AI startup Anthropic has privately informed investors it expects to achieve its first ever profitable quarter in Q2 2026, with revenue more than doubling to approximately $10.9 billion. The operating profit is projected at $559 million, yielding a thin margin of roughly 5.1% — modest by software standards, but a landmark for a frontier AI lab that previously guided it would not reach full‑year profitability before 2028.
The growth has been fueled by surging enterprise adoption of its Claude chatbot and related tools. Products like Claude Code for developers and the Mythos model, used to find security flaws in production software, now generate far stickier recurring income than consumer chatbot subscriptions. This is a key reason Anthropic can show a profit while rival OpenAI — despite soaring sales — remains in the red.
However, the profitability milestone comes with a massive caveat. SpaceX’s S‑1 filing, released amid its own IPO process, revealed that Anthropic has contracted to pay $1.25 billion per month for compute access through May 2029. The deal covers both of SpaceX’s AI training clusters, Colossus and Colossus II, granting Anthropic access to more than 220,000 GPUs and over 300 megawatts of power. At full utilization, this translates to about $5,680 per GPU per month — well above typical public cloud spot rates for comparable Nvidia H100 machines, reflecting not just raw compute but guaranteed capacity, specialized infrastructure, and power access.
The quarterly compute bill to SpaceX alone will consume roughly 34% of Anthropic’s top‑line revenue, before accounting for any other infrastructure spending. Reports indicate the company plans to ramp up compute and training costs in the second half of 2026, which could easily push margins back into negative territory and erase the full‑year profit.
Anthropic’s disclosure arrives on the same day news broke that OpenAI is preparing for its own public listing, highlighting a diverging strategic landscape. While OpenAI seeks public‑market funding, Anthropic is proving — at least for one quarter — that a safety‑focused, enterprise‑driven AI lab can generate an operating profit. Yet the 90‑day termination clause in the SpaceX contract underscores how rapidly pricing and infrastructure needs are shifting, and the durability of this profitability remains uncertain.
