OpenAI Is Not Avoiding the Public Markets. It Is Buying Time.

This one is a little more speculative than the trading system work I usually send out. Not because I think it is random -- I do not. I actually think it may be one of the more important tells in AI right now.

The headline version is simple: SpaceX went public and raised $85 billion. OpenAI raised $122 billion and stayed private.

Most people will look at that and say: of course OpenAI needs money. Frontier AI is expensive. Training is expensive. Inference is expensive. Data centers are expensive. Power is expensive. Chips are expensive.

All true.

But I do not think that is the real story. The real story is not that OpenAI needs capital. The real story is what OpenAI can do because it does not need the public markets to get it.

If you can raise SpaceX-scale capital privately, then staying private is no longer a funding constraint. It becomes a strategic choice. And I think OpenAI’s choice tells us something important about where the AI economy is going.

Because OpenAI is not just trying to build a better chatbot. It is trying to control the intelligence supply chain: models, chips, compute, data centers, government-approved release paths, enterprise access, developer distribution, and now, potentially, the earliest stage of company formation itself.

That is why this move I'm about to dig into matters so much.

The YC investment

Sam Altman was president of Y Combinator before OpenAI. That is not a random biographical detail.

YC is arguably the most important startup accelerator ever created. Stripe, Airbnb, Coinbase, DoorDash, Scale AI, Reddit, Dropbox, Instacart, Kalshi, Replit -- the list is ridiculous. YC has been the ground zero of software company formation for almost two decades.

And now OpenAI is offering to invest $2 million in tokens into every company in the YC spring and summer batches...169 companies!

That is the part I do not think people are taking seriously enough.

A normal AI startup gets API access. Maybe some credits. Maybe a partnership. Maybe a model provider retweets the launch.

But an OpenAI-backed YC company could get something very different: early access, better inference economics, priority compute, internal eval tooling, support from the model ecosystem, enterprise credibility, a cleaner path through safety and compliance expectations, and eventually, maybe, an obvious acquisition path.

That is not venture capital in the normal sense.

That is model-privileged venture capital.

And in an AI economy, that may be one of the most powerful forms of capital that exists. If the most important input into the next generation of companies is not just money, but access to frontier intelligence, then the company controlling that access has a very different kind of leverage.

OpenAI does not need to build every vertical AI company inside the parent. It can fund the teams at the frontier of company formation: legal research, security operations, drug discovery, scientific research, education, government workflow, enterprise agents, finance operations, software engineering, robotics, defense, space.

OpenAI can seed teams in all of those markets, give them privileged access to the model layer, and own part of the upside without becoming the operator of every vertical business itself.

That is a much cleaner strategy.

The public story can remain: we are supporting developers. The strategic reality may be: we are deciding which companies get built closest to the frontier.

OpenAI is not just competing for users. It may be competing for the company formation layer.

And I think what they are really getting is a recursive loop where the best early stage companies doing the most innovative things at ground zero for everything (Dog Patch in San Francisco - you don't have to like it) using their models for their work...and this is where the recursive part comes in.

These companies are then training OpenAI's models.

Arguably the most innovative, intelligent and driven folks in the world and OpenAI gets to train on them, and not a bunch of meatheads like me.

Staying private gives OpenAI room

The other piece is the regulatory layer.

OpenAI is not operating in a normal software release cycle anymore. A normal software company builds a product, ships it, measures usage, fixes bugs, and improves the next version.

Frontier AI is different.

Some capabilities may be too sensitive to release immediately. Some models may need government testing. Some deployments may be acceptable for enterprises but not consumers. Some may be acceptable for trusted government users but not the general public.

Claude Mythos and Fable being restricted is the example that makes this feel very real.

The public model is no longer necessarily the best model. It may just be the model approved for broad release. That changes how you should think about the entire sector.

If OpenAI were public, every model delay would become a market event. Every time a release was held back, analysts would ask about it on the earnings call. Every time a competitor shipped something, shareholders would demand a response. Every unreleased capability would create pressure to monetize.

That is a terrible structure if your most important products may need to be staged, gated, restricted, or selectively deployed.

Staying private gives OpenAI room to do something public markets hate: wait, hold back, segment access, work quietly with governments, and give different levels of capability to different users.

One level for consumers. Another for enterprises. Another for strategic partners. Another for government customers. Another for internal or OpenAI-backed companies.

Public markets want clean release cycles. Frontier AI probably will not have clean release cycles.

That is one reason staying private matters. It gives OpenAI insulation from the quarterly pressure to show the whole hand. And if the market can only price what it can see, OpenAI may be building around what it does not have to show yet.

Jalapeño time

Then there is Jalapeño, OpenAI’s chip.

This matters for a simple reason: a company that rents GPUs is dependent on someone else’s cost curve.

NVIDIA matters. Cloud providers matter. Data centers matter. Power contracts matter. Memory matters. Networking matters. Cooling matters.

If OpenAI is building its own chip, it is not just optimizing margins. It is trying to control the cost of serving intelligence.

That is the whole game.

Training the best model is one layer. Serving that model to hundreds of millions of users, enterprises, developers, agents, governments, and automated workflows is another business entirely.

If inference becomes the dominant cost of the AI economy, then the company with the best inference cost curve gets more than margin expansion. It gets strategic freedom.

It can price differently. It can subsidize products. It can give preferred access to partners. It can fund companies with compute, not just cash. It can decide which workloads are worth running. It can decide which startups get cheap access to frontier intelligence.

That is why Jalapeño matters.

It pushes OpenAI down the stack: models → chips → data centers → power → cooling → networking → distribution → applications → company formation.

That is not a normal software company. That is an AI factory company.

And if you combine that with the YC strategy, the picture gets much more interesting.

OpenAI is not just building the model. It may be building the factory that produces the companies that use the model.

The private market becomes the launchpad

This is where I think the public/private market distinction matters.

If OpenAI were public, the story would be filtered through quarterly revenue, gross margin, capex, guidance, product release timing, and analyst questions.

But staying private lets OpenAI operate more like a sovereign AI industrial company. It can raise enormous capital, build custom chips, secure compute, negotiate power, work with government, control model release timing, fund YC companies, and push capability into the application layer through the startups closest to the frontier.

That is a very different strategy than “sell subscriptions to ChatGPT.”

And it may be the only strategy that makes sense if you believe frontier intelligence is going to become a base input into the economy.

The question is not just who has the best model. The question is who controls the bottlenecks around the model: compute, power, chips, permission, distribution, developer access, and company formation.

OpenAI is trying to move into all of them.

Run it through the bottleneck filter

The filter I keep using is simple:

AI is a bottleneck discovery machine.

If AI capability doubles again (like it has been doing every few months), what breaks first?

At first the bottleneck was model quality. Then GPUs. Then HBM and advanced packaging. Then data centers. Then power. Then cooling. Then networking. Then permission, identity, and agent control.

Now another bottleneck is showing up: access to frontier intelligence.

Who gets it? When do they get it? At what price? Under what rules? With what compute priority? With what regulatory clearance?

That is why the YC move matters.

OpenAI is not just buying exposure to startups. It is potentially shaping who gets to build with the best intelligence first.

And if you are an application company competing against an OpenAI-backed company with better access, cheaper compute, stronger distribution, and patient capital, that is not a small disadvantage.

That is existential.

What this means for public markets

I do not think this is just an OpenAI story. It is an AI infrastructure story.

If OpenAI can raise $122 billion privately, build its own chip, stay outside public market pressure, and seed 169 YC companies at the company-formation layer, then the physical bottlenecks do not get less important.

They get more important.

More private capital means more data centers. More custom chips means more advanced packaging, memory, substrates, networking, and manufacturing complexity. More model access means more inference demand. More spinouts mean more compute consumption. More government gating means more value in companies that can operate inside approved channels.

That keeps pulling forward the boring parts of the stack: power, grid equipment, switchgear, cooling, data-center systems, memory, advanced packaging, and networking.

That is why I keep coming back to names like Vertiv, Eaton, GE Vernova, Quanta, Constellation Energy, NVIDIA, Broadcom, TSMC, Micron, and Arista.

Some are already obvious. Some are expensive. Some need more work. But the direction is clear: if frontier labs keep raising capital, building chips, signing power deals, and pushing into the startup layer, the physical infrastructure stack keeps getting pulled forward.

The weaker part of the market is the thin AI application layer.

If a company is just renting public model access and wrapping it in a workflow, I would be careful. Because if OpenAI can fund a competitor with better model access, cheaper compute, enterprise credibility, and patient capital, that wrapper may not be a business.

It may be a feature waiting to get competed away.

The test for any AI application company is simple:

Does it own distribution? Does it own proprietary data? Does it own a workflow customers cannot easily leave? Does it have regulatory permission? Does it own a physical or operational bottleneck? Does it have access the next startup cannot easily buy?

If not, it may just be standing in front of the model layer. That is a dangerous place to stand.

What would make this wrong

This can be wrong.

If open-source models catch up enough, privileged frontier access matters less. If regulators force equal access or model neutrality, the strategy gets harder. If compute supply catches up faster than demand, the physical infrastructure trade weakens. If enterprises and governments reject dependence on OpenAI, the stack-control strategy loses power. If the YC token investment is more marketing than real access, the company-formation thesis is weaker.

Those are real risks.

But right now, I think the direction is clear enough.

OpenAI is not avoiding the public markets because it cannot raise money. It is avoiding the public markets because the public markets may be the wrong structure for what it is building.

It wants capital without quarterly pressure. It wants chips without vendor dependence. It wants model release flexibility without every delay becoming a market event. It wants the ability to work with government quietly. And maybe most importantly, it wants to be at ground zero of the next generation of AI-native companies.

That is the real play.

Not just ChatGPT. Not just GPUs. Not just another model release.

OpenAI is trying to control more of the stack before the public ever sees the full map: capital, compute, chips, power, permission, distribution, and startup formation.

That is why this matters.

If this is right, the AI trade is still not just a chip trade. It is a bottleneck trade.

And OpenAI just showed us which bottlenecks it wants to own.