The two most prominent AI companies in the world are running out of room to lose money [1].

Anthropic and OpenAI have both signaled through filings, investor communications, and internal planning documents that 2026 is the year the math must start working. Both companies are spending billions annually on compute infrastructure, talent, and model training. Neither has yet demonstrated that revenue can consistently outpace costs.

OpenAI's annualized revenue reportedly crossed $5 billion late last year, driven by ChatGPT subscriptions and enterprise licensing. But its compute costs alone are estimated to exceed $4 billion, and that figure does not include research staff, which numbers in the thousands. Anthropic's revenue is smaller but growing fast, with Claude enterprise contracts providing a more predictable income stream than consumer subscriptions.

The problem is not revenue. Both companies are generating substantial income. The problem is that the cost of staying competitive requires ever-larger training runs, ever-more GPUs, and ever-higher salaries for a finite pool of researchers. Each new model generation costs more than the last. The revenue curve must not only rise — it must rise faster than the cost curve, which it has not yet done at either company.

Investors remain willing to fund the gap, for now. But the patience of capital markets is not infinite. If 2026 ends without a clear path to profitability, the next round of fundraising will look very different. The question is no longer whether AI is transformative. The question is whether the companies building it can survive long enough to profit from it.

-- THEO KAPLAN, San Francisco