Anthropic CEO Dario Amodei told staff and analysts this week that the company’s revenue and usage grew roughly 80-fold in the first quarter on an annualised basis, lifting Anthropic to a $30 billion annualised run rate, up from $87 million in January 2024. Amodei said internal plans had assumed 10-fold growth, and described the actual pace as “just crazy” and “too hard to handle”.
The bulk of the surge has come from Claude Code, the agentic coding product launched in mid-2025, which hit $1 billion in annualised revenue inside six months and is now the fastest-growing product in the company’s history. API volume on the Anthropic platform is up 17x year-on-year, with developers facing repeated rate-limit headaches as supply struggles to keep up.
To plug the compute gap, Anthropic announced a deal to take all the capacity at Elon Musk’s Colossus 1 data centre in Memphis, Tennessee, more than 300 megawatts and over 220,000 Nvidia GPUs coming online within the month. Two days later, on 8 May, Anthropic signed a separate seven-year, $1.8 billion compute deal with Akamai, the largest contract in Akamai’s history. Akamai stock closed up 27% on the news, its biggest single-day rally in more than 22 years. The deal stacks on top of an April $200B Google Cloud expansion, plus existing Amazon, Microsoft and Nvidia commitments.
This is the week Anthropic stopped being the smaller frontier lab. A $30B run rate is one third of OpenAI’s reported $25B run rate this quarter (yes, Anthropic is now ahead on annualised revenue), with about a third of OpenAI’s headcount and almost no consumer reach. The whole story sits on Claude Code and the API platform. Enterprises are voting with developer credit cards.
The compute frenzy is the other half of the story. SpaceX Colossus 1, a $1.8B Akamai deal, an April $200B Google Cloud expansion, plus existing AWS, Microsoft and Nvidia deals all in a single week. Anthropic is now stitching together capacity from every available source because there is no single hyperscaler that can satisfy 80x growth alone. Akamai stock rallying 27% on the news shows the market is starting to price the second-tier “AI cloud” buildouts the same way it priced the hyperscaler trade in 2023.
The bottom line: watch whether the SpaceX and Akamai capacity actually clears the rate-limit queue before Q3, what Q2 revenue looks like once the new compute lights up, and whether the Pentagon thaws on Anthropic now that exclusion is starting to look like a strategic mistake.
The frontier labs are running into the limits of their own success. Anthropic’s 80x quarter pulled forward a SpaceX Colossus deal, a $1.8B Akamai contract and a doubling of Pro and Max rate limits, all because demand has overwhelmed supply. OpenAI’s $25B run rate pulled forward an IPO conversation, with the CFO publicly arguing for delay because $600B of infrastructure commitments need to land first. Sierra’s $950M raise on top of $150M ARR shows the same pattern in the application layer.
The story is no longer “can these companies grow?” It is “can they finance and physically build the compute, talent and operating discipline to absorb the growth they already have?” And just under that, Subquadratic emerged from stealth claiming a 1,000x efficiency gain at 12M tokens. If even a fraction of that holds up, the next year of capex assumptions will need a serious rewrite.
Subquadratic has published claims and a private beta, but no peer-reviewed evaluations and no public weights. Researchers across the field were quick to point that out, and Artificial Analysis is reportedly already lining up benchmarks. The company is launching three products into private beta, an API, a coding agent called SubQ Code and a search tool called SubQ Search, with broader access promised once partner benchmarks land.
Why this matters: frontier labs have been chasing subquadratic architectures for years, and if SubQ holds up it would be the first to ship at frontier capability. Watch for independent benchmark prints in the next two to three weeks. If even half of the 1,000x at 12M tokens claim is real, every 2026 capex assumption (including the $700B hyperscaler trade) will need a quiet recheck.
Read VentureBeat coverage →