I’ve been thinking about OpenAI less as a tech story and more as a balance sheet story.

And the more I look at it that way, the more uncomfortable it feels.

I don’t think OpenAI survives the next three years in its current form. Not because the models are bad, and not because the team lacks talent…

OpenAI spends an extraordinary amount of money. Training frontier models costs billions. Running inference at global scale costs billions more. They also compete for some of the most expensive technical talent in the world. All of that adds up to a business that burns cash at a pace very few companies in history have sustained without a clear path to profitability.

That model works as long as investors believe you’re building something no one else can replicate. For a while, OpenAI had that aura. GPT-3 and early GPT-4 felt untouchable. Today, they don’t really have an edge.

The large language model layer has commoditized faster than most people expected. Anthropic ships strong models with a more profitable target audience. Google finally woke up and Gemini has seen explosive growth last year.

If a company can switch providers without much friction, then pricing power disappears. And if pricing power disappears, you lose the ability to fund extreme levels of spending.

Some people argue that OpenAI can simply monetize harder. They mention ads or higher subscription pricing. I don’t think that solves the core issue.

Ads require both scale and leverage. Google and Meta built empires around distribution and intent. OpenAI has usage, but usage alone doesn’t guarantee an ad machine that can support a multi-billion-dollar cost structure. If they push ads aggressively, they risk damaging trust and user experience. If they move cautiously, the revenue uplift likely won’t offset the burn.

An IPO might look like a solution. It isn’t.

Private investors tolerate massive losses if they believe in eventual dominance. Public markets operate differently. Once quarterly earnings matter, the story shifts from “potential” to “discipline.” Public shareholders won’t accept endless dilution and escalating compute costs without a visible path to margin stability.

So what happens if funding tightens and profitability remains distant?

I see Microsoft stepping in.

Microsoft already integrates OpenAI’s models deeply across Azure, Office, GitHub, and enterprise tooling. Plus, they already owns 27% of OpenAI.

Why not Anthropic collapsing instead? Anthropic positions itself more tightly around enterprise and coding use cases. Their commercial story looks narrower and more disciplined. That focus matters.

OpenAI sits in a strange middle ground. It operates at enormous scale, but it doesn’t clearly dominate a single defensible vertical. It spends like a frontier research lab and markets like a consumer platform, yet it competes in a space where the core technology keeps getting cheaper.

That combination creates risk.