Claude Stops Choking Under the Hood

Anyone using Claude Code or Claude Opus via API has been hitting ceilings for several months. Especially during European afternoon hours — when American and European demand overlap — limited capacity manifested as slowdowns or interruptions in work. Anthropic now comes with a solution that is not just cosmetic: three changes effective today are expected, according to the official announcement, to significantly improve the user experience with Claude.

The first change concerns Claude Code, a tool that allows AI to work directly in the terminal and modify code in real-time. For Pro, Max, Team, and Enterprise plans, the company doubles the five-hour limits. The second change abolishes peak-hour limit reductions for Pro and Max users — they will no longer be penalized for working when the network is busiest. The third change means a significant increase in API rate limits for Claude Opus models, which will be especially appreciated by companies and developers integrating Claude into their own applications.

SpaceX to Provide Entire Data Center

The source of the new capacity itself is attracting the most attention. Anthropic signed an agreement with SpaceX to utilize all computing capacity in the Colossus 1 data center. Specifically, this involves over 300 megawatts of new power and more than 220,000 NVIDIA GPUs, which will be available by May 2026.

The Colossus 1 data center, operated by SpaceX in Texas, is one of the largest specialized AI infrastructures in the world. For comparison: a typical large cloud provider data center usually has a capacity in the tens of megawatts. Colossus 1 thus represents a leap of an entire order of magnitude higher and gives Anthropic immediate access to hardware that the company itself would have waited years for.

Anthropic also stated that, as part of the agreement, it expressed interest in collaborating on the development of orbital computing capacity in the gigawatt range — i.e., AI supercomputers operated from satellite platforms. This sounds futuristic, but in the context of SpaceX and their ambitions in space infrastructure, it is a logical step.

How Much Is Anthropic Actually Building?

The partnership with SpaceX is not an isolated event. Over the past year, Anthropic has announced several massive agreements that together represent the largest expansion of computing capacity in the history of the AI industry:

• Amazon: up to 5 GW total capacity, with almost 1 GW by the end of 2026;
• Google and Broadcom: 5 GW, capacity to start increasing from 2027;
• Microsoft and NVIDIA: 30 billion dollars in Azure capacity;
• Fluidstack: 50 billion dollars in investments in US AI infrastructure.

Together, this amounts to tens of gigawatts and tens of billions of dollars. For context: the entire Czech Republic has an installed electrical capacity of approximately 11 GW. Anthropic therefore plans to operate infrastructure that would itself consume a fraction of the power of a Central European country. This explains why companies like Anthropic, OpenAI, or Google are investing billions in their own power plants and why AI is becoming a geopolitical topic.

Europe on the Radar, but USA Has Priority

Anthropic emphasized in the announcement that part of the new capacity will be located internationally — specifically in Asia and Europe as part of its collaboration with Amazon. The reason for this is regulatory requirements for data localization, which are particularly imposed by the financial sector, healthcare, and government institutions. For Czech companies subject to GDPR and the upcoming EU AI Act, this is good news: inference in the European region reduces legal risks and speeds up response times.

However, the company also stated that it selects partners "very deliberately" — prioritizing democratic countries with legal frameworks that support investments of this size and secure supply chains. In practice, this may mean that while Czech companies will see better availability, the primary capacity will remain in the USA.

When Will the Next Improvement Come?

Anthropic trains and operates Claude on various types of hardware — AWS Trainium, Google TPU, and NVIDIA GPU — and, by its own admission, "never stops looking for ways to connect additional capacity." With the arrival of almost 1 GW from Amazon by the end of 2026, further increases in limits or reductions in latency can be expected.

For Czech readers and companies, it is crucial to monitor how quickly these changes will manifest in practice. While the hardware will be physically connected during May, optimizing software limits and geographical routing may take weeks to months. However, the first signal comes today: limits are increased immediately, not only after new servers are installed.