The artificial intelligence industry is reaching an inflection point where raw computing power is no longer enough, and strategic diversification is becoming the competitive advantage. Anthropic has announced that it is partnering up with Google to use up to one million units of Google Cloud TPUs, a deal worth "tens of billions of dollars", that would potentially bring over a gigawatt of capacity online by 2026.
Thomas Kurian, CEO at Google Cloud, said in the statement:
“Anthropic’s choice to significantly expand its usage of TPUs reflects the strong price-performance and efficiency its teams have seen with TPUs for several years. We are continuing to innovate and drive further efficiencies and increased capacity of our TPUs, building on our already mature AI accelerator portfolio, including our seventh generation TPU, Ironwood.”
Increasing capacity was an inevitable step, as Anthropic now serves over 300,000 business customers and has grown its roster of enterprise accounts spending over $100,000 annually nearly sevenfold in the past year, creating an insatiable demand for compute that a single cloud provider simply cannot satisfy.
By adding Google"s TPUs to its existing infrastructure mix of Amazon Trainium chips and NVIDIA GPUs, Anthropic is effectively hedging its bets across three different chip architectures, three different cloud providers, and three fundamentally different approaches to accelerating AI workloads.
The AI company projected $2.2 billion in annual revenue by 2025, which means it has rapidly outgrown what any single infrastructure partner could comfortably provision. Amazon, Anthropic"s primary cloud provider and an investor with an $8 billion stake, remains deeply embedded in the company"s infrastructure strategy, particularly around Trainium chips that AWS has been developing with close collaboration from Anthropic"s engineering teams.
The deal could eventually protect Anthropic against any future supply shocks, technological obsolescence, or something as simple as a vendor dependency. While Nvidia"s GPUs offer unmatched software maturity, vast ecosystem support, and proven reliability at an unprecedented scale, homegrown AWS Trainium chips come with deep technical integration with Anthropic"s research teams. TPUs (Tensor Processing Units), on the other hand, are Google-designed custom chips that excel at certain workloads and matrix multiplication patterns.
The deal could be a good one for Google as well, as it represents a concrete validation of the company"s TPU strategy and an anchor customer for its new seventh-generation TPU hardware, codenamed "Ironwood." The search giant has invested heavily in proving that TPUs offer compelling price-performance ratios compared to GPUs, particularly for the specific mathematical operations that foundation models rely on.
Despite that, Anthropic said that it will continue working with Amazon on Project Rainier, which is a massive compute cluster with hundreds of thousands of AI chips across multiple U.S. data centers, and continue investing in increasing capacity in the coming months.