OpenAI is below, above and around you

What is the best way to build a $100 Bn company in 2025? Build an AI application company.

Or is it?

I will argue that contrary to popular belief AI application economy will largely be dominated by model companies (OpenAI, Anthropic). Why?

Well:

• OpenAI is cheap

• OpenAI needs money

• OpenAI can

People are concerned. AI bubble. Overvalued. OpenAI raising at $500 billion while burning $30 Bn a year on just $13 Bn in revenue.

It sounds like a bad investment. Very expensive. Bloated capex.

It isn’t.

Check the below table out (link here)

Look at the math.

1. OpenAI’s run rate hits $13 Bn in 2025. Their recent secondary sale pegged them at $500 Bn valuation. That’s a 38.5x revenue multiple on 2025 numbers.

2. They are growing fast with 282% growth in revenue this year. If we assume they will grow an average of the last 2 years, it means they would grow at 197% in 2026, resulting into a $38 Bn forward looking revenue number.

3. 2026 revenues of $38 Bn with a $500 Bn valuation points to a 12.9x forward looking EV/2026E revenue.

Oracle trades at 11-13x forward revenue. So does Microsoft. And neither of them has a serious model lab. Neither of them is building the infrastructure, or better said the index of the AI economy.

OpenAI is cheaper than you think.

Sure, Oracle and Microsoft are profitable. OpenAI bleeds money and will keep bleeding money as they build data centers and buy chips. That’s the one advantage the old guard has.

Here is the catch though:

1. Microsoft owns 49% of OpenAI

2. Oracle has a data center deal worth $300 Bn with OpenAI

3. OpenAI has another deal with Nvidia itself + AMD

Two conclusions jump out:

1. OpenAI is now too big to fail - the single point of failure for the capital markets that came to be indexed almost solely on AI.

2. The AI index ETF is simply OpenAI itself. Not a bucket of companies, only OpenAI.

All good so far. However all of the big tech is largely profitable companies. OpenAI is not. With $13 Bn in revenues this year, it is on track to lose $30 Bn in 2025. Doesn’t this make OpenAI a bad investment? No, because they are simply the gravity of S&P 100 now with 40% of the companies in index having direct exposure to AI in one way or another.

OpenAI simply needs raise more and more money not to fail and in order to grow. To justify that, OpenAI needs to make more money.

Which brings me to my next point:

Supplying models as a neo-cloud API business is good.

You know what is even better: A chat inteface that does everything (ChatGPT).

And an agent builder. And advertising. And an enterprise ChatGPT. And a personal device. And a social media product (Sora). And a web browser. And a music generation app.

OpenAI started chasing superintelligence. Now they’re chasing cash. Maybe they’ll use that cash to get back to superintelligence later. Maybe not. I know I would forget

Think about the last time a model truly mesmerized you. For most developers, it was GPT-4. Or Claude Sonnet 3.5.

That was it. No breakthrough since then.

Maybe we’re still waiting for the next leap. Maybe it’s coming. But I doubt it. Here’s why: GPT-4 and Sonnet 3.5 came from a world where labs were fighting to prove who was better. That fight ended. Claude won on coding. The new fight is about who makes the most money.

And there is no better way to make money than selling applications to the whole damn world.

Look at the last 30 years. Second-comers usually win. Stripe came after Braintree. Facebook came after Friendster. Google came after Yahoo.

Model labs are positioned to do the same in the application layer. Why? Because startups already launched, scaled, and mapped the territory. OpenAI now has a guide. They can see which verticals are large, which workflows matter, which features users want.

The startups using OpenAI’s APIs just taught OpenAI how to compete with them.

This shift is obvious. Look at the product launches. OpenAI keeps shipping products that have nothing to do with improving their models. Anthropic does the same. Claude now has memory, PDF handling, Word, Excel, and PowerPoint creation and more. Google Gemini just released a vibecoding tool, image generation, video products, audio generation and more.

Scary stuff if you are looking to build the next big startup on top of these models.

However I can already hear you saying: “I will build into workflows, embed my product, build a deeper integration and a better vertical product …”

Which bring me to the last point:

Prevailing opinions on the market revolve around:

1. OpenAI team is spreading itself too thin launching too much

2. They cannot go deep

3. They don’t build good products in terms of UI/UX and workflows

These opinions reflect how you would rather see the world rather than how the world actually is.

Here is what I see:

1. The highest amount of enterprise demand a software company has even seen to the extent that they cannot serve it in full

   1. OpenAI has an internal inbound demand agent that vets like 1,000 enterprises per week and routes them to appropriate places including just discarding them and simply sending docs links

2. If they were to serve it, a brand name so powerful that no kind of “focused vertical startup” could beat them in a bake-off

   1. Think about it: If you wanna buy a legaltech solution, and Harvey and OpenAI are pitching you - who are you going with? I’d bet $50 OpenAI wins 9 times out of 10.

3. Great product taste in terms of simplicity and UX

   1. All products they launch are extremely simple (except the agent builder), fully functional and coded by ex-Stripe, Twilio etc engineers - not researchers

All this is to say: OpenAI and Anthropic are moving up the stack. This move is not temporary, it is secular.

When they build products, they make them so capable and generic that 80-90% of use cases get solved immediately.

As developers, we know that the real fight is in the last 10%. However if you are trying to find PMF, you need a way to be radically different than ChatGPT and all the other products OpenAI is launching.

If a prospects asks “How is this different than ChatGPT?”, it’s already too late.

This might have been too much doom and gloom because I kinda started feeling sick around B2B SaaS for some reason. I will touch upon what can be done around this, how to be different or if you should build an AI company at all in a future post.