The revenue figure is real. The $10.9 billion Q2 projection, reported by The Wall Street Journal and confirmed by CNBC, puts Anthropic’s quarterly revenue at roughly 2.3 times its Q1 2026 figure of $4.8 billion. That’s a 127% quarter-over-quarter jump. The company also reportedly projects operating profit of approximately $559 million for the quarter, but that figure excludes stock-based compensation and should be read as a non-GAAP metric, according to investor guidance reported by the Journal. Anthropic has warned investors it may not hold profitability through year-end as next-generation model training capital expenditures ramp up.
The operating number matters less than the architecture behind it.
According to reports citing SpaceX’s IPO filing, Anthropic is paying SpaceX $1.25 billion per month for 300 megawatts of compute capacity, over 220,000 Nvidia GPUs, at the Colossus data center, through May 2029. DataCenterDynamics and Business Insider both reported the contract terms from the filing. The total implied value over the 36-month term: approximately $45 billion. Anthropic is spending more on a single compute vendor every month than most AI startups raise in a year.
Two funding events. One company. Keep them separate.
Bloomberg and The Wall Street Journal have both reported that Anthropic is seeking a new funding round of at least $30 billion at a valuation exceeding $900 billion. That round hasn’t closed. The round that did close, the Series G, confirmed on Anthropic’s own website, raised $30 billion at a $380 billion post-money valuation. These are different events. The $900 billion figure is a target for an ongoing negotiation, not a closed deal. Conflating them overstates Anthropic’s current enterprise value by more than 2.3x.
The compute concentration is the story enterprise buyers should be tracking.
Anthropic runs its inference on a compute stack it doesn’t own, from a vendor with a pending IPO and a CEO whose other company competes in the AI market. That’s not a criticism of the arrangement, the scale and the revenue numbers suggest it’s working. But it means Anthropic’s near-profitability is structurally dependent on a contract with a single counterparty through May 2029. If that relationship changes, in pricing, availability, or governance, the margin math changes with it.
The implied revenue-per-compute-dollar ratio is striking. At $1.25 billion per month in compute spend ($3.75 billion per quarter), $10.9 billion in quarterly revenue implies roughly $2.90 in revenue for every dollar of compute cost at Colossus. That ratio will compress as next-generation training runs scale. Whether it compresses to zero, or stabilizes at a sustainable level, is the central question in Anthropic’s 2026 financial story.
Watch the second half of 2026. Anthropic’s own investor communications signal that CapEx for next-generation model training will pressure margins. The Q2 profitability projection could be the high-water mark before a deliberate investment cycle. The more durable signal is the revenue trajectory, not the single-quarter profit figure.