📖 Reading time: approx. 5 minutes · 952 words · 6,124 characters
AI companies like to talk about models, benchmarks, and security concepts. In the end, however, the question comes back to something rather earthly: Who pays for the data centers? According to Reuters, citing The Information, Anthropic plans to lease and operate its own data center capacity. At the same time, the company is reportedly seeking financial backing from Google for the lease payments. This is notable because Anthropic would then not only be a customer of cloud capacity, but would move more strongly into direct infrastructure control. Anyone who leases and operates their own server environments assumes more responsibility, but can also reduce costs and dependencies in the long term. In AI in 2026, that is no longer a side issue. Compute is strategy.
Symbol image
Reuters reported on June 11, 2026, that Anthropic wanted to lease and operate its own data centers. The Information therefore reported more than a dozen preliminary agreements for US data centers with a combined capacity of more than 1 gigawatt. Anthropic is seeking financial support or guarantees from Alphabet subsidiary Google for the lease payments. These details are based on reporting. Anthropic and Google have not publicly confirmed the specific lease plans in this form as finalized. What is officially documented, however, is that Anthropic is massively expanding its compute infrastructure. Anthropic announced in April 2026 a new agreement with Google and Broadcom for several gigawatts of TPU capacity, which is expected to go online starting in 2027. Back in October 2025, Anthropic had announced that it wanted to expand the use of Google Cloud technologies to up to one million TPUs. The new report therefore fits a clear pattern: Anthropic is scaling not only Claude, but the entire hardware base behind it. Until now, the simpler path for AI start-ups was: rent cloud capacity, train models, sell product. This model is reaching its limits at the largest AI labs. Anyone who continuously trains very large models while also having to serve high inference loads for customers needs predictable access to chips, power, networking, and cooling. Cloud capacity rented at short notice can be expensive and scarce. Own leasing structures can therefore be attractive. Anthropic could control hardware and sites more long term, plan capacity better, and optimize costs over longer periods. At the same time, new risks arise: long-term payment obligations, site lock-in, energy contracts, technical operating responsibility, and dependence on financing partners. This is precisely where Google comes in. Google is already an important Anthropic partner and provides TPU capacity via Google Cloud. If Google were to additionally secure lease payments now, the relationship would become even closer. That would not be just cloud usage, but a mix of infrastructure partnership, financing vehicle, and strategic backstop. The report also shows how AI companies are changing. Anthropic is no longer just a model provider with an API. The company is increasingly becoming an infrastructure user with obligations that are more reminiscent of cloud and data center operators. That is precisely why private credit, guarantees, special purpose vehicles, and long-term offtake agreements are playing an increasingly important role in the AI market. Reuters reported a few days earlier on a $35 billion package for Anthropic compute with Apollo, Blackstone, and Broadcom technology. That also was not just about chips, but about financing, data center capacity, and long-term access.
Taken together, these reports paint a clear picture: The next AI round will not be decided by model quality alone, but by the ability to organize billions in power, silicon, and data center space. Google’s role is also interesting. On the one hand, Google strengthens Anthropic as an important AI partner. On the other hand, Google ties Anthropic more deeply to its own TPU and cloud infrastructure. This is a countermodel to the pure NVIDIA GPU world and helps Google establish TPUs more firmly as a serious alternative in the market. It remains unclear which data center sites are specifically affected, how binding the preliminary agreements are, and to what extent Google actually assumes guarantees. It is also open whether Anthropic will operate the infrastructure entirely itself or delegate operational tasks to specialized data center partners. At capacity levels in the gigawatt range, such details are crucial because power supply, network connectivity, and cooling quickly become the real bottleneck. The financial side also remains delicate. Long-term lease agreements can make compute cheaper, but only if demand for Anthropic services continues to grow strongly. If growth slows or technical requirements change, such obligations can weigh heavily on the balance sheet.
Conclusion
Anthropic is moving from an AI company to an infrastructure company against its will. That sounds less glamorous than a new model, but is probably more important. Anyone who wants to remain at the front in this industry long term needs not only good research, but guaranteed compute at absurd scales. The real question is therefore no longer whether Claude will get better. The question is who will secure the power bill, the chips, and the data centers for it.
Noch keine sichtbaren Antworten im Forum gefunden. Der Thread ist bereits angelegt und kann direkt im Forum geöffnet werden.