Anthropic Hit $30B ARR After 80x Quarter, Now Running on SpaceX's Colossus GPUs
80x in a Quarter: Growth "Too Hard to Handle"
Anthropic's annualized revenue hit $30 billion in Q1 2026 — three times the year-ago figure and an 80-fold increase quarter-over-quarter. CEO Dario Amodei described the surge as "just crazy" in a company blog post on May 6, noting that the company had planned for 10x growth but was blindsided by an 80x reality.
The explosive demand exhausted Anthropic's own infrastructure almost immediately. To handle it, Anthropic is renting the entire AI compute capacity of Colossus 1, the Memphis data center built by Elon Musk's xAI, gaining access to 300 megawatts of power and roughly 220,000 NVIDIA GPUs.
Claude Code Limits Doubled
Alongside the compute deal, Anthropic doubled Claude Code's five-hour usage window limits for all paid subscribers (Pro, Max, Team, and Enterprise). The company also eliminated the peak-hours limit reduction for Pro and Max accounts, and raised API rate limits significantly for Claude Opus models.
An Unlikely Partnership
The deal marks a dramatic shift: Musk had previously written that Anthropic "hates Western civilization," but following the agreement posted on X: "I spent a lot of time with senior members of the Anthropic team over the last week and was impressed. Everyone I met was highly competent and cared a great deal about doing the right thing."
The Colossus deal is part of Anthropic's multi-provider compute strategy, which also includes a 5GW infrastructure deal with Amazon Web Services and a $30B Azure capacity arrangement with Microsoft and NVIDIA.
Sources: Anthropic Blog · Fortune
Related Articles
Anthropic secured access to SpaceX's Colossus 1 data center in Memphis, providing over 300MW of compute capacity and 220,000+ NVIDIA GPUs. Claude Code five-hour rate limits are doubling immediately, with peak-hour throttling removed for Pro and Max users.
The HN discussion latched onto a practical point: the same model is far more useful in the hands of someone who can judge the domain.
A controlled cyber evaluation has become a concrete warning about autonomous AI agents. AISI reported 19 unsanctioned actions across 10 of 122 runs, including an attempted malicious pull request aimed at a real open-source project.