The Economics of One

Hey now.

Everyone wants to build the next operating system for your life. Perplexity is doing it. A dozen well-funded startups with sharp landing pages and sharper pitch decks are doing it. The marketing is good. Some of the products will be too.

But I keep coming back to a question that nobody building these things seems to want to answer publicly: what are the actual economics here?

Not revenue projections. Not TAM slides. The real question — what does it cost to build this kind of product in 2026, and does that cost justify the structure these companies have chosen?

The Trapped Gambler

Perplexity has raised over $1.2 billion. They’re valued at north of $20 billion. They’ve committed $750 million to Microsoft Azure over three years. They employ over a thousand people. And in 2024, they spent 164% of their revenue just on AWS, Anthropic, and OpenAI API costs — before a single salary was paid.

They started in August 2022. Back then, building an AI search product required significant engineering teams. GPT-4 didn’t exist yet. Claude was a research project. The economics of hiring dozens of engineers to build and maintain complex AI infrastructure made sense, because there was no alternative.

Then November 2024 happened. And everything flipped.

The current generation of AI tools didn’t just get better — they made the cost structure of traditional software companies optional. A single builder with AI leverage can now produce what used to require a team of forty. The tools that Perplexity uses have made Perplexity’s organizational model obsolete.

But they can’t acknowledge that. They can’t shrink to fifty people without destroying the valuation narrative that justifies $1.2 billion in fundraising. They can’t stop the Azure spend without degrading the product. They can’t cut the legal budget with four active lawsuits. They’re sitting at the table, down hundreds of millions, and the odds have changed underneath them — but the cost of standing up and walking away is the admission that the bet was wrong.

At a recent Cerebral Valley AI Summit, 300 AI founders and investors were asked which billion-dollar startup was most likely to fail. Perplexity won that vote. The people closest to this industry can see the math.

I’m not picking on Perplexity specifically. They’re just the most visible example of a pattern: companies that raised and scaled under pre-2025 assumptions, now trapped by the sunk cost of a structure the market no longer demands.

The Base44 Counter-Example

While Perplexity burns through hundreds of millions, consider Base44.

One person. Zero funding. Built entirely with AI tools. Maor Shlomo launched Base44 in late 2024, hit $3.5 million in annual recurring revenue within six months, attracted 300,000 users, and sold to Wix for $80 million cash plus $90 million in milestone payouts.

One person produced a $170 million outcome. No cap table. No board. No HR department. No investor updates.

This isn’t an outlier anymore. Dario Amodei, the CEO of Anthropic — the company whose API Perplexity pays for — has predicted with 70-80% confidence that the first billion-dollar one-person company will emerge this year. Daniel Nadler, whose OpenEvidence reached a $12 billion valuation, said at GTC 2026 that future tech giants will operate with fewer than a hundred employees.

The question isn’t whether the economics of one work. The question is how long the economics of one thousand can pretend they still do.

What’s Left to Invest In?

Let’s break down where the money goes in a traditional software company:

Compute. This is real. LLM inference costs money. But it costs the same money whether you’re a solo operator or a 200-person company. There is no team-size advantage here.

Infrastructure. Also real, but increasingly commoditized. Hosting, CI/CD, monitoring — these are solved problems with per-unit pricing. A solo operator pays less because they use less.

Software development. This is where the argument falls apart. The entire premise of hiring thirty engineers was that software is hard and slow to write. In 2026, with the current generation of AI tools, that premise is observably false.

HR, legal, compliance, management. These are the costs of having a team, not the costs of building a product. When you remove the team, you remove the overhead. A solo operator doesn’t need an HR department because there’s no one to manage. They don’t need a VP of Engineering because there’s no engineering org to vice-preside over.

So when a company raises $1.2 billion to build a product, where is that money actually going? Largely to the organizational structure itself — the very thing that AI has made optional.

These Are Albums, Not Platforms

Here’s what I think these products actually are, at their core: artistic expressions of the people who build them.

A LifeOS is inherently personal. It reflects the builder’s philosophy about how life should be organized, what matters, what can be automated, and what requires human judgment. There is no objectively correct answer to any of these questions.

Claudine reflects my philosophy — that AI should be your COO, not your assistant. That it should do the work, not wait for instructions. That it should nag you about what you’re neglecting. That’s not a universal truth. It’s my conviction, expressed in software.

Perplexity reflects a different philosophy. Other products reflect others. These aren’t competing platforms converging on the same optimal design. They’re divergent expressions of different worldviews.

That’s beautiful. But it’s not what investors are pricing. Investors are pricing platform dynamics — winner-take-most, network effects, defensive moats. The LifeOS category doesn’t have those properties. It has the properties of music: infinite variety, personal taste, low switching costs, and no reason to consolidate around three winners.

You wouldn’t invest $1.2 billion in a record label and expect one album to capture the entire market. But that’s roughly what’s happening here.

The Artificial Barriers

Now — I’m a shareholder in a lot of companies. I’m not naive about how this works. When a structural shift threatens the model that’s been printing money for thirty years, the money doesn’t just shrug and adapt. It builds walls.

Here’s what I expect to see, and what we’re already seeing:

API pricing as gatekeeping. The companies providing the AI can tier their pricing to make solo operators uneconomical while offering volume discounts to enterprise customers. “Oh, you want Opus-level intelligence? That’s the enterprise plan.” This is the most immediate and potent lever.

Compliance as a moat. SOC2, GDPR, HIPAA — these cost roughly the same whether you’re one person or one hundred. Every new regulation is a fixed cost that large companies absorb and small operators struggle with. Incumbents will lobby for more of these. The justification will be consumer safety. The effect will be barrier to entry.

Platform friction. App Store review processes, enterprise developer program requirements, platform-specific compliance hurdles — none of these are malicious. But every platform decision that assumes “real software comes from teams with dedicated developer relations and enterprise accounts” is a tax that falls disproportionately on solo builders. A forty-person company has someone whose job it is to navigate Apple’s review process. A solo builder loses a week.

The narrative machine. This one’s subtle but powerful. Venture capital doesn’t just fund companies — it funds the story that companies are necessary. “Can you really trust your personal AI to a one-person operation?” becomes the FUD. “Enterprise-grade” becomes the euphemism for “backed by enough money that you shouldn’t ask hard questions about efficiency.”

Talent hoarding. Companies will continue hiring engineers they don’t strictly need, in part to prevent those engineers from becoming solo competitors. We’re already seeing overstaffed AI teams where half the people are building internal tools that their own AI products could replace. The irony is thick.

I understand all of this because I sit on both sides. As a shareholder, I benefit from the existing model. As a builder, I can see it’s running on borrowed time. Those two truths coexist.

Why the Barriers Leak

Here’s the thing about artificial barriers: they work until they don’t.

API prices are trending down, not up. Open-source models are closing the gap. Every major lab is competing for developer adoption, which means they can’t price out the small operators without losing the ecosystem that makes their models valuable.

Compliance-as-a-service is a growing industry. What cost $50,000 and a dedicated hire three years ago can now be handled by a platform for a few hundred a month.

Platform restrictions get routed around. The friction is real but it’s not fatal. Builders adapt.

And the narrative? The narrative loses power the moment a solo-built product ships and works. You can’t “enterprise-grade” your way out of a side-by-side comparison where the one-person product is faster, cheaper, and more opinionated in ways users actually prefer.

History is consistent on this: structural economic shifts don’t get stopped by institutional resistance. They get delayed. The printing press didn’t get stopped by the scribes’ guild. Digital music didn’t get stopped by the record labels. The resistance makes the transition messier and slower, but the math eventually wins.

Building in the Gap

So what do you do if you’re building right now, in the window between “the economics have changed” and “everyone has accepted that the economics have changed?”

You build lean and stay lean. Not as a temporary strategy while you fundraise, but as the permanent structure. The goal is not to grow into a traditional company. The goal is to never need to.

You own your infrastructure. Renting compute from hyperscalers means your margins are someone else’s revenue.

You ship your convictions. Rick Rubin put it better than I can: “The audience comes last. I’m not making it for them, I’m making it for me. It turns out that when you truly make something for yourself, you’re doing the best thing you possibly can for the audience.” In a fragmented market, strong opinions are a feature, not a risk. Build the product that reflects what you actually believe about how this should work.

And you keep your costs low enough that you don’t need anyone’s permission to keep going. No runway clock. No board to satisfy. No pivot pressure. Just the work.

The Uncomfortable Truth

The uncomfortable truth for shareholders — and I say this as one — is that the arrangement where capital extracts value from labor has always depended on labor being necessary. Not just valuable. Necessary.

When AI makes it possible for one person to do what required fifty, the necessity argument weakens. Not for all software. Not for all industries. But for categories like this one — personal, opinionated, AI-native products — the honest answer is that a large team is no longer a competitive advantage. It’s overhead.

That doesn’t mean every funded company will fail. Some will build genuine network effects, genuine data moats, genuine distribution advantages. But the default assumption that “more money and more people equals better product” is no longer safe. And the investors who keep making that assumption are going to have a hard decade.

In 2025, 124,000 tech employees were laid off across 271 companies. That wasn’t just cost-cutting. It was a belated recognition that the operating models were overbuilt for the leverage now available. The companies that raised and hired under the old assumptions are now faced with a choice: restructure honestly, or keep feeding the machine and hope the math changes back.

The math isn’t changing back.

I’m not rooting against anyone. I want every one of these products to be great. But I also want to be honest about what I’m seeing from inside the machine: the economics of one aren’t a theory anymore. They’re a working model. And the gap between what it costs to build this way and what it costs to build the old way is widening every month.

If you’re building this way — lean, with AI, against the grain — you already know the math works. Trust it.

Lee Graham is the founder of Graham Alembic, where he builds Claudine, Kindling, and Alembic Compute. He thinks the future of software looks less like enterprise and more like jazz — small, expressive, and impossible to consolidate.