Google’s reported plan to invest as much as $40 billion in Anthropic is not just another oversized AI funding round. It is the latest sign that the frontier AI race is now as much about power, chips and data center capacity as it is about model quality.
The reported deal, first reported by Bloomberg, calls for Google to invest $10 billion now, with another $30 billion possible if Anthropic hits performance milestones. It would deepen a relationship that already makes Google both a supplier of cloud infrastructure and AI compute—and a strategic backer of one of its most important AI rivals.
The compute side may matter even more than the cash. Anthropic said this month it signed a new agreement with Google and Broadcom for “multiple gigawatts” of next-generation TPU capacity expected to come online starting in 2027. In plain English: Anthropic is locking up a guaranteed slice of the chips, electricity, networking and data center infrastructure it needs to train and run Claude at scale.
The pattern continues.
Amazon said on Monday it will invest $5 billion more in Anthropic, with up to $20 billion to follow. Anthropic, in turn, committed to spend more than $100 billion on AWS technologies over 10 years and secure up to 5 GW of Trainium chip capacity.
Microsoft and Nvidia inked a similar deal in November. Anthropic committed to buy $30 billion of Azure compute and contract up to 1 GW of additional capacity, while Nvidia and Microsoft said they would invest up to $10 billion and $5 billion, respectively, in Anthropic.
Deloitte expects inference – jargon to describe the everyday running of AI models after they are trained – to account for roughly two-thirds of AI compute in 2026. And when that day comes, every chatbot response, coding request, AI agent task and enterprise workflow will add another pull on a growing scarce number of available chips and data center to run them.
OpenAI is following the same infrastructure logic. It said in February that Amazon would invest $50 billion in OpenAI, while OpenAI and AWS expanded an existing $38 billion cloud agreement by another $100 billion over eight years. The deal includes roughly 2 GW of Trainium capacity.
The circular nature of these deals are hard to miss.
Cloud providers invest billions into AI labs. AI labs commit billions back in cloud spending. Both sides get growth, capacity, market share. Today’s capital spending, they hope, will allow platform dominance in the future.
The reality is that compute power now matters even more than the amount of cash on hand these companies have. In today’s market the priority is whoever can guarantee sustained access to scarce GPUs, power, and data center capacity ultimately dictates how fast models can be built, deployed and monetized.
The more gigawatts of processing power equals more chips, data center space, power, networking, and AI dominance.
S&P Global said new spending plans from Meta, Google and Amazon helped push expected 2026 capital spending by the top five U.S. hyperscalers—the giant cloud and infrastructure companies that build and run massive global data center networks—above $700 billion.
Image by Pete Linforth from Pixabay
