The Next Big Business Models: Where New Markets Are Coming From

Every technology shift produces two kinds of companies. The first kind uses the new capability to do an old thing more cheaply. The second kind notices that the new capability changes what customers are willing to pay for, how they are willing to pay, and who they are willing to pay, and builds a different kind of business around that change. The first kind earns a few good years. The second kind occasionally rewrites an industry. This study looks for the second kind in the autumn of 2026. It asks where new markets are coming from, which of the business models being tried right now show signs of lasting, and how a founder, especially one outside the usual technology centres, can tell a durable model from a fashionable one. As with my other reviews this year, I have tried to rely on named sources, to separate measurement from forecast, and to say plainly when a number comes from a party with something to sell.

I want to begin with a distinction that will carry through the whole piece. A business model is not a product and it is not a price list. It is an answer to four questions: who pays, for what, how much, and why will they keep paying when a cheaper alternative appears. New technology changes the answers to all four. It changes who pays, because the buyer may be a company instead of a person, or a software agent acting for a person. It changes what is being bought, because customers increasingly buy a finished result instead of access to a tool. It changes how much, because the cost of delivering a unit of work is falling in some places and rising in others. And it changes why they stay, because the barriers that protected old businesses, such as the cost of building software or the scarcity of expertise, are eroding. Everything below is a different answer to one of those four questions.

The clearest case is pricing, so start there. For roughly two decades, enterprise software was sold by the seat, on the assumption that the number of people using a product tracked the value it created. That assumption is under strain wherever software now does work that a person used to do. If an agent resolves a customer enquiry, the company needs fewer human agents and therefore fewer seats, even though the value delivered has gone up. Analysts have been predicting the end of seat pricing for a while, and early 2026 gave the prediction a market test. According to a Bain analysis cited in trade coverage, around two trillion dollars of software market value was wiped out in the first months of the year in a sell-off that the press called the SaaSpocalypse, as investors reassessed how much of the sector's revenue depended on the headcount of its customers. I would treat the exact figure with caution, since it appears in secondary reporting, but the direction is not in dispute: the market has decided that seat based revenue is more fragile than it looked.

What replaces it? Vendors and analysts offer a range of data, and the data are mostly consistent in direction. A survey described by one pricing consultancy found the share of software companies using pure seat pricing falling from twenty one percent to fifteen percent within a year, while hybrid models rose from twenty seven percent to forty one percent. Gartner is quoted as expecting at least forty percent of enterprise software spending to move towards usage, agent or outcome based pricing by 2030, and IDC as expecting seventy percent of vendors to move away from pure per seat models by 2028. A survey of three hundred software chief executives in April found that almost all planned to retire seat pricing within two years, even though nearly as many said seat pricing currently matched the value of their product, a contradiction that tells you how unsettled the thinking is. Several of these numbers come from vendors and consultancies that sell pricing advice, so I would hold them loosely, but they agree with each other and with what I observe in the market.

The most discussed alternative is outcome based pricing, in which the customer pays only when the software delivers a defined result. Customer support is the pioneering category because a resolved conversation is easy to define and count. Intercom has charged about ninety nine cents per resolution, Zendesk has quoted figures in the range of one fifty to two dollars, and HubSpot reportedly cut the price of its customer agent to fifty cents per resolved conversation in April 2026, a move interpreted as a signal that outcome pricing at volume can undercut per seat pricing and that the vendor willing to cut first can capture the adoption window. Sales tools that charge per meeting booked and legal tools experimenting with usage based billing follow the same logic. In principle, outcome pricing is elegant: the vendor is paid for the thing the customer wants, and the customer takes no risk.

In practice, pure outcome pricing runs into three walls, and the market's response has been instructive. The first wall is measurement: someone has to decide what counts as a resolution, and both sides have incentives to argue. The second is predictability: customers like to forecast their spending and vendors like to forecast their revenue, and variable pricing makes both harder. The third is cost: when the vendor pays for computing on every task, an outcome price that ignores cost can scale losses. The result, according to several pricing analyses, is that hybrid models are winning, combining a fixed platform subscription with a variable fee for usage, credits or outcomes. I think this is right and I would go further. The deepest reason hybrids win is that they split risk sensibly. The fixed fee pays for the vendor's costs of keeping the system running and gives the customer a budgeted number. The variable fee shares the upside of heavy use. A founder designing pricing in this environment should expect to build a measurement system as part of the product, because whoever controls the definition of an outcome controls the revenue.

A second source of new business models is the arrival of software agents as customers. In my review of retail I described how agent mediated shopping is being built. The protocols are real: OpenAI and Stripe's checkout protocol, Google and Shopify's Universal Commerce Protocol announced in January, payment network programmes for agent transactions, and an open standard for connecting models to data and tools. The status is early. A mid 2026 assessment relayed by a commerce software vendor judged that humans still decide and complete checkout in the vast majority of cases, ChatGPT's instant checkout was reported as paused after a limited rollout, and a transaction fee of about four percent was reported for merchants using it. The model to watch is less the shopping assistant itself than the infrastructure around it: product data feeds that agents can read, identity and authorisation systems that let an agent spend within limits, payment tokens designed for software, dispute resolution when an agent makes a mistake, and analytics for merchants who need to know why an agent chose a competitor. Every new channel in the history of commerce produced a layer of businesses serving the people who sold through it, and I expect agent mediated commerce to be no different.

The opportunity for a small business in this layer is genuine and underappreciated. A merchant that makes its catalogue clean, consistent and machine readable, with accurate stock, clear return rules and structured reviews, will be easier for agents to recommend. A service that helps small merchants do this, across platforms and protocols that are changing quickly, is a plausible business with real demand. The caution is that platform owners can change the rules, and a business that depends entirely on one protocol or one platform's goodwill is exposed. The sturdier model is a neutral layer that works across several agents and several storefronts, and treats each protocol as an integration and not as a foundation.

Third, content is being turned into a licensed asset, which creates a new kind of seller. In my review of media I described the emerging market in which AI companies pay for content. Disclosed deals number in the dozens, news dominates, and the structure has shifted from one off payments for training data towards ongoing payments for live access and attribution. The biggest reported deal gives a single publisher group up to fifty million dollars a year for three years, while the long tail of independent publishers has so far earned little. The business model that interests me is the intermediary that makes a market for the long tail: a collective licensing body that pools the rights of many small publishers, negotiates with the buyers and distributes the revenue. Music has a parallel in licensed services built on catalogues whose owners have agreed to participate, and image and video libraries already sell to model builders. A second layer is the business that produces content specifically for this market, such as specialised datasets, expert annotated examples and verified, structured knowledge in languages and domains that are thinly represented online. For readers in Bangladesh, I would flag Bengali as precisely such a domain: high quality, properly licensed Bengali text, speech and domain knowledge is scarcer than the number of speakers would suggest, and scarcity is what creates a market.

Fourth, physical automation is acquiring its own business model, which is robots as a service. Instead of selling a machine for a large upfront price, a company charges per task or per hour, takes responsibility for uptime and maintenance, and absorbs the risk that the technology does not yet work reliably. One humanoid maker has reported more than a hundred thousand totes moved for a logistics customer under such a contract. The logic is the same as cloud computing's: customers who are unsure about a new technology will try it if they can pay for results and walk away, and the vendor can learn from many deployments and improve faster than any single buyer could. The risk is on the vendor's balance sheet, since the robots are assets that must be financed, maintained and replaced. I expect financing and insurance to become distinct businesses in this space, as they did around aircraft and shipping containers, and I think the companies that finance and operate fleets may prove more durable than some of the manufacturers.

Fifth, and most relevant to anyone starting out, the cost of building has collapsed, which makes the tiny company a viable business model in its own right. A small team with good tools can now ship products that needed a department five years ago. This is visible in the way software is being built, in the number of one person companies that report real revenue, and in the way venture funding has polarised, with the largest rounds going to a handful of firms while many small companies grow without outside capital. It also changes what a company is for. If the cost of producing software approaches zero, the scarce things become distribution, trust, taste, domain knowledge and the ability to keep a system correct over years. A company that sells deeply specific expertise wrapped in software, such as accounting for a particular industry, compliance for a particular regulator or a school management system for a particular country, can serve a market too small for a large vendor to bother with and too specialised for a general tool to serve well. I believe this is where a great many sensible new businesses will come from, and it is a pattern that favours people who know a specific market intimately.

Sixth, the creator economy deserves a careful look, partly because its size is so frequently misstated. Estimates for the global market in 2026 vary enormously, from around two hundred billion dollars to more than three hundred and twenty billion, depending on the research firm and on what it counts. One report puts it at about three hundred and ten billion, another at about three hundred and twenty three billion, a third at around two hundred and sixteen billion, and others between two hundred and two hundred and fifty billion. That spread of more than fifty percent in one year should tell you something about the reliability of the whole genre of market size reports. The more solid observation is structural: well over two hundred million people worldwide describe themselves as creators, a few million earn a living from it, and most of the income is concentrated at the top. The platforms that carry creators set the economics through revenue shares and algorithmic distribution, and creators have learned the lesson that dependence on a single platform is a risk.

What is new in the creator economy is the shift from renting an audience to owning one. Newsletters, memberships, paid communities, direct sales of digital products and courses, and creator run software are all attempts to turn attention that lives on someone else's platform into a relationship that does not. AI changes the economics again by making it cheaper for a creator to produce, translate, adapt and package content, and to build tools for a community. It also floods the market with competent content, which raises the value of a distinctive voice and a trusted relationship. My view is that the sustainable creator business of the next few years looks less like an influencer with a large following and more like a small media and education company with a clear audience, several revenue streams and its own distribution. For a creator who works in Bengali, the audience is large and underserved in paid, high quality formats, and the opportunity is substantial for those who build carefully.

Seventh, the old idea of vertical specialisation is returning in new form. General purpose models are available to everyone, which makes the model itself a poor moat, and so value is migrating to those who combine the model with proprietary data, regulatory permission, specialised workflow and distribution into a particular industry. Legal, medical documentation, insurance claims, construction, logistics, accounting and agriculture are all seeing companies that take a capable general model and wrap it in the knowledge and integrations of a single field. These businesses tend to price on outcomes or hybrid terms, since they replace labour in a defined process, and they tend to grow by embedding themselves so deeply in a customer's operations that replacing them is costly. I think this is the most reliable category of new business for the next five years, because it follows an old and tested pattern in which specialised software survives by knowing its customer better than anyone else.

There is a different kind of opportunity in the infrastructure of trust. As more content is generated by machines, more transactions are initiated by software and more decisions are influenced by models, the value of verification rises. Businesses that can verify identity, prove the origin of content, audit an automated decision, certify that a model behaves within limits, or reconcile what an agent did with what a customer authorised are selling something that becomes more valuable as the volume of automated activity grows. Payment infrastructure is the clearest example. The stablecoin rules now moving through American regulators, which I described in my review of finance, create a regulated space in which compliance itself is a product, and the companies that make it easy for others to comply may prove sturdier than those issuing the tokens. In the same spirit, anyone who builds reliable reconciliation, monitoring and audit tools for automated systems is addressing a need that every organisation using automation will eventually feel.

How should a founder tell a durable model from a fashionable one? After reading all of the above, I have come to rely on five questions, and I offer them as a practical framework. First, who pays, and are they paying because the product reduces a cost or creates revenue they can measure? A model whose value is a feeling, however pleasant, is fragile. Second, what happens to your price when the underlying technology gets cheaper? If your price is tied to a cost that is falling, your revenue will fall with it, unless you are selling something that costs more to provide than the technology, such as trust, integration or accountability. Third, what stops a well financed competitor from copying you in a quarter? The answer should be something other than a feature: proprietary data, a licence, a relationship, a network, or deep embedding in a customer's process. Fourth, who controls the rules? If a single platform can change the terms on which you reach customers, you have a dependency, not a business, and you need a plan for it. Fifth, can the unit economics survive an unfavourable year? Plan for the case in which computing costs rise, a key customer leaves, or a regulator changes the rules, and see whether the business still works.

Let me apply that framework to a few models to show how it works. Outcome based pricing for customer support passes the first test, since a resolved enquiry has clear value, but it is exposed on the second and third, since cheaper models and aggressive competitors can compress prices quickly, as the fifty cent price cut suggests. It becomes durable only when the vendor owns something more than the model: integration with the customer's systems, accumulated knowledge of the customer's products, a track record of accuracy. A neutral layer that makes merchants ready for agent shopping passes the first and second tests but depends on the third and fourth, since platforms may build the same capability themselves. A collective licensing body passes the third test well, since it aggregates something no individual can, but depends on the willingness of buyers to pay, which is uncertain. A specialised vertical product built on a general model passes most tests when it has a regulatory or data advantage and fails when it is only a thin interface. A creator business with a direct audience passes the fourth test, ownership of the relationship, and is sturdy to the extent that it passes the third, through a voice that cannot be copied.

There is a set of failure modes I have seen repeatedly, and naming them may save someone a costly mistake. The first is mistaking usage for revenue: a product that many people try and few pay for is a hobby with a server bill. The second is pricing below the true cost of the computation behind each task, which works while venture money subsidises it and breaks the moment it stops. The third is building on a single provider or protocol and discovering that the terms have changed. The fourth is confusing a large market size estimate for evidence of demand, when, as the creator economy figures show, such estimates can differ by half between two reputable firms. The fifth is neglecting the unglamorous parts of a business, such as billing, support, compliance and measurement, which decide whether an interesting product becomes a company.

Now for my own predictions, which are judgements and not forecasts from any institution. Over the next eighteen months I expect hybrid pricing to become the default for software that includes agent features, with pure seats surviving mainly in categories where headcount still tracks value, and pure outcome pricing surviving where results are easy to attribute, such as customer support and collections. I expect a visible consolidation among vendors that cannot explain their cost per task. I expect agent mediated commerce to grow slowly through 2027, with the first large wins in repeat purchases, travel and business procurement, where the buyer's preferences are clear and the cost of error is small, while impulse and fashion purchases stay human. I expect collective licensing arrangements for independent publishers and creators to appear, and I expect disputes about attribution and payment to increase. I expect robots as a service to expand in logistics before manufacturing, since the tasks are simpler and the contracts easier to measure. And I expect the share of new companies that are small, profitable and bootstrapped to rise, as the cost of building continues to fall and as the concentration of venture capital makes outside funding harder to obtain for ordinary businesses.

What do I think will turn out to be overrated? I would be sceptical of any model whose only moat is access to a general model that everyone else can also buy, and of forecasts that treat a large addressable market as if it were a customer. I would be sceptical of the idea that outcome pricing will simply replace everything else, since the evidence points to hybrids. I would be sceptical of claims that humanoid robots will transform manufacturing within two or three years, since the deployments to date are narrow. And I would be sceptical of the notion that a single platform will capture all of agent mediated commerce, since retailers have strong reasons to build their own agents and keep their customers.

What do I think is underrated? The infrastructure of trust and measurement seems underrated to me, because every new automated activity creates demand for verification, audit and reconciliation, and these are businesses that do not depend on any single technology winning. Specialised local businesses are underrated, because the combination of cheap building and deep domain knowledge opens markets that large vendors will not serve. Language specific businesses are underrated, especially in languages that global products serve poorly. And financing and insurance for new kinds of assets, from robot fleets to computing capacity, are underrated, because every new asset class has needed its own financiers and they have often earned more than the operators.

I should speak directly to the perspective of a founder in Bangladesh, since that is where I work and where I see the least coverage of these ideas. There are advantages here that the global conversation overlooks. Engineering talent is plentiful and affordable. Mobile payments are already part of daily life, which means the infrastructure for small, frequent, digital transactions exists. There is a large language community whose needs are served poorly by products built elsewhere. Small businesses are numerous and many remain under digitised, which means a tool that helps them is solving a real problem and not marketing to the already convinced. The disadvantages are equally clear: energy and currency pressures raise costs, a dependence on foreign platforms and foreign currency pricing for computing creates exposure, and access to patient capital is limited. The model that I think fits these conditions best is the focused, profitable company that solves a specific problem for a specific kind of customer, charges in a way that covers its costs from the first customers, and builds a moat from local knowledge, language, payments integration and trust. It is less glamorous than a global platform and much more likely to survive.

There are of course limits and risks. Business models are easy to describe after the fact and hard to predict before it, and I have no special immunity from that difficulty. Several of the numbers cited here come from consultancies and aggregators with a commercial interest in the story, and I have tried to flag them. Some of the models above may fail for reasons that are not yet visible, such as a regulatory shift, a change in a platform's terms or a sudden fall in the cost of computing that erases a margin. The agent commerce market in particular is in flux, with protocols competing and products being launched and paused within months. I would urge anyone building on these ideas to test them cheaply, to keep their commitments small until revenue arrives, and to review their assumptions every quarter.

To summarise, new markets in 2026 are coming from five directions. They come from the repricing of software as it shifts from seats to usage, outcomes and hybrids. They come from software agents acting as buyers and the infrastructure that serves them. They come from content, data and expertise becoming priced assets, particularly in underserved languages and domains. They come from physical automation sold as a service, with its attendant financing and insurance. And they come from the collapse in the cost of building, which makes small specialised companies viable and shifts value towards trust, distribution and knowledge. The sustainable businesses among them will be those that can answer four questions clearly: who pays, for what, how much, and why they will keep paying. The fashionable ones will be those that can answer only the first. If you remember one thing from this study, let it be that a new technology rarely creates a durable business by itself. It creates the conditions in which a business with real customers, honest economics and something hard to copy can be built more cheaply than ever before, and the founders who understand that will have an advantage that no tool can supply.

Sources: Fluenta on the end of per-seat pricing and the early 2026 software sell-off (https://www.fluentaone.com/blog/the-end-of-per-seat-software-pricing-was-predicted-but-the-market-decided-otherwise); Userpilot on SaaS pricing models after the AI sell-off (https://userpilot.com/blog/saas-pricing-models/); Pickaxe on AI agent pricing models (https://pickaxe.co/post/ai-agent-pricing-models); Deepak Gupta on outcome-based pricing (https://guptadeepak.com/ai-native-outcome-based-pricing-2026/); NxCode SaaS pricing strategy guide (https://www.nxcode.io/resources/news/saas-pricing-strategy-guide-2026); commercetools, agentic commerce radar, autumn 2026 (https://commercetools.com/blog/the-agentic-commerce-radar-key-market-shifts-insights); Google, Universal Commerce Protocol announcement (https://blog.google/products/ads-commerce/agentic-commerce-ai-tools-protocol-retailers-platforms/); Opascope on agentic commerce protocols (https://opascope.com/insights/ai-shopping-assistant-guide-2026-agentic-commerce-protocols/); Media and the Machine, AI content licensing deals (https://mediaandthemachine.substack.com/p/ai-content-licensing-deals-june-2026); PR News licensing tracker (https://everything-pr.com/ai-licensing-tracker); Digiday on publisher AI licensing revenue (https://digiday.com/media/media-briefing-publishers-cautiously-count-ai-licensing-as-notable-revenue-amid-programmatic-strain-in-q1-earnings/); Solid Market Research on humanoid robots and robots-as-a-service deployments (https://www.solidmarketresearch.com/post/humanoid-robots-cross-the-pilot-threshold-where-factory-deployment-actually-stands-in-2026); Research and Markets, Creator Economy Market Report 2026 (https://www.researchandmarkets.com/reports/6226071/creator-economy-market-report); Grand View Research, Creator Economy Market (https://www.grandviewresearch.com/industry-analysis/creator-economy-market-report); Coherent Market Insights, creator economy forecast (https://www.coherentmarketinsights.com/industry-reports/global-creator-economy-market); Crunchbase News on H1 2026 venture funding (https://news.crunchbase.com/venture/global-startup-exits-ipo-ma-soar-ai-q2-h1-2026/); Federal Register, OCC stablecoin proposed rule (https://www.federalregister.gov/documents/2026/03/02/2026-04089/implementing-the-guiding-and-establishing-national-innovation-for-us-stablecoins-act-for-the).