Your Market Knowledge Is Worth More Than a Million-Dollar Token Budget

There's a narrative forming in tech right now that goes roughly like this: intelligence is becoming a commodity, the companies that can afford the most tokens win, and everyone else should probably start updating their CVs.

It's a tidy story. It's also wrong.

The token economy is real. Per-token inference costs are cratering — somewhere between 10x and 200x per year depending on the model tier. Enterprise LLM spend has risen from $4.5 million to $7 million on average, with projections blowing past $11 million in 2026. OpenAI is reportedly pricing agent tiers at up to $20,000 a month. The big players are building horizontal platforms, running agents around the clock on broad workflows that every Fortune 500 shares.

But if you stop the analysis there — at the headline numbers and the enterprise dashboards — you miss the most interesting part of the entire story. The part that matters if you're not Goldman Sachs.

Intelligence Is Purchasable. Distribution Is Not.

Goldman will spend more on inference this year than most startups will spend in their entire existence. JP Morgan will commit to consumption contracts that would make a bootstrapped founder weep. If the competitive axis were simply “who can buy the most intelligence,” the incumbents would win by definition. They have the revenue base. They have the infrastructure. They have the procurement departments to negotiate custom API agreements with every hyperscaler on the planet.

And none of that matters if they can't sell what they build.

Intelligence is a commodity. You can purchase it by the token from half a dozen providers. But distribution — the channel, the relationship, the trust, the understanding of what a specific customer in a specific market actually needs — that is not purchasable. It never has been. And in a world where building software is collapsing in cost, distribution becomes the only durable advantage.

Goldman can run more inference than a five-person startup building AI-powered inventory management for independent restaurant chains. But Goldman cannot sell to those restaurants. Goldman doesn't know their pain. Goldman doesn't understand that the real problem isn't “inventory optimisation” in the abstract — it's that the owner of a 50-location chain is losing £40,000 a year because their Tuesday morning produce orders are based on gut feeling and a spreadsheet from 2019. Goldman will never build that product. It's too small, too specific, too far from their core business.

This isn't a consolation prize. It's the entire game.

Think about what's happened to retail banking. The big banks have spent billions on digital transformation. They have the best infrastructure, the deepest pockets, the most sophisticated models. And yet: Monzo, Starling, and Revolut ate their lunch in the UK. Not because they had better technology — the underlying banking infrastructure is largely the same. They won because they understood a specific customer segment (digitally-native consumers frustrated by branch-era UX) better than incumbents who were trying to serve everyone. They had distribution insight. They had taste. They had specificity.

The same pattern is about to play out across every industry, accelerated by the token economy. The big players will build broad platforms. The specialists will build the products people actually want.

The Total Addressable Market Is Exploding

Here's the mechanism most people miss: when the cost of building software falls, the number of problems worth solving with software expands.

This is Jevons' paradox applied to the total addressable market for software itself. Deloitte's 2026 software industry outlook puts it plainly: creating software is faster and cheaper than ever, and the definition of what constitutes a software company is expanding accordingly. Every enterprise has a backlog of projects that were never economically viable. The internal tool that would save 200 hours a year but cost 2,000 to build. The integration that would unlock a new revenue stream but require a team of four for six months. The niche vertical that could work but couldn't justify the engineering allocation.

When intelligence gets 10x cheaper, those backlogs become gold mines. But here's the crucial bit: the enterprises themselves often won't mine them. They'll focus on the big horizontal wins. The niche stuff? That's where the specialists come in.

Six months ago, building AI-powered compliance software for independent dental practices wasn't a viable business. The market was too small, the build cost too high, the unit economics didn't work. Today, a founder who spent 15 years managing dental practices can build that product in weeks. Not because they're a brilliant engineer — because intelligence is cheap and their domain knowledge is the part that can't be purchased.

This is happening in real time across dozens of industries. Construction estimation. Veterinary practice management. Independent brewery operations. Marine logistics. Every one of these niches has people who know exactly what software they need and have never been able to justify building it. The token economy just made all of those niches viable simultaneously. The total addressable market for software isn't growing linearly — it's growing combinatorially, because every niche that was previously too small to serve is now big enough.

Why This Is the Specialist's Moment

Bessemer Venture Partners' vertical AI playbook makes the case explicitly: vertical AI rewards insider expertise more than horizontal SaaS ever did. You're not digitising generic processes. You're reimagining complex, nuanced workflows in specific industries — and the person who understands those workflows has an advantage that no amount of compute can replicate.

The construction scheduling expert is now a developer. The insurance compliance analyst can now build tools instead of just using them. The e-commerce operator who's spent a decade in the weeds of fulfilment logistics can now encode that knowledge into a product and sell it back to their own industry.

These people don't need to manage token budgets in the millions. They don't need to negotiate custom API agreements with hyperscalers. A £200/month AI subscription aimed precisely at the right problem creates more downstream value than a £20,000/month agent budget pointed vaguely at the wrong one.

That's not motivational poster material. That's arithmetic.

The Numbers Behind the Niche

AI-native companies are running at 3-5x the revenue per employee of traditional SaaS businesses. Bessemer's data shows top AI companies hitting $1.13 million ARR per full-time employee — four to five times the typical SaaS benchmark. A $10 million ARR AI startup might operate with 15 people where a traditional SaaS company would need 55 to 70.

But here's what makes that number transformative for specialists rather than just interesting for VCs: it means the minimum viable business is shrinking. If you can build and run a profitable product with two or three people — or even alone — then the market only needs to be big enough to sustain that team. You don't need a $500 million TAM to justify VC funding. You need a £50,000 MRR niche to build a genuinely good life.

The solo founder speculation — the breathless “when will the first one-person billion-dollar company happen” discourse — misses the point entirely. The interesting insight isn't one exceptional person hitting an arbitrary milestone. It's that the minimum viable team for software is converging on one, which means going independent is no longer a lifestyle trade-off. It's a rational economic choice for anyone with deep domain knowledge and sufficient AI fluency.

There are thousands of those people. Most of them don't think of themselves as founders yet. Many of them don't think of themselves as technical. Both of those self-assessments are becoming outdated.

The maths is stark. If a solo founder can build a product generating £30,000 MRR in a niche they know intimately — and their operating costs are a few hundred pounds in AI subscriptions plus their own time — that's a £360,000/year business with margins that would make a traditional SaaS investor cry with joy. No VC required. No team of 50. No office. No seven rounds of funding before you can pay yourself.

Scale that to a two or three person team, and you're looking at a £1-2 million ARR vertical AI company that's profitable from month six, built on expertise that took a decade to accumulate and costs nothing to maintain. This is the business model that VCs aren't talking about because it doesn't need them. And it's the business model that big tech can't replicate because their cost structures demand bigger markets.

What the Big Players Can't Do

Enterprise AI strategy is, almost by definition, horizontal. When you spend $7 million a year on LLM costs, you need to justify that spend against broad business outcomes that affect the entire organisation. You build platforms. You standardise. You create centres of excellence and governance frameworks and acceptable use policies.

All of which is sensible and necessary and completely irrelevant to the owner of a 12-person HVAC company in Birmingham who needs a system that handles his quoting, scheduling, and compliance paperwork.

The enterprise playbook optimises for breadth. The niche playbook optimises for depth. And depth — real, hard-won understanding of how a specific market actually works — compounds in ways that breadth cannot.

A specialist who knows the HVAC industry builds a product that handles the specific compliance requirements of F-Gas regulations. That handles the scheduling constraints of emergency callouts. That integrates with the three wholesalers that 80% of independent HVAC firms actually use. Every one of those details is invisible from the enterprise vantage point. Every one of them is the difference between a product that gets adopted and one that gets abandoned after the free trial.

This is why horizontal AI companies keep hitting the same wall. They build impressive demos. They raise enormous rounds. They hire brilliant engineers. And then they struggle with adoption because the product works in the abstract but fails on the specifics. The AI can process natural language beautifully, but it doesn't know that HVAC engineers in the West Midlands call a particular part by a different name than the manufacturer's catalogue lists. It doesn't know that the quoting process for a commercial refit is fundamentally different from a domestic installation, even though they look identical in a feature requirements document.

That kind of knowledge takes years to acquire. It can't be scraped from the internet. It can't be fine-tuned into a model. It lives in the heads of the people who do the work, and those people are about to become the most important product designers in the software industry.

This is why AI-native companies are reaching $100 million ARR in one to two years while traditional SaaS companies take five to seven. The AI-native builders aren't smarter. They're operating in a world where the cost of building has collapsed and the cost of knowing your customer hasn't changed at all. In that world, domain expertise is the competitive advantage. Everything else is a commodity input.

The Time Window

This window won't stay open forever, which is why it matters that you understand this now rather than in 18 months.

Right now, most domain experts haven't realised they can build software. Most haven't connected the dots between their industry knowledge and the AI tools available to them. Most are still thinking of AI as something that writes blog posts or generates images — not something that can encode their operational expertise into a sellable product.

That gap between awareness and capability is your window. Every month that passes, more people figure it out. More niches get claimed. More vertical AI companies plant their flags in markets that were invisible six months ago.

The specialists who move first compound their advantage. They get the customer relationships, the feedback loops, the domain-specific training data, the word-of-mouth in tight-knit industries where everyone knows everyone. By the time a horizontal platform decides a niche might be worth pursuing, the specialist has an 18-month head start in a market where switching costs are high and trust is everything.

Consider what “18-month head start” actually means in a vertical market. It means you've iterated through dozens of customer conversations. You've built the integrations that matter — the ones the customer asked for, not the ones that look good on a feature comparison page. You've earned the referrals. In a market where the owner of a successful plumbing company asks the owner of another successful plumbing company “what software do you use,” that referral is worth more than £100,000 in paid acquisition. You can't buy it. You can't hack it. You can only earn it by being genuinely useful to people who talk to each other.

This is the moat. Not intellectual property. Not a proprietary model. Not a data flywheel. Trust, in a tight market, among people who know each other. It's the oldest competitive advantage in business, and the token economy is making it more valuable, not less.

This isn't about outspending anyone. It's about knowing something that no token budget can teach.

The Uncomfortable Inversion

For decades, the technology industry has operated on a simple hierarchy: technical skill at the top, domain knowledge as a supporting function underneath. Engineers built things. Business people told them what to build. The further you were from the code, the less power you had in the organisation.

The token economy inverts this. When building is cheap and intelligence is a commodity, the scarce resource isn't the ability to write code. It's the ability to know which problems are worth solving, for whom, in what order, with what constraints. That knowledge lives in the heads of people who've spent 10 or 15 or 20 years in a specific industry. People who have never opened a code editor. People who couldn't explain what an API is if you asked them.

Those people are now, functionally, product designers. They just don't know it yet.

The ones who figure it out first — who combine their hard-won domain knowledge with AI tools that are becoming cheaper and more capable every quarter — will build the next generation of vertical software. Not from Silicon Valley. Not from venture-funded offices with catered lunches. From the industries themselves. From the people who actually understand the work.

If you're reading this and you've spent years in an industry that frustrates you with its outdated tools and manual processes, you're not behind. You're exactly where you need to be. The engineers who could build software but didn't know your industry? They've been trying and failing for years. The enterprise platforms that promised to modernise your sector? They built something generic and moved on to the next vertical. Your knowledge — the stuff that feels so obvious you don't even think of it as expertise — is the scarce resource now. The intelligence to build on top of it is a commodity you can buy for the price of a monthly subscription.

That's the real story of the token economy. Not the $20,000/month AI employees at the top. The thousands of niche products at the bottom that nobody sees coming until they've already won their markets.

Sources: A16Z Enterprise AI Survey 2026 | Bessemer Vertical AI Playbook | Deloitte 2026 Software Outlook | Bessemer AI Supernova Revenue Data | SaaStr — Threat Vectors in B2B Software 2026