Anthropic has announced second-quarter 2026 revenue of around $10.9 billion, up roughly 130%, and a first operating profit of around $559 million. The company reaches profitability two years ahead of its own schedule. At the same time, it has committed to a twenty-year data centre lease covering 191 megawatts, for around $9.1 billion.
We have been tracking the question of AI profitability for months, from financing structures to Microsoft's results. Here is the first frontier lab to announce an operating profit. That matters, and it deserves careful reading.
Why this figure matters
Until now, the sceptics' argument was solid: these companies generate substantial revenue and lose even more money. An operating profit, even modest relative to revenue, changes the nature of the conversation.
It shows that a frontier lab can cover its operating costs with its revenue. That is no small thing, and it validates the strategy we described in our comparison between Anthropic and OpenAI: targeting the enterprise segment, where customers pay, rather than the consumer volume that is hard to monetise.
The caveats, which are real
Three precautions are in order before jumping to conclusions.
One quarter is not a trajectory. Analysts note that an exceptional ramp-up at a major client may have boosted this quarter, and pressure is expected next quarter. A first profit can be followed by a return to losses.
Operating profit is not net profit. This metric measures what the core business generates, before certain financial items. It is an important milestone, but it is not the same as making money in the full sense.
These figures are self-reported. They have not been subject to an audited regulatory filing. That is precisely what will make OpenAI's IPO prospectus so instructive: it will be the first document of this kind subject to legal liability.
The same move comes with a substantial infrastructure commitment: a data centre lease running until 2048, covering 191 megawatts, for around $9.1 billion in baseline revenue, with expansion options that could push the total well beyond. In other words: the company posts a quarterly profit in the hundreds of millions and commits to billions over twenty years. That is not contradictory; it is the very model of this industry, which we described in our article on the compute race: locking in access to electricity before competitors do.
What it reveals about the business model
The gap between these two orders of magnitude is the real issue. A profit is counted in hundreds of millions, commitments in tens of billions.
That means the announced profitability rests on an assumption of continued growth. A twenty-year lease is only sustainable if revenue keeps growing at a sustained pace for the entire duration. It is a bet on demand, not just on technology.
That bet may be reasonable: demand for compute is indeed rising, and securing electricity today costs less than hunting for it in five years, as the current tension on components shows. It remains a bet.
What to take away
This announcement shifts the debate without closing it. It becomes harder to argue that frontier AI cannot be profitable: the demonstration exists, at least for one quarter.
It remains just as hard to argue that the question is settled, because the observed profitability is nowhere near the commitments made. What will decide things is not this quarter, but the ability to sustain that growth over the duration of the leases signed. The coming quarters, and the prospectus expected at the competitor, will tell us far more than this first figure.