A US$10 billion partnership gives Anthropic access to 1 million Google TPUs — signalling a turning point in the AI chip race long dominated by Nvidia.
Google’s decade-old tensor processing units (TPUs) — once seen as internal tools powering its own AI models — are now emerging as a formidable challenger in the high-stakes AI hardware arena. A new multibillion-dollar deal with Anthropic marks the strongest validation yet for Google’s AI chips outside its own ecosystem.
A Billion-Dollar Boost for Google’s AI Chips
On October 23, 2025, Anthropic PBC — the AI startup behind Claude — announced a massive partnership with Google Cloud, granting access to over one gigawatt of computing power worth tens of billions of dollars. The agreement provides Anthropic with up to 1 million Google TPUs, expanding its capacity to train and run complex AI models.
The deal represents one of the largest TPU deployments to date, highlighting growing industry demand for alternatives to Nvidia’s graphics processing units (GPUs). Analysts see this as a key win for Google’s cloud business, which has long trailed Amazon Web Services and Microsoft Azure.
Breaking Nvidia’s Grip on the AI Chip Market
For years, Nvidia’s GPUs have been the gold standard in artificial intelligence — powerful, flexible, and backed by a mature software ecosystem. However, their high cost and limited supply have led companies to seek new options.
TPUs, by contrast, are application-specific integrated circuits (ASICs) designed solely to accelerate machine learning. They can outperform GPUs on certain AI workloads while consuming less energy, making them more cost-efficient at scale.
According to Seaport analyst Jay Goldberg, the Anthropic deal serves as “a powerful validation” of Google’s long-term chip strategy. “Many were already considering TPUs — now even more are likely to explore them,” he noted.
A Decade of Refinement
Google began developing its first TPU in 2013, launching the initial version in 2015 to speed up its own search engine and AI systems. By 2018, TPUs were made available to external developers through Google Cloud, offering access to the same infrastructure used internally by DeepMind and Google’s Gemini AI team.
Over seven generations, Google has improved the TPU’s processing power, energy efficiency, and cost-performance ratio. Its latest model, TPU v7 “Ironwood”, unveiled in April 2025, features liquid cooling and comes in configurations of up to 9,216 interconnected chips — enabling faster inference and training capabilities.
Strategic Partnerships and Industry Adoption
Beyond Anthropic, several major AI players — including Salesforce, Midjourney, and Safe Superintelligence (founded by OpenAI’s co-founder Ilya Sutskever) — are now leveraging Google’s TPUs. This growing list of adopters points to the technology’s rising relevance in commercial AI development.
Bloomberg Intelligence analysts Mandeep Singh and Robert Biggar suggested the Anthropic deal could pave the way for TPUs to be used beyond Google Cloud, potentially in smaller “neo-clouds” offering AI computing power.
Complementing, Not Replacing, GPUs
Despite the breakthrough, experts caution that TPUs are not replacing Nvidia GPUs anytime soon. Google remains one of Nvidia’s largest customers to ensure flexibility for varied workloads. “If a customer’s algorithm changes, GPUs still handle a broader range of tasks,” said Gaurav Gupta, an analyst at Gartner.
Key Banc’s Justin Patterson echoed this, noting that TPUs are “less versatile” but “strategically vital” for scaling Google Cloud’s share of the AI infrastructure market.
The Ripple Effect of Google’s Chip Legacy
The TPU project has spawned an influential alumni network shaping the broader AI hardware landscape. Jonathan Ross, who helped design the first TPU, now leads chip startup Groq. Other TPU veterans like Richard Ho and Safeen Huda have joined OpenAI, contributing to new generations of AI hardware innovation.
Their work continues to amplify Google’s influence across the AI industry, even beyond its own products. “There really is no substitute for this level of experience,” said Mark Lohmeyer, Google Cloud’s VP of AI and computing infrastructure.
Google’s renewed push into AI hardware through TPUs marks a strategic inflection point in the global AI arms race. By offering a scalable, energy-efficient alternative to Nvidia’s GPUs, the company positions itself as both a hardware innovator and a key enabler of the next wave of AI breakthroughs — reshaping the competitive landscape of cloud computing and artificial intelligence worldwide.
Sources: Bloomberg Technoz (2025) , Yahoo Finance (2025)
Keywords: Google TPU, Anthropic, Nvidia, AI Chips, Cloud Computing, Machine Learning
