Laika AI
Last Updated
April 23, 2026

Anthropic announced an expanded collaboration with Amazon on Tuesday to secure up to 5 gigawatts of dedicated compute capacity for training and deploying its Claude family of large language models. The agreement marks one of the largest disclosed AI infrastructure deals to date and deepens the strategic ties between the AI lab and Amazon Web Services.
According to Anthropic, nearly 1GW of the new capacity is expected to come online by the end of 2026. The remaining 4GW will be phased in over subsequent years as data center construction and energy procurement schedules allow. The compute will be used for both pretraining next-generation Claude models and for high-volume inference workloads serving enterprise and consumer customers on AWS.
The 5GW figure dwarfs most public compute commitments in the AI sector. For context, 1GW can power roughly 750,000 U.S. homes. Industry analysts note that securing power, not chips, is now the primary bottleneck for frontier AI labs. By locking in 5GW, Anthropic aims to ensure long term training capacity and reduce latency for Claude users globally.
Amazon will provide the capacity through a mix of dedicated AWS Trainium and Inferentia clusters, plus GPU based instances for workloads that require them. The companies said the buildout will span multiple AWS Regions, with a focus on locations where Amazon has secured clean energy agreements. Both firms reiterated prior commitments to match data center electricity use with 100% renewable energy.
Claude model training runs require dense clusters that operate for months at a time, consuming hundreds of megawatts per training job. After release, inference demand scales with user adoption. Anthropic cited rising enterprise use of Claude 3.5 Sonnet and upcoming multimodal features as drivers for the expansion. The company said the additional compute will help reduce queue times, improve reliability, and enable longer context windows and faster real-time responses.
No comments yet. Be the first!
The deal also supports Anthropic’s safety research. Larger pretraining runs let researchers test alignment techniques at scale. More inference capacity allows broader red teaming and monitoring before wider releases. Executives framed the 5GW as critical to keeping Claude competitive with rival models while maintaining the company’s safety standards. For context on Claude’s evolving capabilities, see the recent Claude Mythos preview detailing its reasoning upgrades.
Amazon invested $4 billion in Anthropic in 2023 and made AWS the primary cloud provider for Claude. The new compute agreement extends that partnership beyond capital into infrastructure planning. Amazon gains a marquee AI workload to optimize its Trainium chips and power procurement strategy. In return, Anthropic gets priority access to next generation AWS silicon and energy backed data center capacity.
Neither company disclosed financial terms. However, analysts estimate that 5GW of dedicated AI compute, including power and cooling, could represent tens of billions in capital expenditure over the life of the deal. The phasing to 1GW by the end of 2026 suggests sites are already under construction or in advanced permitting. The expansion comes as Anthropic faces product scrutiny from users, including crypto traders reacting to recent Claude AI changes.
The announcement intensifies the race among AI labs to secure power purchase agreements and grid interconnections. Microsoft, Google, and Meta have all announced gigawatt scale projects in the past year. Energy analysts warn that AI load growth could strain local grids, prompting calls for co-located nuclear, natural gas with carbon capture, and advanced geothermal projects.
Anthropic and Amazon said sustainability remains a priority. New data centers tied to the deal will aim for high water efficiency and use AWS's custom cooling designs. The companies plan to share periodic updates on renewable energy matching as the 1GW tranche comes online in 2026.
For developers building on Claude via Amazon Bedrock, the near term impact should be improved availability and lower latency during peak hours. Longer term, the 5GW runway gives Anthropic headroom to train larger models without pausing inference services. The company did not announce new model versions today, but said the compute will support its roadmap through 2028. That roadmap includes broader platform support, such as the Claude for Windows launch, bringing feature parity to AI agents.
The deal underscores how AI competition is now as much about megawatts and transformers as it is about algorithms. As Claude, GPT, and Gemini scale up, partnerships that guarantee power will separate the leaders from the rest of the pack.