay. It need not own the screen: it can build the tool, supply the assistant, or partner with whoever completes the task. What matters is not who controls the interface, but who owns something the decision cannot do without.

1  The Business Newspapers Already Lost

For most of the twentieth century, a newspaper ran two businesses under one masthead. The first told readers what had happened. The second helped them act: find a job, rent a flat, sell a car, hire a plumber. That second business was classified advertising, and it was enormous. In 2000, American newspapers earned $19.6 billion from classifieds, about 40% of all their advertising revenue. Recruitment ads alone brought in $8.7 billion.

Within a decade most of it had gone. By 2009, classified revenue had fallen to roughly $6 billion; by 2012 it was $4.6 billion, down about 77%. Help-wanted advertising, the most valuable category, lost more than 90%. Many newspaper groups saw it coming and invested in online sites such as Cars.com, CareerBuilder and Apartments.com, which, as a Congressional Research Service report notes, drew classified advertising away from the papers’ own pages. Reach and a famous masthead were not enough. The winners had better software: search, filters, alerts and a market that grew more useful with every listing.

Figure 1. Publishing’s first second business, and how fast it went.

Publishers have had a second business before. They lost it because matching people with things went to whoever built the best software.

Some media groups did hold on to large second businesses, and those businesses grew bigger than the journalism. When Axel Springer split itself in two, its classifieds businesses, including the StepStone jobs platform and the Aviv property group, accounted for about €10 billion of a €13.5 billion valuation. Schibsted sold its newspapers to its main owner for about $570 million; its marketplaces stayed listed as a separate company. The Washington Post Company earned $2.9 billion, or 62% of its revenue, from its Kaplan education business in 2010, and when it sold the newspaper in 2013, it kept Kaplan.

These cases prove less than they seem to. Kaplan was bought, not built from the newsroom; it entered higher education by acquiring a chain of colleges. Marketplaces won on network effects and software, not on reporting. They show that media companies can own valuable businesses beyond journalism. They do not show that journalism itself can be turned into one.

Publisher Second business Built from the journalism? What happened
US newspapers Classified advertising Partly: local reach, not newsroom knowledge Lost most of it to specialist marketplaces
Axel Springer Job and property marketplaces No: separate marketplace businesses About three-quarters of group value; separated from the newsroom
Schibsted Online marketplaces No: marketplace software Newspapers sold to the founding trust
Washington Post Co. Education No: acquired Grew to 62% of revenue; newspaper sold, Kaplan kept
Dow Jones Professional information Closer: news archives, data and expertise About 40% of revenue and growing (section 4)

Table 1. Most successful second businesses were acquired or built as separate software businesses.

The distinction matters for what follows. A publisher can acquire a second business, as media groups have done for decades. The more interesting opportunity is to build one from the knowledge, relationships and expertise that publishing itself creates. History offers few examples of the second kind. Diversification of the first kind is already well under way: according to WAN-IFRA, revenue from “other” activities such as events and business services reached 25.4% of surveyed publishers’ revenue in 2025, up from 13.2% in 2021. This essay is about why knowledge-led second businesses may now become possible, and when they make sense.

2  Know, Decide, Do

Every reader has three jobs. Know: what happened? Decide: what does it mean for me? Do: what is my next step? Publishing has concentrated on the first, and the first has a public value no business model fully captures. But commercially, the middle job went to whoever stood nearest the decision: the search box, the comparison site, the broker, the adviser, the software vendor. Much of the third went to marketplaces.

Consider a local paper reporting that the city’s schools have changed their admission rules: new catchment boundaries, a lottery for oversubscribed schools, earlier deadlines. That is Know, and it is good journalism. But a parent reading it has a bigger problem. Which schools is my child now eligible for? How do they compare on the things I care about? When must I apply, and what documents do I need?

The paper already holds much of the knowledge to answer those questions. It has reported on every school for years. It knows the rules and the officials who wrote them. Yet it publishes the story and stops, and the parent takes the next step somewhere else, or alone.

Figure 2. Three jobs. Publishers kept the first; others captured the value in the next two.

The article informs the decision. The larger commercial opportunity often lies in helping the reader make it.

3  What Publishers Already Hold

A publisher setting out to help readers decide and act does not start from nothing. It holds six assets that most new entrants would have to build or buy:

  • Damaged, but still real. Trust in news stands at 37% across the markets the Reuters Institute surveys, a record low, yet a known masthead still carries more authority than an anonymous site.
  • Specialist expertise. Reporters who have covered schools, markets or one industry for years know where the traps are.
  • Verified archives. Decades of checked facts, figures and context.
  • Readers who come back on their own and have said what they care about.
  • Local knowledge. People on the ground who know which rule changed and which provider cut corners.
  • Convening power. The ability to bring people facing the same decision together.

Assets are not advantages until they are tested. An archive that any assistant can summarise is not proprietary intelligence. Trust in a headline is not permission to give regulated advice. And audience size is missing from the list on purpose: newspapers had huge audiences when they lost classifieds. An audience is a reason to try a second business, not a reason it will work.

The right question is not “what can we sell to our readers?” but “what can we help them do better than anyone else?”

4  When Knowledge Becomes Software

Until recently, turning editorial knowledge into a product meant hiring software teams and data engineers and waiting years. Only the largest publishers could afford it, and those that did built some of the best businesses in the industry. Dow Jones’s professional information arm, which sells risk, compliance and energy intelligence to companies, now accounts for around 40% of Dow Jones revenue, and management says a significantly greater share of its profit. Its Risk & Compliance unit grew 19% to $100 million in a single quarter this year. It is the clearest case of a knowledge-led second business at scale, built on news archives, data and editorial judgement.

AI changes who can try. A regional publisher can now build a working prototype of an answer service over its archive, a calculator that applies this year’s rules or a checklist that knows the local deadlines, in weeks rather than years. That is a prototype, not a product. Turning it into a service people rely on still means ingesting and verifying data, updating it when rules change, supporting users, securing their information and clearing the rights. Those costs can dwarf the cost of the first demonstration.

So AI does not remove the hard part; it moves it. Building gets cheaper; maintaining matters more. A wrong answer in a decision tool costs the reader far more than a wrong fact in an article. Accuracy, provenance and human oversight become the main operating cost of a second business, and its main source of value.

AI’s biggest opportunity for publishers may not be cheaper journalism. It may be more businesses that journalism can support.

5  The Assistant Wants the Same Job

There is an obvious objection, and it could sink the argument. The companies building AI assistants say plainly where they are heading. When OpenAI let users buy products inside ChatGPT in September 2025, it described AI as becoming “a key interface for how people discover, decide, and buy”. That is Know, Decide and Do in a single sentence. Assistants that summarise the news will also compare schools, model a loan and book an appointment. If they keep the answer, why would they not keep the decision and the action too?

Often they will. Generic decisions, such as the cheapest flight or the best-rated vacuum cleaner, will go to whoever has the widest reach and the best interface. That is the classifieds lesson again. Publishers without a differentiated asset have little reason to compete head-on with general-purpose assistants. Three assets are harder for an assistant to supply:

  • Verified, proprietary or local data. This year’s admission rules for one city’s schools, the cut-offs for a regional exam, the tradespeople who did the job properly. Information gathered and checked on the ground is often not on the open web, or not reliably.
  • A named institution that answers when it is wrong. When the stakes are money, health or a child’s education, people want someone answerable, not just an answer.
  • People who trust each other through the publisher: parents in the same district, investors in the same sector. Much of the value comes from the others in the room.

Holding one of these is necessary but not sufficient. A publisher may have trusted reporting on schools but no accurate, current admissions data. It may have a community but not the expertise to run a service reliably. It may be accountable but unable to serve thousands of users economically. A second business has to pass three tests:

  1. Do we hold data, accountability or community that an assistant cannot cheaply reproduce?
  2. Can we turn that advantage into a reliable, maintained product?
  3. Will readers or institutions pay enough, repeatedly, to make it profitable?

Figure 3. Three tests for a second business, and what to do when one fails.

Passing the first test creates an opportunity. Passing all three makes it a plausible business.

Job Assistant’s advantage Publisher’s possible edge Where the publisher stands
Know: general news summary Reach, speed, convenience Original reporting, provenance Best as a licensed source
Decide: generic product choice Breadth, comparison at scale Little Stay out
Decide: local or high-stakes choice Convenience Verified data, named experts Can compete, if all three tests pass
Do: generic purchase Checkout inside the conversation Little Stay out
Do: local service or application Convenience Verified providers, community Can compete, often through a partner

Table 2. The assistant’s default advantage, and where a publisher can still compete.

6  The Interface Isn’t the Business

Return to the parent. Suppose a regional publisher keeps the most accurate, current information about local schools: rules, catchments, deadlines, results. The parent does not visit the publisher’s site. She asks her assistant which school her child is eligible for. Three outcomes are possible.

  • The publisher runs its own eligibility checker and deadline tracker, and earns from readers who use it.
  • The assistant answers using the publisher’s verified data, under a paid licence, with the source named.
  • The publisher guides the parent, and an accountable specialist, such as a school-admissions service, completes the application, with any commercial relationship disclosed.

Only the first requires the parent to come to the publisher’s own screen. As assistants become the place where decisions start, the second and third may matter more. Supplying is different from licensing an archive for training: what is sold is live, maintained, verified knowledge, priced on use, and its value depends on being kept right. That is a business an assistant cannot easily build for itself, because it rests on people on the ground and an institution that answers for errors.

Figure 4. Three ways to turn trusted knowledge into a second business.

What matters is not who controls the interface, but who owns something the decision cannot do without.

7  Four Second Businesses

Four domains show where the tests are passed most often. Two deserve depth, because they show both the promise and the hardest question: who pays.

Professional intelligence is the proof that a knowledge-led second business can be large. Industry news tells a manager what changed. The second business tells the company what a new regulation means for it, whether a counterparty is safe and what to watch next, delivered inside the work through screening, monitoring and alerts. The advantage is proprietary, structured data plus accountability, and the customer is a company that pays for reliability because errors are expensive. Dow Jones shows the scale this can reach. It also shows the cost: the data must be maintained continuously, enterprise buyers are demanding, and liability comes with the territory. It proves the category, not that a small newsroom can copy it.

Education shows where the economics get hard. Education pages already report exam dates, results and policy changes. The second business helps a student or parent decide which exam, course or school fits, then do the work: eligibility checks, practice, assessments, applications and learning plans. The advantage is verified local data and a high-stakes need. But who pays for a school-admissions tool? A parent may use it intensely for a few weeks and never again. Schools paying for placement create an immediate conflict. The options each change the business: a membership that bundles the tool with other services, a contract with an education authority, sponsorship kept visibly separate from rankings, or supplying the verified data to the services and assistants parents already use. A reader’s need does not automatically mean a willingness to pay.

Finance follows the professional model for consumers: calculators, plain-language explainers and investor education, with next steps routed to licensed partners. Its boundary is regulation; in most markets, explaining is journalism, but personalised investment advice needs a licence. Local turns civic reporting into verified directories, bookings, events and memberships. Its advantage is local data and community; its risk is becoming one more directory, so “verified” has to mean something.

Domain Decide Do Likely model Who pays
Professional What this means for my company Screening, monitoring, alerts Build Enterprises, through licences
Education Which exam, course, school Eligibility, practice, applications Build, supply or partner Hard: members, institutions, licensees
Finance What this means for my money Next steps via licensed partners Build and partner Subscribers; disclosed partner fees
Local Which contractor, clinic, school Verified directory, bookings, events Build and partner Members, verified providers, events

Table 3. Four second businesses, each built on knowledge the publisher already holds.

8  The Trust Tax

Every second business tempts a publisher to spend the trust that makes it possible. A lender will pay for leads. A school will pay to be listed first. A retailer will pay a commission on every sale. Readers may trust a publisher’s journalism without wanting it to become their broker or their adviser.

Kaplan is the cautionary case, even though it was acquired rather than built from the newsroom. A Senate committee found that Kaplan’s higher-education business drew 85.9% of its 2010 revenue from federal student aid, and 87.9% including military education benefits. The same investigation documented aggressive recruiting and high withdrawal rates, while also crediting the company with significant later reforms. Meanwhile the parent company’s chief executive led the industry’s effort to soften new federal rules, as his newspaper covered the same policy fight. A second business had put the first in an awkward position.

The boundaries are not complicated, but they have to be built in from the start:

  • Editorial independence. The newsroom does not write for the second business’s clients.
  • Every commercial relationship and ranking method is visible at the point of decision.
  • Reader data is used only for what the reader agreed to.
  • Information, advice and advertising are labelled and kept apart.
  • Licences and oversight. Where advice is regulated, the publisher holds the licence or works through someone who does, with expert review and a visible process for complaints and corrections.

The hardest case is a service that succeeds commercially while trust in the newsroom falls. A publisher should decide in advance which trust measures would trigger a review, and be prepared to change or close the service if they move.

Trust is the asset only if the second business does not spend it.

9  Who Can Afford a Second Business?

Smaller publishers are under the most pressure. Over two years, small publishers lost about 60% of their search referrals, mid-sized ones 47% and large ones 22%. But pressure is not capability. Many smaller publishers lack product teams, capital, proprietary datasets or enough customers to support a software service. Some will do better through membership, philanthropy or specialist reporting than through any second business.

For those that do have an advantage, three structures are feasible without a large balance sheet: a focused local utility that does one job well; a shared service built with peers, so that several publishers spread the cost of the data and the technology; or a supply or partner arrangement that turns their verified knowledge into revenue without building a product at all.

Smaller publishers may need a second business most, but need is not evidence that they can build one.

10  The Economics of the Second Business

The test is contribution, not traffic. Contribution means revenue minus the costs that a second business adds: maintaining data, expert review, technology, support, liability, distribution and acquiring customers. A decision tool used by a few thousand people who renew every year can be worth more than an article read by a million, or it can lose money on every user. Only the arithmetic, against a fair baseline, will say which.

Revenue comes in several forms: subscriptions to tools, service fees, licences to businesses and assistants, disclosed partner fees, and events and memberships. Most second businesses will combine two or three.

The organisation sits beside the newsroom, not inside it. Journalists supply the knowledge; product builders turn it into tools; editorial standards govern both. Start narrow: one decision readers face repeatedly, with high stakes, where all three tests can plausibly be passed.

Governance needs deciding early. Table 1 showed what happens when a second business succeeds: when it is worth more than the journalism, investors want the two apart. A publisher should decide at the outset whether its second business exists to fund journalism, and design ownership accordingly.

11  What Would Prove This Wrong

Each of these tests can be run within 12 to 24 months, against a suitable comparison group:

Hypothesis Evidence that would challenge it
Publishers hold a real advantage Users of a publisher’s decision tool report no added value over a general assistant, and usage falls as assistants add the same features
Someone will pay Free use is high, but trials do not convert into recurring payments, licences or other revenue that covers costs
AI lowers the cost enough Verifying, updating and supporting the product consume most of its gross profit
Trust survives commerce Trust in the publisher falls among users exposed to the service, relative to those who are not
The interface need not be owned Publishers cannot earn from their knowledge either directly or by supplying assistants and partners
It is not only for giants After 24 months, profitable examples remain limited to the largest publishers

Table 4. Six falsification tests.

12  The Publisher of 2030

The publisher of 2030 will still be defined by its journalism. Reporting earns the trust, the expertise and the relationships that every second business depends on. Without the first business, there is no second one.

But journalism no longer needs to be the only product. Not every publisher will become a software company; not every archive will become a product; not every reader wants a publisher to act on their behalf. Those with a clear right to win will turn their knowledge into something readers and institutions can use: to choose a school, manage their money, run a business or find their way in their own city. Some will own the service. Others will power it from behind an assistant, or through a partner. The question for 2030 is not how many articles a publisher produces, but what people can reliably do because that publisher exists.

The first business of publishing was making information worth reading. The second may be making knowledge worth using.

Thinks 2146

FT: “The AI trade is no longer new and it is no longer “a” theme. Instead it has morphed into “the” theme — a vast, suffocating blob slathered on top of the global financial system, starving everything else of oxygen, distorting market reactions and leaving every portfolio, everywhere, alarmingly reliant on the fortunes of just one bet. Stocks, real estate, infrastructure, energy, emerging markets, some vintages of private equity . . . they’re all, to a large extent, the same trade.”

Mint: “Two reforms can help India move from a slow, expensive manufacturing destination to a fast, cheap one: averaging regular hours over a longer period and reducing the overtime premium.”

WSJ: “When we eat, the energy in the food is shared between our own cells and the trillions of microbes that inhabit our intestines. How much energy we absorb depends on the quality of the food and our gut microbiome. Controlled trials show how important this can be. On highly processed diets, the body can absorb roughly 97% of consumed calories. On minimally processed diets rich in intact foods, absorption can fall to 85% or lower. For a 2,500-calorie diet, that difference is about 300 calories a day. Yet the calorie counts appear identical. Calorie totals also don’t tell us where the absorbed energy goes. The quality of what we eat influences hormones, immune function, inflammation, insulin signaling and our gut bacteria. Together, these systems determine how nutrients are partitioned in the body.”

NYTimes: “Fourteen authors in the New Directions stable have received the Nobel Prize in Literature, including this year and last. How does this small press pack such a punch?”

Sadanand Dhume: “China still holds lessons for India—and any democracy. Julian Gewirtz, a China expert at Columbia University, calls the country’s economic takeoff “as much a story of bottom-up experimentation and exploration of international ideas as anything else.” By embracing meritocracy, China has made its universities world-class. China sees its cities as engines of growth, not merely cash cows for rural hinterlands. But these are arguments for improving Indian democracy—so it may one day offer the developing world a better example than China.”

Making AI Affordable for All

Private, Useful AI for Every Employee, at Emerging-Market Prices

**

What would it take to give every employee in an emerging-market company the AI tools that a few people in a rich company take for granted?

AI has an access problem hidden inside a pricing problem. The tools that make office work faster are priced in dollars, per seat and increasingly per task, for buyers in the richest economies. The leading office-suite AI add-on lists at $21 to $30 per user per month, on top of the base licence. A company in Mumbai, Jakarta or Lagos often pays much the same dollar price for employees whose salaries are a fraction of their counterparts’ in New York or London. So it buys a few seats for senior people, and everyone else goes without, or uses whatever free tool they can find.

This essay completes a series of three. The Compute Strait asked how much inference a job needs. The Compute You Already Own asked where the inference that remains should run. This one asks who gets it, and at what price. The goal is not cheaper AI for a few. It is useful, private AI for every employee, at a price an emerging-market employer can sustain.

My answer is not a set of cheaper substitutes for famous AI products. It is a shared utility inside the enterprise: a few everyday AI jobs, joined by one layer that remembers the work, running first on compute the company already owns, under rules the company sets. The rest of the essay explains why that combination could cost a tenth of today’s prices, what competes with it, what it would take to make the economics work for the supplier as well as the buyer, and how it could fail.

Figure 1. Three essays, three questions

1   AI is priced for the few

Run the arithmetic for a company of 500 people. At $20 per employee per month, below today’s list prices for the leading business add-ons, covering everyone costs $120,000 a year. For a services firm in India, that is the salary of several employees, spent on software for all of them. Few companies make that trade. They ration instead: seats for managers and a handful of enthusiasts, and nothing for the support agent, the operations clerk or the sales coordinator who would gain the most from help with routine writing and reading.

Figure 2. Covering every employee in a 500-person company (prices assumed)

When AI is priced per seat in dollars, an emerging-market company does not decide how to use it. It decides who may use it. Metered pricing makes this worse. A finance team that cannot predict the bill caps the usage, and the cap falls hardest on the people with the least authority to ask for an exception.

The market is not standing still, and an honest argument has to start there. The largest office-software vendors are folding AI into what companies already buy. One has bundled its AI assistant into its business plans, funded by a base-price increase of $2 to $4 per user a month rather than a separate add-on; the other has offered promotional prices on its business AI add-on. At the same time, employees without a company seat do not wait. Many paste work into free consumer tools, with no governance over what leaves the company or what comes back. The competitor is not only the $30 seat. It is AI bundled into software a company already pays for, and free tools its employees use without asking.

So the opportunity is not to introduce AI where none exists. It is to offer a coherent, governed and affordable alternative to fragmented tools and uncontrolled use. A bundled assistant serves the company that runs its whole office on one vendor’s suite. Many emerging-market firms do not: email may sit with one vendor, documents with another, accounts in a local system, customers on a messaging app. What they lack is AI that works across that mixed environment, keeps the company’s memory under its own control and costs a predictable amount. That is the gap this essay is about.

2   Four jobs, one suite

Most everyday AI use in a company reduces to four verbs. Ask: a private assistant that answers questions from approved company knowledge. Act: a mail and work agent that drafts replies and follow-ups, with nothing sent without a person’s yes. Listen: meetings transcribed and written up, ideally on the machine that recorded them. Watch: company records checked against rules, with each exception sent to its owner, with evidence.

None of these is new on its own. Self-hosted chat over company documents is already available as open-source software, and cheap tools exist for each of the other three. Four inexpensive applications would not be much of a business. What changes the picture is the work that passes between them.

Take a single meeting. On Tuesday morning, a team agrees three actions. The Listen job writes up the meeting and records who promised what, by when. The Act job drafts the three follow-ups, and each goes out only when its owner approves. On Friday, the Watch job checks the records: two actions are done, one has slipped, and its owner is told, with the evidence. The following week, a manager asks where Tuesday’s actions stand, and the Ask job answers from the same records, showing only what that manager is allowed to see. Nobody copied anything from one application to another.

Figure 3. One meeting, four jobs, one shared record

Separate subscriptions can be connected, of course, through integrations. But integration across four vendors is usually partial, expensive to maintain and fragile when any one of them changes. A shared architecture makes the connection the default rather than a project. The value is not four inexpensive AI applications. It is one inexpensive system that remembers what happened across them and helps finish the work.

What runs underneath matters less than buyers assume. The leading frontier models are proprietary; their weights cannot be installed on an office server. Open-weight models can be, and they are closer than most buyers think. Epoch AI finds that the best open-weight models lag the closed frontier by about four months on its capability index, and that open models able to run on a single top-end consumer graphics card typically match the frontier of six to twelve months earlier. Last year’s frontier now fits on a well-equipped desk. Not on every desk: that finding is about a high-end graphics card, not an ordinary office laptop. But narrower models doing narrower jobs reach far more modest hardware, and for summarising a meeting, drafting a reply or answering from a policy manual, that is often enough. It has to be proved task by task rather than assumed. The company is buying completed work, not a model name.

3   The office already owns much of the infrastructure

Every office already owns an AI computing infrastructure. Most of it has never been organised to work together. It sits on desks: well-specified laptops and desktops, some with enough memory to hold a capable open model, idle for much of the day while their owners are in meetings, at lunch or at home.

The architecture that uses it has three parts, and the order of importance matters.

  • An always-on office server provides the service guarantee. It routes every question, holds the models and their updates, and runs anything sensitive. If every employee laptop were switched off, the service would still work.
  • Idle employee machines contribute optional capacity. They take eligible, independent tasks while their owners are away and the machines are plugged in, each running a whole model so that nothing is split across machines.
  • The router uses an idle machine only when it helps. It sends work there only when the gain is measurable and the owner is unaffected; otherwise the server does the work.

In The Compute You Already Own I set out why such an office pool is plausible where a public pool of strangers’ devices is not, and the questions that decide whether it pays. The most important is timing. What matters is not total idle computing power. It is dependable compute available in the hours employees need AI. An office might have twenty idle machines at lunchtime and three at eleven in the morning, precisely when questions peak. If one dedicated server handles the load more cheaply, use the dedicated server. The business does not depend on proving that distributed office compute always wins.

It is tempting to call this an “Airbnb for compute”. As an image, it works: an underused asset put to productive use. As an operating model, it misleads. Airbnb rents strangers’ property to strangers. This is a private pool inside one company, among colleagues, managed by IT.

The economics need stating carefully too. On hardware the company already owns, the marginal compute cost of an answer is mostly electricity. The service around that compute is not free. Models have to be kept ready in memory, machines’ availability detected, work moved when a machine disappears, employee endpoints secured and failures diagnosed. Those costs belong in any honest comparison, and they are the reason the dedicated server comes first.

4   The meta-model decides what runs where

If every question goes to a large model, the system is expensive. If every question goes to a tiny one, it fails on the hard cases. The answer is a meta-model in the practical sense: a routing layer that chooses the cheapest way to complete each task correctly. It follows the ladder from The Compute Strait. Code handles dates, sums and rules. Validated memory answers what is already settled. A small local model extracts, classifies and drafts. The office server’s larger model handles harder shared work. A premium external model is called only for the cases that need it, and only where policy allows that data to leave the approved environment.

The router must itself be cheap: rules, capability checks and past evaluations first, never an expensive model asked whether an expensive model is needed. And it should record what it learns: completion, errors, corrections, response time and cost per completed task, by type of work. Over time the company comes to own an evidence-based map of what its own work requires.

Figure 4. One suite, one shared layer, compute the office owns before compute it rents

5   A tenth of the price, if the economics hold

The commercial target should be bold: roughly a tenth of developed-market per-seat prices, about $2 to $3 per employee per month for everyday workplace AI. Four things make it plausible. Open-weight models carry no per-token licence. Much of the compute is already owned. One shared layer serves four jobs instead of four vendors each building their own. And frugal routing keeps premium spending to the cases that need it.

A low price has to work for the supplier as well as the buyer, and that is where the argument is most exposed. At $2.50 per seat, a 500-person company brings in $1,250 a month and a 50-person company $125. Against that sit installation, updates, security patching, monitoring, compatibility testing and support. Those costs do not shrink in proportion to headcount. Supporting a 50-person office does not cost a tenth of supporting a 500-person one.

Figure 5. Supplier economics per office at $2.50 a seat (all costs assumed)

Office size Monthly revenue Cost to serve Monthly margin
50 seats $125 $425 –$300
200 seats $500 $500 $0
500 seats $1,250 $650 +$600
2,000 seats $5,000 $1,400 +$3,600

Assumes $400 a month fixed per office (updates, monitoring, security, support) plus $0.50 per seat. Illustrative, not measured.

Three design choices follow. A tenth-price seat needs a minimum commitment, either a seat floor or a fixed office fee, so that small offices cover the cost of serving them. It needs standardised installation and predominantly self-service operation, because a support visit can wipe out a year of margin. And it probably needs local partners who install and support it, paid from the hardware and services they sell rather than from the subscription.

The buyer’s comparison must be complete too. For the 500-person company, the full cost includes the subscription, the office server, IT time and approved premium use.

Figure 6. The buyer’s full annual cost, 500 employees (all inputs assumed)

Item Assumption Annual cost
Subscription $2.50 per employee per month $15,000
Office server $4,000, written off over three years $1,333
IT time Two hours a week at $15 an hour $1,560
Premium AI Administrator cap $3,000
Total   about $20,900

Against $120,000 for developed-market seats, that is about a sixth, not a tenth, once everything is counted. Against AI bundled into an existing suite, the money may be similar, which is why price cannot be the whole case. The case is coverage across a mixed environment, memory the company keeps and control over where data goes. The $2 to $3 seat is a design constraint to be proved in pilots, not a price the architecture has already earned. The test is cost per completed task, including hardware, IT effort and human rework, not the licence fee alone.

A flat price also cannot mean infinite work for infinite users. It means a predictable price per employee, with fair-use and concurrency limits set by the office’s own capacity: when the office is busy, answers queue or slow down; they do not generate a bill. Premium external inference is different. It should be approved by the administrator, priced separately and visible before it is spent. A flat price is a promise about the bill, not about unlimited compute.

6   Sovereignty is an operating promise

For a bank, a manufacturer, an insurer or a services firm, the attraction is not only cost. It is control. Questions, documents and answers can be processed on machines the company owns, under permissions the company sets.

The promise needs precise wording. “No data leaves the building” is too narrow: companies have remote employees, several offices, existing cloud software and, sometimes, a good reason to use a premium external model. The real promise is this: no enterprise data leaves the approved processing environment without explicit policy authorisation. That environment may span offices and a private cloud the company controls. Telemetry, software updates, connectors to other systems and premium escalation can all carry data out unless they are configured not to, which is why the promise is a policy to enforce and audit, not a slogan inherited from where the server sits. Sensitive classes of work, such as legal, HR and finance, should run only on machines IT controls, by default rather than by request.

The record-keeping needs the same precision. Storing every raw prompt indiscriminately would defeat the purpose. But a ledger that records nothing but usage is too thin for work that needs evidence. Audit records may need source references, decisions, approvals, model versions and access logs, kept under retention rules the company sets. The objective is that the enterprise keeps control of sensitive information and can prove where it went, who saw it and what was done with it.

Regulation is a reason for some buyers, not all. In India, the data-protection law does not impose blanket localisation: its cross-border framework allows transfers except to countries the government restricts, while sectoral regulators such as the RBI can impose stricter rules. Local processing can simplify governance and satisfy contractual or sectoral requirements. For most firms, it is a choice about control rather than a statutory necessity, and the essay should not claim otherwise.

Open source helps, but it is not a guarantee. Code that can be inspected makes a backdoor findable, not impossible. The stronger promise is verifiability: published code, signed releases, builds that independent parties can reproduce, documented components and external audits, practices of the kind NIST’s Secure Software Development Framework sets out. Trust comes from what can be checked, not from what is promised.

7   The layer that lasts

Underneath the four jobs sits the asset that matters most: the shared operating layer. It records what each application knows, what each employee is authorised to see and do, where each task should run and whether completed work met the agreed standard. Applications are replaceable. A company might change its meeting writer, its chat assistant or its local model next year. It should not have to replace its memory, its policies or its operating history when it does.

This is the argument of From Rented to Owned, carried into the office: own what compounds, rent the rest. Here, what compounds is the record of the company’s work and the rules around it. The models underneath can be swapped as better ones appear, and the routing adjusts to use them.

It also clarifies what kind of business this is. A suite of cheaper substitutes competes on price. A trusted, portable operating layer that helps several applications work together competes on continuity, convenience and control. The first can be undercut by any incumbent willing to cut a price. The second becomes more valuable the longer a company uses it.

8   The business model

The opportunity grows in three levels. At the first, a company saves on AI bills by running routine work on hardware it owns. At the second, it gains private AI infrastructure: local routing, permissions, memory and controlled escalation, valued for control as much as cost. At the third, several applications share one operating layer for office intelligence. The lasting advantage sits at the third level, and it comes from making many applications cheap and dependable to run, not from idle processors alone. It would be a mistake to claim all three before proving the first.

Figure 7. How the business earns, and what it refuses

Revenue line How it works Why it fits
Base subscription A flat price per employee, about a tenth of developed-market seats, with a seat minimum or office fee Predictable for the buyer; covers everyone; covers the cost to serve
Premium escalation A disclosed commission on approved external inference, capped by the administrator Pays for hard cases without hidden metering
Hardware and partners The always-on office server, sold or leased, installed and supported by local partners One machine turns idle capacity into a dependable service
Advertising Not offered A product sold on keeping data inside cannot be paid for by anyone who wants to know what is in it

Two cautions belong in the model itself. If the escalation commission becomes the main source of margin, the incentive tilts towards sending work outside; disclosure and caps exist to prevent exactly that. And the natural first market is emerging-market enterprises, where the price gap is widest, software environments are most mixed, local languages matter and data rules are tightening. Those buyers do not need a lesser version of AI. They need a different economic architecture for delivering it.

9   How the thesis could fail

Three bets should be tested separately. Product fit: does one integrated suite complete everyday work better than separate tools or the AI bundled into existing software? Economics: does a seat at around a tenth of developed-market prices, with a minimum commitment, cover models, support, updates and margin? Trust: will security and compliance teams approve a shared local service, and will employees use it without feeling their machines or their data have been commandeered?

Figure 8. Eight ways the thesis fails

Failure What would show it
Bundled AI is good enough Companies on one vendor’s suite see no gain worth switching for
Open models fall short on real office work The share of tasks escalated to premium models stays high, or human rework rises
The shared record adds little Users rarely move work between the four jobs; the suite behaves like four separate tools
The idle pool adds nothing One office server matches the pool at equal cost; then sell the server and drop the pool
The price cannot fund the service Cost to serve exceeds revenue even above the seat minimum
Buyers trust global brands more than local control Pilots convert slowly even where the product performs
Escalation becomes the real business Commission revenue outgrows subscriptions
Global vendors cut prices for emerging markets Developed-market tools arrive at local prices, removing most of the gap

A ninety-day pilot in one office of 20 to 50 consenting users would answer much of this. Run the four jobs on real work against current practice, including whatever bundled or free AI tools employees already use. Serve them three ways: one dedicated office server, the server plus an idle-machine pool, and a hosted small model. Measure completion, errors, response time, availability at peak hours, human rework, energy, the share escalated outside, the share of work that crosses between jobs, the supplier’s cost to serve and cost per completed task. Set the stop rules before starting.

If the suite needs constant human rescue, the shared record adds nothing, or incumbents close the gap on price and control, this particular route is wrong. The need it addresses does not go away.

**

The first wave of AI sold extraordinary intelligence to those who could afford the invoice. The next should make dependable intelligence ordinary, private and affordable across whole organisations, in the countries where most of the world’s office workers live. The answer is not merely to make intelligence cheaper. It is to make it a shared utility inside the enterprise: available to every employee, across applications, with the company keeping control of its data, its memory and its costs.

Affordable for one person is a subscription. Affordable for everyone is infrastructure.

Thinks 2145

Cody Finke: “My proposed explanation is that in order to impact growth, the innovation must impact a market large enough to matter or a set of markets large enough to matter. While it is statistically possible for hundreds or even millions of innovations to add up to high growth (and for indoor plumbing to lead to millions of innovations), it is much more likely that a smaller number of innovations impacting very large markets occurred. I argue that 7 times in the special century, innovations dramatically cut cost for products that, by themselves, were more than 1% of economic growth. This has never happened since 1970. One way to drive economic growth is to reduce the cost of things the economy needs. When something gets cheaper, the economy can produce the same amount of it and still have resources left over to make other things. Spending the excess money from cost reductions pulls freed resources into new uses, and, via Jevons paradox, we also, often, end up buying more of the thing that got cheaper.“

Henry Mintzberg: “As a manager, get out of your office, get on the ground, and find out what’s going on.”

NYTimes: “A superhabitable world — as scientists imagine it — is about 20 to 30 percent wider and twice as massive as Earth, and is placed at the right distance from its star for water to flow on its surface. The stronger gravity of this bigger Earth would hold on tighter to ambient molecules than Earth does, giving the planet a thicker atmosphere and a warmer and more stable climate, fueling the growth of extraterrestrial plants and animals (or their equivalents). The super-Earth’s bigger iron core would erect a stronger magnetic field around it, better blocking cosmic radiation. As for the superhabitable planet’s surface, picture paradise. An enlarged orb would be rounder and smoother than Earth, as its stronger gravity erodes continental shelves and fills in trenches. Water would spill evenly over its surface as shallow lakes and seas.”

Adam Grant: “There are three important roles that people play in our lives: allies, pals, and companions. Allies share our goals; they’re our most reliable and capable collaborators. Pals share our interests and hobbies; they’re our go-to people for a pickleball game or a karaoke night. Companions share our values; they’re our best sources of support and compassion.”

When Search Keeps the Answer

Published November 15, 2026

Publishers and brands are losing the same click. ActionAds and NeoNet could turn their two problems into one market.

1 The Click That Does Not Come

On 29 October 2026, Alphabet had the worst day in its history as a public company. Its shares fell about 13%, and roughly $500 billion of market value disappeared during the session. Most of the selling was about spending, not search: capital spending had more than doubled and free cash flow had turned negative.

One line on the earnings call deserves longer attention than the share price. Google’s chief business officer attributed the slowdown in search revenue mainly to currency and a handful of industries, and then added that fast-growing AI Mode queries monetise less well than classic search. The day before, Google had announced ads in AI Mode in ten more countries, including India, and ads inside follow-up questions in the US. Google is not standing still. It is moving its toll booth into the conversation.

What happens to everyone else is becoming clearer. When Google shows an AI summary, users click a traditional result in 8% of visits, against 15% when there is none. A randomised experiment by Saharsh Agarwal of the Indian School of Business and Ananya Sen of Carnegie Mellon, which hid AI Overviews from some users, found that showing them cut outbound organic clicks by about 40%, with no detectable effect on sponsored clicks. That experiment tested AI Overviews inside classic Search; AI Mode, the conversational experience, is newer and less studied, though early evidence points the same way.

Figure 1. Two studies of different design, one direction: fewer organic clicks leave where the AI answers

The finding on sponsored clicks narrows the argument, and it should. The evidence does not show search becoming too expensive for advertisers. It shows search becoming less dependable as a route outwards. For twenty-five years the bargain was simple: you search, Google points, you leave. Increasingly, Google answers and you stay.

Search brought customers to businesses. AI Search increasingly brings answers to customers.

2 Two Businesses, One Dependence

Two groups feel this most, and they are rarely discussed together.

Publishers first. Data from Chartbeat for the Reuters Institute showed Google organic search traffic to more than 2,500 news sites falling by a third globally in a year, and small publishers losing 60% of their search referrals over two years. The Reuters Institute is careful to say that AI summaries are not the only cause; changes in search and in audience behaviour matter too. But the direction is not in doubt, and advertising follows pageviews down. In How Publishers Can 10X Their Revenue I described what this means: the visit that paid for journalism is being replaced by an answer that pays for nothing.

Brands next. A lapsed customer has several ways back to a brand: its app, its website, its emails, search, and increasingly an AI assistant. When she has stopped responding to the brand’s own channels, search is one of the most important. She types the brand’s name or the category, clicks and buys. Some of those clicks were free organic results; others the brand paid for. The free part of that route is the part now narrowing. The paid part still works, but it is rented, and the brand pays again for a customer it already had. The claim here is weaker than for publishers: fewer outbound clicks need not mean fewer purchases, and the evidence of harm to brands is still emerging. In The AI Revolution’s Hidden Sponsor I argued that this reacquisition toll helped finance the AI build-out. It is now reshaping where attention lands.

Put the two side by side and the symmetry is plain.

Publishers depended on Search for their readers. Brands depended on it for part of the way back to their customers. Both now find that journey controlled by an intermediary that increasingly keeps the answer.

Neither can fix this by doing more of the same. Both need an additional route back that they control, and each, as it turns out, holds something the other needs.

3 The Publisher’s Answer

The publisher’s first move is to change the unit of its economics, from the pageview to the relationship. The audience that survives AI search is the one that comes back on its own, and the largest such audience most publishers have is their email list. Most run it as a pipe, built to send readers back to a website that search no longer feeds.

Run as a product, the list becomes something else. A NeoLetter is composed when it is opened rather than when it is sent, so the reader sees what is current, and it lets her act without leaving the inbox: answer a poll, save a story, request a quote. That builds the habit first. Monetisation comes after, and in a different form.

To raise yield, publishers must sell actions, not impressions.

An ActionAd is a sponsored action inside earned attention: a test-drive booking in a motoring newsletter, a fund factsheet requested from a markets digest, a re-permission for a partner brand. Because AMP for Email does not allow third-party ad components, these are modules the sender controls, with plain HTML fallbacks for inboxes that cannot run them. The first advertiser should be the publisher itself – its events, subscriptions and verticals – before any third party pays.

Set out against search, the difference is in what is sold and who keeps the relationship.

Classic Search AI Mode Inbox ActionAds
Signal of interest A typed query A conversation A publication the reader chose to open
Where attention sits Google’s results page Google’s answer The publisher’s inbox relationship
What is sold A click A sponsored recommendation, and increasingly a completed action A completed action
Who holds the relationship Google Google The publisher, and through it the brand
Pricing Auction per click Auction, folded into Search campaigns Per incremental action, tested against holdouts

Search monetises expressed intent. ActionAds monetise earned attention.

Google is not standing still here either. On its earnings call, Sundar Pichai described agents that book reservations, fill in forms, schedule appointments and make purchases, and Google already offers direct checkout for hundreds of thousands of brands. So the difference is not that Google sells clicks while email sells actions. It is between actions coordinated by an intermediary that controls discovery, and actions initiated by a brand that already knows its customer, carried through attention she has chosen to give. Google’s actions run through Google’s relationship with the shopper. ActionAds run through the publisher’s relationship with the reader, and CRM Ads through the brand’s own.

The question is not who completes the action. It is who holds the relationship after it.

4 The Brand’s Answer

The brand’s problem is different. It does not lack customers; it lacks their attention. Its database holds millions of people who bought once and now ignore its emails. The question is how to reach them before paying to rent them back.

In CRM Ads: Your Customer, Someone Else’s Attention I proposed an order: owned attention first, relationship attention second, rented attention last. The middle rung is the new one. Relationship Media is a brand reaching its own known customer through attention another party has earned.

Consider one customer. She bought from a fashion retailer last year and earned loyalty points. She has not opened the retailer’s emails in months. She does open a travel publisher’s Digest every morning, and when she subscribed she granted a “Your Brands” permission: brands she already buys from may send her useful reminders there, for a stated purpose, and she can switch any of them off. This month her points are about to expire. NeoNet matches the retailer’s unfinished action to the Digest at the moment she opens it, and a clearly labelled panel appears: a brand you shop with has loyalty points expiring this month. She taps, and her balance and redemption options open on the retailer’s own secure page, not in the publisher’s email. The Digest carries the invitation; the brand’s surface carries the detail. The relationship is restored, the panel cancels itself, no retargeting budget is spent, and the publisher earns a fee for carrying a useful action.

Figure 2. One customer, three rungs: the relationship rung does the work before the rented one is needed

That is a CRM Ad. It is not prospecting: it reaches only customers a brand already has. It rests on rules that are not negotiable:

  • No reach across brands without permission the customer can see, understand and revoke.
  • No payments, passwords or identity checks on a third party’s surface, and no personal details such as balances until the customer chooses to look, on the brand’s own surface.
  • No banks as hosts, and no prospecting or lookalikes in the first version.

Now the two halves meet. Publishers rebuilding direct audiences create earned attention that needs monetising. Brands with customers who no longer respond hold relevant actions that need a route. NeoNet sits between them, matching deterministically where the same permissioned identity exists on both sides, and deciding at the moment of open rather than days earlier when a campaign was set up.

Figure 3. Two casualties of AI search, one market – and a loop that can compound

One group’s problem is the other’s supply.

The market can compound. More recurring, permissioned attention creates more places to route useful actions. Better routing raises publisher yield and lowers the brand’s cost of recovery. Higher yield funds better publications, which earn more attention. And supply is not limited to publishers. A retailer, airline or jobs platform that builds a successful Brand Digest earns attention too, and can carry other brands’ actions while sending its own; some members will sit on both sides of the market. The loop runs in reverse just as easily. Irrelevant actions, poor matching or over-monetisation would spend the attention the whole network depends on.

A brand can measure its side with one number, Alternative Reach: the share of its attention-dark customers who are reachable, with permission, through another member of the network. NeoNet connects unfinished customer intent to attention that already exists.

5 The Market That Has to Be Earned

None of the parts is new on its own. Google’s Customer Match already lets a brand upload its customer list and reach matched customers across Search, Gmail and YouTube, including dormant ones. Google Ad Manager serves ads in participating publishers’ newsletters, largely as image creatives, because inboxes cannot run the scripts display ads rely on. LiveIntent built an advertising network across thousands of newsletters before Zeta Global bought it in 2024. The claim here is narrower: actions completed inside the email, routed through permissioned earned attention, and priced on incremental outcomes measured against holdouts. That combination is harder to copy than any single piece. It has to be demonstrated, not presumed.

Google could build it, partner with it or compete with it. It has the advertisers, the ad server and newsletter distribution through participating publishers. What it does not hold automatically is the reader’s permission for one brand to reach her through another’s publication, or the authority to put actions inside other senders’ emails; owning Gmail confers no right to commercialise those messages. This is less a new technology than a new market design, and its economics will be decided by the publishers and brands who choose to join.

Every participant will ask why it should switch. A brand will ask why it should use NeoNet instead of Customer Match; a publisher, why ActionAds instead of its existing newsletter advertising. Each needs its own scoreboard.

Participant What must be demonstrated
Publisher Higher net revenue per thousand engaged readers, after costs, with no loss of engagement, subscriptions or retention
Brand Lower cost per incrementally recovered customer, or more incremental gross profit, than paid and owned alternatives
Network Permissioned Alternative Reach, match rates, completed actions and positive contribution economics

The decisive test is simple to describe. Take eligible dormant customers and split them at random into three groups: a holdout that receives only the brand’s usual activity, a group reached through conventional paid reacquisition such as Customer Match, and a group reached through CRM Ads. Measure incremental recovery, gross profit and fully loaded cost. Separately, measure whether ActionAds dent the publisher’s future engagement.

The test is not whether NeoNet produces actions. It is whether it recovers customers more profitably without consuming the attention that makes the network valuable.

Order matters more than ambition. Earn the attention first. Monetise it with ordinary ActionAds. Only then earn the permissions and identity coverage that CRM Ads need. Prove the economics before expanding the permission; expand the permission before expanding the specificity.

What would prove this wrong

  1. Readers decline the “Your Brands” permission, or find it unsettling.
  2. Publishers cannot build enough recurring attention to supply a market.
  3. Publisher net revenue per thousand engaged readers is no better than existing newsletter advertising, or engagement falls.
  4. In the three-way test, CRM Ads recover customers no more profitably than paid reacquisition or the holdout.
  5. The permission design cannot satisfy India’s data protection rules.
  6. A better-funded player assembles the pieces first, which would confirm the idea and lose the race.

Search will keep sending customers for many purchases, and paid search will remain a powerful route. But the businesses that depended on it for the way back now need a second road. The first generation of internet advertising monetised pages. The second monetised searches and feeds. The third may monetise useful actions inside relationships that already exist.

The future of advertising may not be about finding a better place to display an ad. It may be about finding the most trusted route to complete the customer’s next action.

Thinks 2144

Dwarkesh Patel: “I think we’re a couple years away from systems that match humans across all cognitive domains and which vastly outnumber humanity and are thinking much faster. I think this will have a big impact on the world. It’s very hard to think about what kind of impact it will have. Some of it is just inherently unpredictable, but trying to understand what is happening and articulate it seems like an important thing to do right now.”

WSJ: “Aldaron is part of a new breed of companies centering their entire business models around applying new technologies…Aldaron’s investments are aimed at increasing head count and improving operations at portfolio companies by accelerating the adoption of AI. Aldaron avoids businesses whose strategy involves laying people off, Mendez said.”

NYTimes: “There are roughly 700 companies pursuing innovations in the overall care economy, according to Anna Steffeney, executive director of FamTech.org, an association of care companies in the technology sector. Within that niche is an emerging market for parent-focused companies, which are often founded by exhausted parents themselves, Steffeney said. Apps including Ollie, Fambot, Ohai and Ava aim to function as household managers, and the majority of users so far are mothers, according to the chief executives of these start-ups. Other parents are creating their own D.I.Y. systems by leveraging general-use artificial intelligence systems — like ChatGPT, Claude or, more recently, Muse — that are more capable but also require more skill to prompt successfully. The agents typically demand access to an individual’s emails, calendars and other communication platforms.”

Kevin Kelly on Swarm Intelligence: “Here is a possible world 50 years in the future, one I think is more likely than the scenario most people expect, which is one huge AI in the sky. Fifty years from now most AI is invisible. There is nothing to see. It is sort of like air – always there, always on, everywhere. Something you need and use but take for granted. AI is so prevalent we become blind to it. You don’t need to log onto it, or even to summon it. Rather you inhale it. Instead of a central AI brain, AI is like a gas, composed of zillions of tiny interacting AIs that create an atmosphere. These zillions of minds are like eukaryotic cells in your body, or hundreds of varieties of bacteria in your gut. They form a presence. They are woven together into a vast network of intelligences that create an AI background. This society of AI minds is not so much seen by humans, as it is felt. Unlike the AI of today, it will be deeply embodied – into robots, vehicles, buildings, and devices. But its embodiment will be so wide, so deep, so pervasive, that it will become the background.”

FT: “The wisdom of prediction markets…Trades can contain valuable cues for businesspeople and investors.”

The AI Revolution’s Hidden Sponsor

Published November 14, 2026

Who paid for the AI boom, and what happens to the businesses paying the toll

1 The Profits Behind the Revolution

Every technological revolution has a financier. The railways were built on the bond markets. The telephone network was built on regulated monopoly rents. The early internet was built on venture capital and an equity bubble. The AI revolution has many financiers – venture funds, sovereign wealth, chipmakers, enterprise software profits – but one of the largest hides in plain sight. Among the companies building the most AI infrastructure, two of the most powerful cash engines were built on advertising. Alphabet and Meta turned the sale of attention and intent into extraordinary profits, and those profits are paying for much of the foundation on which the next technological era is being built.

Add Amazon, the third giant of digital advertising, and the scale becomes clear. In 2023 the three companies earned about $417 billion from advertising between them. By 2025 that had risen to about $560 billion, up 34% in two years. Over the same two years, their capital expenditure – the money spent on data centres, chips, networking and power – rose from $112 billion to $293 billion, up 162%. In 2023, capital spending absorbed 43% of the cash their operations generated. In 2024 it absorbed 52%. In 2025 it absorbed 70%. The figures come from the 2025 annual filings of Alphabet, Meta and Amazon.

The cash engines grew strongly. The capital demands of the next era grew far faster.

The 2026 plans extend the curve sharply. Alphabet has guided to capital expenditure of $195–205 billion, Meta to $135–145 billion, and Amazon to about $220 billion. Together that is roughly $560 billion: almost double 2025, five times 2023, and a third more than all the operating cash the three companies generated last year. Not every dollar is for AI; some serves warehouses, logistics and ordinary computing. By the companies’ own accounts, most of it supports AI and cloud infrastructure, and the direction is unambiguous: Alphabet and Amazon have both said they will spend meaningfully more again in 2027.

Figure 1. Alphabet, Meta and Amazon combined, 2023–2026. Capex is gross cash purchases of property and equipment; 2026 is guidance and includes non-AI spending.

Precision matters about who is paying, because the three companies are not alike.

Meta is the purest case. Advertising is about 98% of its revenue, and its Family of Apps – Facebook, Instagram, WhatsApp and Messenger – earned $102.5 billion of operating income in 2025. Everything else Meta does, from its AI labs to its data centres, is funded by that one engine.

Alphabet is the central case. In 2025, Google Services earned $139.4 billion of operating income against $13.9 billion for Google Cloud. Services includes subscriptions, devices and the Play store, but its core is advertising across Search, YouTube and the partner network. Cloud is growing fast and is profitable, but last year Services earned about ten dollars for every dollar Cloud earned.

Amazon is the different case. AWS earned $45.6 billion of Amazon’s $80 billion operating income in 2025, about 57%. Advertising brought in $68.6 billion of revenue, at margins widely believed to rival AWS, but Amazon does not disclose its advertising profit. Amazon illustrates diversification better than it illustrates advertising dependence.

Figure 2. Where the profits came from: 2025 operating income by segment

One concession belongs at the start. Microsoft, the second-largest cloud provider and one of the largest AI spenders of all, is not an advertising business. Its build-out rests on enterprise software and cloud. So the claim is not that advertising funds all of AI. The claim is that advertising funds the ad-funded AI builders: Alphabet and Meta almost entirely, Amazon in part. Remove Amazon and the pattern sharpens rather than weakens.

The clean case: Alphabet and Meta alone. Capital expenditure absorbed 34% of their combined operating cash flow in 2023, 41% in 2024 and 57% in 2025. Their 2026 guidance of about $340 billion is roughly 120% of the $281 billion of operating cash they generated in 2025.

Two cautions keep the argument straight. Revenue is not profit, and profit is not cash. Advertising revenue measures the toll collected; operating cash flow measures what is available to spend. Measured either way, the conclusion holds: much of the money building the AI economy has come from businesses paying to reach customers. And when a build-out outgrows the engine that funds it, the funding question leaves the income statement and enters the capital markets.

2 From Surplus to Capital Markets

For two decades, the advertising giants financed their futures out of surplus. Search and social generated more cash than even their ambitions could absorb. For most of the past decade the excess went back to shareholders through buybacks, joined in 2024 by dividends. The model was simple: earn from advertising, invest what the business needs, return the rest.

In 2026 that model reached its limit. In the third quarter, Meta’s capital spending of $34.9 billion exceeded the $32.8 billion its operations generated, and its quarterly free cash flow turned negative for the first time since its 2012 listing (results). Meta repurchased no stock in the first nine months of the year, against $26.2 billion a year earlier. Alphabet’s free cash flow turned negative in the second and third quarters, and its trailing twelve-month figure fell 73% to $19.7 billion. Amazon’s trailing twelve-month free cash flow went negative in the middle of the year and had reached an outflow of $42.4 billion by the end of September.

None of these companies has run out of money. Meta’s free cash flow for the first nine months was still positive, at about $11 billion, and all three hold large cash balances. The change is at the margin, and it is structural rather than quarterly: investment has begun to outrun internally generated surplus, and the companies are supplementing operating cash with debt, equity and long-term commitments.

The equity raises need careful reading. In June 2026, Alphabet announced an equity raise with a headline value of $84.75 billion, and its parts served different purposes. About $35 billion of underwritten common and mandatory convertible preferred stock, together with a $10 billion private placement with Berkshire Hathaway, was raised for general corporate purposes, including AI infrastructure. A further $40 billion at-the-market programme, drawn over time, was designed primarily to change how Alphabet meets tax obligations on employee equity awards, and should not be counted as AI funding. In October, Meta priced $37.5 billion of stock: $25 billion of common shares through forward sales, whose proceeds arrive on settlement by November 2027, and $12.5 billion of mandatory convertible preferred. A further programme of up to $60 billion from 2027 is contemplated, not committed.

The cash in hand is therefore smaller than the headlines. But it answers the question that matters: does the market value of these companies now help pay for their AI? At the margin, it does.

For twenty years, advertising paid for the future out of surplus. In 2026, it began to lean on the share price as well.

The mechanism is indirect rather than mechanical. A market capitalisation does not pour concrete or buy chips. But once internal cash runs short, the share price sets how much of the company must be sold to raise each dollar, and the credit markets price their confidence in future cash flows. That confidence still rests on the advertising franchise. On 29 October 2026, the day after Alphabet reported slower Search growth alongside a doubling of Cloud revenue, it lost close to $500 billion of market value, the worst day in its history. Investors value Alphabet as an advertising business building a cloud business, not the reverse.

This creates a loop worth naming. Advertising profits support market value. Market value makes capital available at scale and at a tolerable cost. Capital funds AI infrastructure. AI improves ad targeting and creates new businesses. Better ads support advertising profits.

Figure 3. The reflexive funding loop behind the ad-funded AI builders

George Soros called loops like this reflexive: perception and reality reinforce each other. They compound on the way up. They are tested on the way down, because a weaker advertising engine would lower market value, raise the cost of capital and slow the very investment meant to replace it.

The loop ends only when AI can pay for itself, and that has three stages. Revenue validation asks whether customers pay meaningful sums for AI services; they do. Operating profitability asks whether those services cover their running costs; Google Cloud’s segment profit says yes for one of them, though segment figures leave out shared AI research costs carried at the corporate level. Capital independence asks whether AI generates enough cash to fund its own maintenance, chip replacement and growth without leaning on the old franchise. Nobody has reached that stage. Accelerators wear out within a few years, depreciation compounds, and an AI business can show rising revenue and operating profit while still needing outside capital to grow. Amazon makes the point: AWS earned $17.8 billion of operating income in the third quarter of 2026 alone, while the company as a whole ran a trailing free cash outflow of $42.4 billion. Until capital independence arrives, the advertising engine remains the guarantor.

If advertising is the sponsor, the next question follows. Who pays the advertising bill?

3 The Bill Behind the Profits

Every advertising dollar is a cost line in some other company’s profit and loss account. The sponsor of the AI revolution is, in the end, the world’s brands: retailers, banks, airlines, consumer goods makers, direct-to-consumer start-ups. Most of what they buy is what the invoice says – introductions to strangers, and demand that would not otherwise exist. But part of what they buy is stranger. It is access to people they already know.

Consider how it happens. A customer buys from a retailer. The retailer holds her email address, her mobile number and her purchase history. Over the following months its emails go unopened, its offers blur into the inbox, and its systems eventually mark her as unresponsive. Six months later she searches for the category, clicks a paid listing and buys again. The advertising dashboard records a conversion. The profit and loss account tells a different story: the brand paid a platform to restore a relationship it already possessed.

The pattern runs across the stack: search ads bought on the brand’s own name, retargeting of past visitors, and “new customer” campaigns whose audiences quietly overlap the customer file. The cause is the same each time. Attention in owned channels decays, and the only scalable way back is paid reach.

Not all of this is waste, and the distinction matters. Spend on known customers falls into three tiers. The first is the pool worth investigating: spend that reaches people the brand already knows. Call it potential AdWaste exposure – measurable, probably substantial, and not wasteful in itself. The second is spend that is not incremental, because the customer would have bought anyway. The third is avoidable reacquisition: spend that worked, but that an owned channel kept warm could have matched far more cheaply. Only the second and third are economic waste, and only a controlled test can verify them: a holdout group that receives no contact, compared with the paid route and the owned route.

Figure 4. Exposure is not waste: three tiers of spend on known customers

The platforms have a fair rebuttal. Knowing a customer’s email address is not the same as having an effective, free route to her. A lapsed customer who ignores email may respond to a search ad, and paying for that response can be rational.

The evidence supports the concern without settling its size. A field experiment at eBay, by Thomas Blake, Chris Nosko and Steven Tadelis, found that paid search on eBay’s own brand name produced no measurable short-term benefit, because shoppers simply arrived through the free link below the ad. Returns from other paid search came mainly from infrequent users; frequent users who already knew eBay were barely influenced. On the other side, an analysis of 640 Meta incrementality experiments by the measurement firm Haus found that Meta lifted brands’ primary results by 19% on average, with 65–70% of that impact coming from new customers. That is real acquisition. It also means roughly a third of Meta’s incremental impact came from customers the brands already had – incremental, and therefore not waste, but a measure of how much paid media is spent on people already known.

Nobody knows the size of the wasteful slice, and the reason is structural. Each system measures the job it was built to do. Adtech reports conversions. Martech reports engagement. The agency reports media efficiency. Nobody owns the question that sits between them: was this customer new to the company? The number that should exist is what I call REACQ% – the share of customers that paid channels report as new who were already in the brand’s database before the paid touch. It is a count of people, not a split of spend. It is a warning light, not a verdict: it shows where to look, and the holdout test shows how much of the spend was wasted.

My own estimate, developed across the NeoMarketing essays, is that AdWaste runs to about $500 billion a year globally. That is an estimate of the scale at stake, not a measurement; the real figure will emerge brand by brand, as REACQ% gets counted and tested. Even a modest fraction of it is a material share of the revenue the AI builders collect.

Part of the money building frontier AI is a tax on forgetting: brands paying to rent back customers they already own.

That tax does not sit idle. It is now funding the systems that will reshape how customers are found.

4 The Reinvention Paradox

The paradox of the AI build-out is that it strengthens the advertising engine before it threatens it.

The strengthening comes first. AI improves targeting, creative, bidding and recommendations. Meta’s case for its capital spending rests largely on advertising performance: better models put more relevant ads before more engaged users, and advertisers pay more for them. For brands, the toll becomes more productive, which also makes it harder to escape.

The threat follows. AI answers reduce the clicks that search advertising was built to sell. Agents acting for consumers do not see banners. Discovery shifts from results pages and feeds to conversations and delegated decisions. A question answered is a click not sold, unless the answer itself carries a sponsor.

Three futures follow from this tension.

Future What happens
1. Advertising finances the transition AI lifts the old engine while cloud and AI services scale. The incumbents cross from one profit pool to the next on their own cash and credit.
2. The old engine weakens first Discovery moves into AI answers and agents faster than it can be monetised. Earnings disappoint, capital gets dearer, and the build-out slows.
3. The toll is reinvented Agent gateways charge brands for recommendation, visibility and transactions. The old toll fades and a new one replaces it, often collected by the same companies.

The third future is the most consequential, and on today’s trajectory the most likely – though not inevitable. Gatekeeper rents rarely disappear; they migrate. Classified advertising moved from newspapers to search. Attention moved from portals to social feeds. Each time, whoever controlled the moment of choice found a way to charge for access to it. If agents become the moment of choice, the companies best placed to run them are those that already own the models, the distribution, the identity graphs and the advertiser relationships.

Much depends on whom the agents serve. A platform-controlled agent can monetise recommendations, placement and transactions. A customer-aligned agent, paid for by subscription, could refuse sponsored recommendations and optimise for its user alone. Will agents become the next advertising gatekeepers, or the first effective guardians against them? The answer is open, and few commercial questions this decade matter more.

Either way, two loops are already visible. The first is the incumbent loop: advertising profits fund AI capital, AI produces better products and new businesses, and those strengthen the profit engines. The second is the toll-payer’s loop: brands lose direct attention, re-buy customers through paid advertising, that revenue helps fund AI, and AI builds more powerful gateways to customers – deepening the dependence that started the cycle. Its magnitude remains a hypothesis. Its direction is the point.

Figure 5. Two loops: the incumbent’s and the toll-payer’s

Brands may be helping finance the intelligence that will decide which brands their customers discover, consider and buy from tomorrow.

There is a deeper contest inside the paradox. AI is being put to work in two economies with opposite incentives. In one, platforms use AI to rent attention: sharper targeting, better recommendations, higher ad yield. The objective is to maximise the value of paid access. In the other, brands use AI to earn attention: relevant communication between purchases, better service, retention and recovery through direct relationships. The objective is to minimise the need to buy access again. Both draw on the same improving capabilities. One makes the advertising auction more effective; the other makes some of those auctions unnecessary.

The next great contest in marketing may be between AI that makes advertising more valuable and AI that makes advertising less necessary.

Better AI targeting is therefore not automatically a victory for brands. It can raise short-term returns while leaving the underlying dependence on paid access intact. Which economy a given brand ends up in depends on what it does next.

5 Lessons for the Gatekeepers and the Toll-Payers

Step back from advertising and the story carries lessons for any business with a profitable core and an expensive future.

For every business with a cash engine

Every moat has a payer. Behind every cash cow is a customer paying more than they would in a more contestable market, or paying for something they could have had more cheaply. The first strategic question about any franchise is who funds it, and whether they can stop.

Cross-subsidies are fragile. For decades, newspaper classifieds paid for newsrooms. Readers believed they were paying for journalism; the economics said otherwise. When classifieds moved online, journalism did not become less valuable, but it lost its sponsor, and newsrooms shrank. Any business whose future is funded by a single profit pool should know exactly how that pool could dry up.

Beta can fund Alpha, but Alpha must earn the right to finance itself. Protecting the base business creates the cash and stability to build the next one. Until the new business reaches capital independence, it is an expensive experiment funded by yesterday’s advantage, and its funding is only as secure as that advantage.

Surplus and dilution are different kinds of freedom. A company that funds its future from earned profit controls its own pace; this is the idea behind what I have called the Proficorn. A company that funds its future through its share price depends on a story staying intact. The advertising giants have moved, at the margin, from the first condition towards the second.

For the businesses paying the toll

Never pay twice. The first move is measurement. Count how many of the customers your paid channels call new were already in your database, then use holdout tests to find out how many of those purchases the ads caused. Until both numbers exist, every acquisition budget hides a reacquisition bill.

Owned attention is the asset the toll taxes. An email address is not an audience; it is permission to earn one. Every relationship kept warm in an owned channel is one the brand never has to buy back, and AI now makes daily, individual, relevant engagement affordable in a way it was not three years ago.

Agent-mediated discovery raises the stakes. If agents increasingly decide what customers see, a direct relationship may be the only access to a customer that a brand does not have to rent. A customer who hears from a brand directly is a customer whose choices an intermediary does not fully control.

This is the agenda I have been developing under the name NeoMarketing: measure what you pay to re-buy, earn attention daily rather than renting it back, and pay vendors for the outcomes they deliver. I have set out parts of it in The Profit You Already Own, Every Business Has a Strait and From Rented to Owned. It remains an argument to be proved in practice. But the question it asks has grown larger than marketing efficiency, because the toll brands pay is now financing the gateways they will have to pass through.

What would prove this wrong

  1. Reacquisition turns out to be a small share of platform revenue once REACQ% is counted and tested brand by brand.
  2. AI makes paid retargeting so effective that it becomes the cheapest route back to a lapsed customer, not an avoidable one.
  3. Cloud and AI businesses reach capital independence within two or three years, and advertising’s role in financing the build fades.
  4. Agents bypass owned channels entirely, which would weaken the case for owned relationships as the remedy.

Each of these is testable, and each will show up in the numbers within a few years. Until then, the pattern in the filings is plain.

The profits of one generation of gatekeepers are financing the next. The open question is who keeps the customer when they are done.

Thinks 2143

Aviv Shamny: “When you advertise to humans, you target eyeballs, emotions and dopamine, but when you advertise to agents, you advertise to logic, a very mathematical logic.”

Stijn Van Nieuwerburgh: “The rapid progress in artificial intelligence is often described as a software revolution. In economic terms, however, the current phase of AI development is better understood as a large-scale investment cycle in physical capital. Training and deploying modern AI systems requires vast quantities of specialized hardware, electricity, and purpose-built facilities. As a result, the expansion of AI is tightly linked to the construction of a new layer of industrial infrastructure…. The institutions, contracts, accounting practices, and investor base supporting this infrastructure are still being formed.” [via Arnold Kling]

Rana Foroohar: “One thing to remember from both the old and new Great Games is that outside nations can swing the battle. Afghans and Persians repeatedly foiled the plans of both Britain and Russia. And it was ultimately the rise of Germany that ended the Great Game, by shifting the entire geostrategic chessboard westward. Today, India, Japan, Canada, Turkey, the Gulf states, Nordic nations, Asean and the EU are hardly passive. Their choices, not just the aspirations of America and China, may decide our 21st-century power struggle.”

Ben Thompson: “Are you getting that messaging — chatbots now, natural interfaces later — is how we will communicate with AI? Are you getting that not only will we not program computers, we won’t use them — AI will? Are you getting that pre-built UI — write once, run everywhere, for everyone — is dead? These are not three separate predictions: this is reality, right now, in 2026.”

WSJ: “Hope is a muscle. Give it exercise. To be free and fearless, it helps to think, “What’s the worst that can happen?” Then assess the answer.”

Beyond the Base Fare: What IndiGo, Couriers and Messaging Platforms Can Teach Email

Published November 13, 2026

The carriers that escaped commodity pricing followed a recurring pattern: they kept the journey cheap and learned to price the job. Email has priced only the journey.

The argument in one line. Airlines, couriers, messaging platforms and verification services all sell a cheap, standardised journey and charge separately for the jobs customers value. Email sells only the journey, at about a paisa a send. A NeoMail can keep the 1-paise fare and add prices for what the email gets done: a premium for capability, a fee for each completed action, advertiser revenue for attention and a share of verified profit for recovery.

A note on prices. Every price in this essay is an indicator for controlled pilots, not a tariff. The base is a standard HTML email at about 1 paise, or ₹10 per thousand. Where the essay puts a rupee price on an action, it is an assumption chosen to make the arithmetic visible.

PART I

The Pattern

1   The 1-Paise Email

A send is the cheapest thing a brand buys, and it is priced like a pipe.

A brand pays its email provider about a paisa to deliver a message. For that it gets authentication, delivery, a version that renders everywhere, an unsubscribe link and a report. It is a good deal for the brand and a hard business for the provider. The price falls with every efficiency, and, as I argued in Martech Beyond Messages and MAUs, the meter shrinks whenever the provider helps the brand send less and better.

The email itself has changed far more than its price. A NeoMail can let a customer pay a premium, confirm a renewal, book a slot, record consent or answer a question without leaving the inbox. It can show a price that is current at the moment of opening. It can carry a Digest worth opening between purchases, and an action from a partner brand. 10+1 Email Innovations described these capabilities, and Paid to Send. Paid to Keep. argued that they deserve a different price.

This essay takes up the question those essays left open: how to price them. Not in principle, but as a rate card a brand can understand, a procurement team can audit and a provider can run. The answer comes from industries that met the same problem first.

The question is not whether email can charge more. It is what, exactly, the extra charge is for.

2   How Carriers Escape the Commodity

Four very different businesses solved the same problem in the same way: a cheap journey and a priced job.

A carrier moves something from one place to another. Once the network exists, carrying one more unit costs little, competitors can match the price, and the journey becomes a commodity. The carriers that escaped did not raise the price of the journey. They kept it low and found the jobs customers would pay for separately.

IndiGo. India’s largest airline built its business on a low base fare and a standardised operation: a fleet of 441 aircraft at the end of March 2026, built mainly around one aircraft family, with fast turnarounds and simple processes. The fare buys the seat. Seat choice, extra baggage, meals and priority boarding are bought separately. Ancillary income reached ₹2,446 crore in the third quarter of FY26, about a tenth of total income. In July 2026 the airline went further with a Lite fare for passengers travelling with cabin baggage only. The base fare stays low because the extras carry margin.

Couriers. Carrying a parcel is the commodity. Proof of who received it is not. FedEx sells a range of signature options on delivery, from no signature to an adult’s signature checked against identification, and several carry a surcharge. The shipper is not buying more transport. It is buying certainty that a particular job was done.

WhatsApp Business. Meta prices business messages per delivered message, by category: marketing, utility or authentication. In India, a marketing message has cost around seven times a utility or authentication message. When Meta lowered utility and authentication rates in July 2025, it said the aim was to bring them on par with alternative channels for those jobs. The pipe is the same for every message. The price follows the job, and the job is priced against what it replaces.

Verification services. Twilio Verify charges $0.05 per successful verification, plus the channel fee, which for a US text message is under a cent. The charge for the job is about six times the charge for the transport, and it is due only when the job succeeds.

Table 1.  Four carriers, one pattern. Email has done the first column well and the other two not at all.

Carrier The journey The job, priced separately Priced against
IndiGo A seat on the flight Seat choice, baggage, meals, priority boarding What passengers value enough to pay for
Couriers Carrying the parcel A signature and proof of delivery The cost of a disputed or lost delivery
WhatsApp Business Delivering the message The job’s category: marketing, utility, authentication Alternative channels for the same job
Verification services Sending the code A successful verification Building and running it yourself
Email today The send, at about 1p Nothing Other email providers
NeoMail The send, still at 1p Capability, completed actions, attention, verified lift The workflow each job replaces

The pattern has three parts. The journey is cheap, standardised and simply priced. The job is priced separately, often only when it succeeds. And the job’s price is set against what the customer would otherwise do: dispute a delivery, send an SMS, build a system, or be split from family on a flight.

The four models are not identical. IndiGo and the couriers sell optional extras; WhatsApp prices classes of job, some cheaper than others; Twilio prices success. What they share is a separation between the price of delivery and the price of what delivery accomplishes.

The carriers that escaped commodity pricing kept the journey cheap and learned to price the job. Email providers have mastered the first part. They have barely started on the other two.

PART II

The Email Fare

3   Five Products, Five Mechanisms

The send stays at 1 paise. Everything above it is priced by where its value can be seen.

Translate the pattern to email and the rate card has five products a brand can buy (Figure 1). The prices are indicators.

  • Email, the base fare: about 1p per send, including authentication, the HTML fallback, the preference centre and reporting.
  • NeoMail: about 1.5p per eligible send for interactive email built from standard modules, and about 2p for a Living Email composed at the moment of opening, with live price, stock and status.
  • NeoMail Actions: a fee per verified action, or a monthly minimum plus usage, for secure tasks completed inside the email: tell, consent, pay, book and verify.
  • Brand Digest: a price per Engaged Edition, meaning one reader, one edition and at least one verified action. No attention, no charge. Like the prices, the unit is a hypothesis for pilots.
  • Recovery: no upfront cost. The provider funds the programme and earns Carry only on incremental revenue verified against a concurrent holdout. Everything the brand spends, fees and incentives included, carries a floor of six times its value in incremental revenue, backed by a capped make-good; for low-margin categories the floor is set on contribution instead. Martech Beyond Messages and MAUs explains why the fee must be the residual, so that the floor is a guarantee and not an arithmetic identity.

Figure 1.  The NeoMail rate card. One base fare, four ways to earn above it. Prices are indicators for pilots.

Two further sources sit outside the brand’s bill. Advertisers pay per completed ActionAd, with net revenue shared between the provider and the publishing brand. An ActionAd is a clearly labelled sponsored action built from the components email permits, not a display advertisement; AMP’s ad components do not run in email. And three add-ons attach where they are relevant: a placement premium (NiVO) that applies only while measured inbox placement beats an agreed baseline, a one-off fee for bespoke builds, and an agent-readable part of the email that is included at no charge until agents complete actions that can be measured.

Behind the products sit five charging mechanisms and one rule that keeps them honest.

The pricing rule. Base price for delivery. Premium price for capability. Event price for completed actions. Advertiser price for attention. Carry for verified profit. Never charge twice for the same value.

Each mechanism follows from where value can be seen. A capability present in every eligible email is priced as a premium on the send. A discrete, verifiable event, such as a payment or a consent record, is priced per event. Attention a partner wants to reach is paid for by the partner. A result that shows up only against a holdout is paid for from the profit it creates.

The email’s job decides what to optimise; the contract decides what to bill. Usually the two line up, and one primary meter per type of email is the right default: premium sends for Sell emails, base sends plus actions for Notify emails, Engaged Editions for Digests and Carry for recovery. An eligible send, here and throughout, is an AMP-enabled email from a registered sender to a recipient in an agreed cohort of mailbox providers that support interactive email. Section 8 explains why the definition has to be an agreed approximation.

One unit needs more caution than the rest. The Engaged Edition is the least proven of the five, because a verified action is a proxy for attention, not attention itself. A reader can value a Digest without tapping anything; another can answer a trivial quiz without caring. Counting one unit per reader per edition, however many actions, removes the incentive to clutter an edition. It does not remove the temptation to add a trivial action to every edition. The pilots have to show two things: that action-qualified editions predict later attention and purchases better than delivered editions do, and that the unit earns enough without manufactured interactions.

4   Where the Money Is: Actions, Not Formats

A format premium lifts the email bill. A priced action can double it.

Start with the obvious upgrade, a 50% premium for interactive email. Applied only to the share of a list on supported mailbox providers, say 40%, it raises email revenue by 20%, not 50%. That is worth having, but it is modest.

Now take a bank sending 10 million renewal and payment reminders a month. At 1p each, the stream earns ₹1,00,000. With an interactive premium on 40% of sends, it earns ₹1,20,000. Now suppose 1% of recipients complete a payment or renewal inside the email, and each verified action is priced at ₹1. That price is an assumption, and it would need to sit far below the cost of the reminder calls, SMS chains and branch visits it replaces. The actions add ₹1,00,000 and the stream earns ₹2,00,000 (Figure 2). The 1% counts completions across all 10 million recipients, through the interactive email and through the HTML version’s continuation to a secure page.

Figure 2.  The same 10 million emails priced three ways. Illustrative: service emails use base sends plus actions, so the premium and the action fee are not added together.

One completed action at ₹1 is worth a hundred sends. That is the arithmetic WhatsApp and the verification services discovered: the job is worth more than the journey, and pricing it is where a carrier’s revenue can grow faster than its volume.

The action business has a second property the format premium lacks: it does not depend on AMP. The secure service that records a payment, a consent or a preference from inside an interactive email also powers the HTML version’s continuation to a secure page, and can power a WhatsApp or SMS link. Actions are counted at the provider’s servers, not in the mail client. That makes NeoMail Actions channel-independent, auditable and much harder to copy than any interactive format.

Each action needs its own billing event, defined tightly enough that neither side can argue about it:

  • Tell: a validated submission, not a field touched.
  • Consent: a valid, purpose-specific record. Consent is never inferred from a click.
  • Pay: confirmed settlement, not initiation, reconciled against duplicate callbacks.
  • Book: a confirmed booking.
  • Verify: a completed check.

Each price is set from the workflow it replaces, never from the provider’s cost of delivering it. That is Meta’s logic for utility messages, applied to email. And for a production service the right structure is often a monthly minimum plus usage, with the brand paying whichever is greater. Uptime, security, audit logs and incident handling do not shrink when action volume does.

5   The Single-Fleet Lesson

IndiGo’s low fare rests on a standardised operation. So does a 1-paise NeoMail.

Ancillaries are the visible part of IndiGo’s model. The less visible part is standardisation: largely one aircraft family, one way of turning a plane around, one set of training and spares. That is what keeps the base fare low enough to fill the plane, and the extras sell on top of it.

Email has had the opposite habit. Interactive email has mostly been sold as custom work: a spin-the-wheel for one brand, a calculator for another, each designed, tested and maintained once. Bespoke work cannot support a low base fare or a cheap upgrade.

The NeoMail Factory is email’s single fleet. Its modules fall into three classes (Figure 3). Standard modules, such as polls, ratings, carousels and quizzes, are included in the premium send. Configurable modules, such as an EMI calculator, a product selector or a booking or consent flow, carry a one-off set-up fee and then ordinary usage. Bespoke work, such as a game or workflow unique to one brand, carries a development fee.

Figure 3.  Three classes of module. The Factory’s job is to move every capability towards standard.

The Factory’s job is to move capabilities from bespoke to configurable to standard. The contract can help. A brand can pay more for an exclusive build, or less if the provider keeps the right to generalise the reusable parts for others. The first bank to fund an EMI calculator pays for its creation; the tenth pays for its configuration.

Standard modules are not enough on their own. The Factory has to standardise their deployment, security, billing, measurement and support too; otherwise the hundredth AMPlet is cheap to build and as expensive as the first to integrate, test and run. A factory that produces a different bespoke project for every client is not a factory. It is an agency with better tooling.

The hundredth AMPlet must cost less to deliver than the first, or the fare cannot stay low.

6   What Airlines Got Wrong

Unbundling works until customers feel the basic journey is being held hostage.

Paid seat selection showed the limit. Over a decade, Indian airlines moved most seats into paid categories, and complaints mounted, especially from families split across a cabin. In March 2026 the civil aviation ministry directed that at least 60% of seats on every flight be free to select. Within weeks it put the rule in abeyance after airlines argued that base fares would have to rise, while keeping the requirements to disclose charges clearly and to seat passengers on one booking together.

Both sides had a point. The airlines’ argument was sound: the extras fund the low fare. So was the passengers’ grievance: charging for what feels basic turns a choice into a toll, and invites a regulator. What survived the reversal was transparency.

For email, three rules follow.

  • Never charge for the basics. Authentication, the HTML fallback, the unsubscribe link and honest reporting are in the fare, and a regulated notice never carries unrelated advertising. Deliverability is not an add-on. NiVO is priced only as measured improvement beyond an agreed baseline, never as a promise of the primary inbox.
  • Keep the bill short. One primary meter per type of email, plus an event fee only where an event carries its own value. A dozen line items per email is the procurement equivalent of a seat map priced row by row.
  • Never charge twice. A send to a customer inside a recovery programme is not billed, because the provider is paid through Carry. An action an advertiser paid for does not also justify a premium. An outcome counted once is never counted again.

PART III

The Limits

7   Where the Analogy Breaks, and Why That Helps Email

An airline sells scarce seats. Email sells useful capability, and its scarce resource is attention.

IndiGo can charge for a front-row seat because there are only a few. An interactive module is not scarce: once built, it can go into every email at almost no extra cost. So email cannot copy airline ancillaries by charging for scarcity. It can charge only for usefulness: the action completed, the moment captured, the call avoided.

What is scarce in email is the reader’s attention, and that changes the rules. An airline can sell an extra without making the flight worse. An email that carries too much loses the attention every price depends on. That is why the Digest is priced per Engaged Edition rather than per send, why an edition carries at most one partner action, and why a provider paid on actions must never manufacture them.

The analogy breaks in a second, more useful way. An airline’s passenger pays for almost everything; the main exception is the co-branded card, where banks buy miles, which is the model behind Mu that I explored in Can Mu Make Email Invincible?, Email has more payers than that. An advertiser pays for a completed action inside a Digest. The profit pool pays for verified recovery. Where a Digest earns enough from ActionAds, its revenue is credited against the brand’s bill and the brand’s net cost falls towards ZeroCPM.

No airline can make the seat free because a partner paid for the passenger’s attention. Email can.

8   How to Bill What You Cannot See

A price is only as good as the unit both sides can audit.

An airline knows who sat in which seat. An email provider does not always know what the reader saw. An interactive email travels as one message carrying AMP, HTML and plain-text parts, and the mail client decides which to display. Interactive email is supported principally by Gmail, Yahoo Mail and Mail.ru, with a few smaller clients, and only for senders registered with each provider; to anyone else, the provider shows the HTML version. AMP’s own analytics and pixel components do not run in email, and requests pass through the provider’s proxy. Ordinary image-based tracking can sit in the message, but it cannot show which part of the message a person saw. A provider that promises to bill only when the interactive version renders is promising a number it cannot produce. Nor can it price the premium on message size: the extra parts add bytes, not value.

So each meter needs a unit both sides can check, and a plain statement of what that unit cannot prove (Table 3).

Table 3.  Five meters, what each counts and what each cannot prove.

Meter Billing unit Evidence and its limits
Interactive An eligible send: AMP-enabled, from a registered sender, to an agreed cohort of supported providers, identified from the domain’s MX records and other available evidence MX records show where mail is received, not which app a person reads it in or whether interactive email is switched on. The unit is an agreed approximation, not a record of renders
Living The same eligible send, at the Living rate Live-content requests at open reach the provider’s servers and show the live components are working, but caching makes them a check, not a meter. Heavy custom lookups can be sold separately as a Live Data Service on a minimum plus metered requests, never billed twice
Actions A validated, deduplicated completion recorded at the server Reversals, refunds and retries must be defined before launch
Digest One Engaged Edition per reader per edition, however many actions the reader takes Quiet reading may have value the unit does not count
Recovery Lift against a concurrent randomised holdout Statistical uncertainty; small cohorts cannot be read with confidence

One discipline sits above the meters. Every pilot reports four numbers: the brand’s incremental value, the provider’s incremental revenue, the provider’s incremental cost, and the share of the value the provider captures. A feature that creates ₹10 of value and lets the provider capture ₹0.50 at a cost of ₹0.80 is a favour, not a product.

PART IV

Tests

9   How This Could Be Wrong

Each mechanism has a condition under which it fails, and each can be read within a year.

Table 2.  Kill conditions, mechanism by mechanism.

Test Kill condition What it would mean
Premium sends Fewer than half of pilot brands renew interactive sends at a premium of at least 30% after twelve months Interactivity is a feature, not a revenue line; the format premium reverts to the base fare
Actions In-email completions on service emails stay too rare to move revenue, or disputes over what counts as an action persist after billing events are defined Actions belong on websites and apps; the action fee does not scale through email
Basics Brands treat the premiums as charges for basics and push back, as passengers did over seats The rate card needs fewer, more visible premiums
Platform Mailbox providers restrict or withdraw support for interactive email The format premium shrinks; Actions survive through the HTML version and secure web pages
Digest unit Action-qualified editions predict later engagement and purchases no better than delivered editions, or the qualifying actions turn trivial The Engaged Edition is a poor proxy for attention, and the Digest needs a different meter
Attention Email revenue rises while engaged reach and retained attention fall, or complaint rates rise Yield is being bought with the attention it depends on
Capture Across pilots, the provider’s incremental revenue does not cover its incremental cost to serve The pricing is generous to brands but is not a business

The platform risk deserves emphasis, because it is the one the provider does not control. If interactive email lost support, the premium on eligible sends would disappear. The Actions business would not, because it was never an AMP feature. That is one more reason to build it first.

10   What a Provider Should Do on Monday

Three pilots, three mechanisms, one matrix.

  • A retail Sell email: 1p HTML against a 1.5p NeoMail, measured on completed actions and contribution per thousand delivered, against a concurrent control.
  • A banking or insurance service stream: NeoMail Actions on a monthly minimum plus a fee per verified action, measured on cost per completed task and on the calls and reminders it avoids.
  • A publisher NeoLetter: existing send economics plus ActionAds, measured on net yield per thousand engaged editions and on reader retention.

Alongside the pilots, build a pricing matrix: one row per capability, with columns for the job it does, whether it is standard, configurable or bespoke, what the provider must keep running, the billing event, the indicative price, the cost to serve, the margin and the proof a customer can check. Every proposed premium has to justify itself there before it reaches a rate card. One test sits above the matrix: can the brand explain the extra charge to its finance team in terms of completions, lower operating cost, retained attention or avoided spend? If not, the feature belongs in the fare.

For thirty years email has been sold like a bus ticket: one price, whatever the journey achieved. Airlines learned to sell the seat cheaply and charge for what passengers valued. Couriers charge for proof, messaging platforms for the job, verification services for success. Email can keep the 1-paise send and get paid for what the email gets done.

The send is the fare. The job is the business.

**

Key points

  • Airlines, couriers, messaging platforms and verification services followed a recurring pattern out of commodity pricing: a cheap, standardised journey and a separately priced job.
  • Email has priced only the journey. A NeoMail keeps the 1-paise send and adds four ways to earn: NeoMail at about 1.5p and 2p per eligible send, NeoMail Actions, the Brand Digest and Recovery.
  • Five mechanisms: base price for delivery, premium for capability, event price for actions, advertiser price for attention, Carry for verified profit. Never charge twice.
  • A format premium lifts the bill modestly; priced actions can double it. One completed action at ₹1 is worth a hundred sends.
  • Actions are counted at the server and work in AMP, in the HTML version and on the web, which makes them the most defensible engine.
  • The Factory is email’s single fleet: the hundredth module must cost less than the first.
  • Never charge for the basics, keep the bill short and bill only units both sides can audit. Email’s scarce resource is attention, and it has payers airlines lack: advertisers and the profit pool.

Beyond the Base Fare  ·  Rajesh Jain  ·  26 October 2026  ·  V2