AI Pulse · Frontier Labs Watch

Anthropic just posted its first profitable quarter. Revenue jumped to $11.5B in Q2 2026, up from $4.8B in Q1, with adjusted operating income around $559M. That sounds like a company that’s cracked the code on AI economics. It’s not quite that simple.

Critics looking at the number point out a few things worth knowing before you repeat this stat in a meeting. The profit shows up while Anthropic is on a discounted compute ramp rate, a temporary deal that goes away once its full $1.25B a month contract kicks in at full price. The figure also leaves out stock compensation and the amortization of training compute, two costs that are very real and very large for a company building frontier models. Anthropic itself has told investors not to expect this profitability to hold up through the rest of the year, according to the original report.

Here’s why this matters if you run a business and not a headline. “Profitable” is doing a lot of work in that sentence, and the definition changes depending on which costs get counted and which quarter you catch the company in. The same trick applies to plenty of vendor pitches you’ll see this year, AI tools claiming ROI numbers that hold up for one billing cycle under one set of assumptions. When you’re evaluating any AI spend, ask what’s excluded from the number, not just what’s included.

This kind of scrutiny, tracking real cost against real output over time rather than trusting a single good quarter, is the kind of thing we build into an AI command centre so you’re working from your own numbers, not someone else’s press release. Anthropic’s story is still developing. Treat the “profitable” label as provisional until a full quarter at full price tells you more.