On November 2, 2026, The New York Times reported that Anthropic had faced a computing-power shortfall late in 2025 and later pursued dozens of contracts and other ways to secure capacity. The company needs computing resources both to develop advanced AI models and to deliver Claude to business and consumer users.
Anthropic pursued contracts and other capacity measures
The Times described contracts involving former cryptocurrency miners, data-center startups, chipmakers and major technology companies. It named Amazon, Google, Microsoft and SpaceX among the counterparties. The report also described equipment purchases, data-center hires and meetings with developers. Anthropic had explored converting former coal plants and factories, and considered smaller capacity deals.
A related financing angle appears in NeoTeo’s earlier report on Nvidia’s reported guarantee for Nscale.
What the Colossus 1 agreement covers
In a report published May 6, 2026, Constellation Research said Anthropic had agreed to use all computing capacity at SpaceX’s Colossus 1 data center. Anthropic described the agreement as covering more than 300 megawatts of new capacity and over 220,000 NVIDIA GPUs. The megawatt figure describes power capacity; the GPU figure counts accelerators at that facility.
Constellation reported that Anthropic expected access within May 2026. It also reported changes to Claude usage limits alongside the agreement: five-hour Claude Code limits doubled for Pro, Max, Team and seat-based Enterprise plans; peak-hour reductions were removed for Claude Code Pro and Max; and rate limits for the Claude Opus API were raised.
Earlier reports put the figures in context
An English-language CNBC segment published November 18, 2025, discussed a reported $30 billion in compute through NVIDIA and Microsoft. That reported dollar amount is a separate kind of figure from Colossus 1’s power capacity and GPU count.
An English-language Forbes video published October 30, 2026, described Anthropic’s IPO filing as setting out a plan to spend $1 trillion on chips and data centers over the next two decades. Forbes framed the figure as planned spending over that period.