The arrangement, announced Thursday, gives Anthropic vastly more computing capacity while cementing Google’s position as both a major investor and key infrastructure provider in the escalating race to power large AI models. The tensor processing units, or TPUs — Google’s custom-designed chips that accelerate machine learning workloads — are slated to be deployed in 2026, bringing more than a gigawatt of capacity online relatively soon, Google said in a statement. The deal ranks among the largest commitments yet in the AI hardware arms race, underscoring the soaring cost of developing cutting-edge AI systems that require immense computing power to train and run large language models.