In the face of tightening AI compute resources, Google Cloud and AI startup Anthropic on Thursday formally announced a strategic partnership worth several billion dollars. The agreement will provide Anthropic access to up to one million Google‑designed tensor processing units (TPUs) along with additional Google Cloud services. The deal is expected to bring online more than 1 GW of AI compute capacity by 2026, underscoring Google's expanding ambitions in AI infrastructure and Anthropic's flexible multi‑cloud approach.
According to Google Cloud CEO Thomas Kurian, the collaboration deepens a long‑term partnership. "Anthropic's decision to massively expand its use of TPUs reflects the team's long‑standing recognition of TPU price‑performance and efficiency," Kurian said. Google's TPU family, especially the upcoming seventh‑generation "Ironwood," is widely regarded as a machine‑learning accelerator designed for large‑scale training and inference, and can offer energy‑efficiency advantages over NVIDIA GPUs on certain workloads. Anthropic CFO Krishna Rao added, "A long‑term partnership with Google will help us continuously scale compute to define the frontiers of AI."
Founded in 2021 by former OpenAI researchers and based in San Francisco, Anthropic focuses on building safe, controllable AI systems. Its flagship Claude series of large language models has stood out in enterprise applications. Claude Code, Anthropic's agent‑style coding assistant, reached $500 million in annualized revenue within two months of launch, becoming one of the fastest‑growing AI products on record. Anthropic's revenue run‑rate is approaching $7 billion, and it serves over 300,000 enterprise customers, with the number of large customers spending more than $100,000 annually growing nearly sevenfold in the past year. This rapid expansion has been driven by Claude's strengths in code generation, natural language processing and multimodal tasks, but it has also created unprecedented demand for compute.
At the heart of the deal is Anthropic's multi‑cloud strategy. Unlike many AI companies that rely on a single vendor, Anthropic spreads workloads across Google TPUs, Amazon Trainium chips and NVIDIA GPUs. This diversified approach optimizes cost and performance while reducing supply‑chain risk. For example, Google TPUs are used for specific training tasks, Amazon Trainium supports Anthropic's "Project Rainier" supercluster, and NVIDIA GPUs handle intensive inference workloads. The hybrid‑cloud model ensures flexibility during model iteration and helps avoid bottlenecks caused by shortages of a single chip type.
From Google's perspective, the agreement is a key step in expanding the TPU ecosystem. TPUs have long served Google's internal needs, such as training the Gemini model and search optimization. But with NVIDIA GPU supply constraints in recent years, Google has accelerated external access to TPUs via Google Cloud leasing. Supplying one million TPUs—the largest single TPU commitment on record—elevates Google Cloud's AI infrastructure. Industry estimates put the cost of building a 1 GW data center at roughly $50 billion, with chips accounting for about $35 billion. The deal also cements Google's role as a major investor in Anthropic (about $3 billion in total, including $2 billion in 2023 and $1 billion in early 2025) and strengthens its competitiveness in the AI chip market.
The scramble for AI compute has become a central battleground in global tech. Rival OpenAI has recently signed multiple agreements that could total more than $1 trillion to secure roughly 26 GW of compute capacity—enough to power some 20 million U.S. households. This reflects the "compute hunger" of scaling AI models: training a cutting‑edge large model requires tens of thousands of GPU/TPU days, while real‑time inference demands are enormous. Anthropic's multi‑platform strategy is seen as an innovative response; CFO Rao emphasized, "By keeping three big suppliers in competition, we secure better pricing and capacity guarantees." Nevertheless, the approach faces challenges: cross‑platform compatibility, network latency and energy management require significant optimization. Google's Ironwood TPU introduces advanced interconnect architectures that may help address these pain points.
More broadly, the agreement highlights a surge of investment in AI infrastructure. Amazon has committed $8 billion to Anthropic, providing Trainium chips, while Google's investment targets the TPU ecosystem. Anthropic recently closed a $13 billion funding round led by Iconiq Capital, lifting its valuation to $183 billion (including the raised capital), and attracting international investors such as Abu Dhabi's MGX fund. The funds are being used not only for hardware procurement but also to support Anthropic's "responsible scaling" initiatives, including model alignment, safety audits and ethics research.
Despite the promising outlook, the AI compute race raises concerns. Global data center energy consumption is climbing, driving up carbon emissions; chip supply chains remain concentrated among a few giants, creating geopolitical risks. Although Anthropic emphasizes explainable AI in Claude, balancing rapid innovation with safety remains a test. Experts warn that while multi‑cloud strategies are efficient, they must carefully address data privacy and regulatory compliance, particularly under the EU's GDPR framework.
Looking ahead, the Google‑Anthropic agreement could accelerate AI democratization. A wider range of enterprise applications—from automated coding to intelligent decision‑making—stand to benefit from more efficient compute. Kurian said optimistically, "We will continue to innovate TPUs to help move AI from the lab into reality." Anthropic's Rao stressed, "Compute is the foundation of the AI frontier; this expansion will help us build more reliable intelligent systems."
In the AI "arms race," Google's million‑TPU supply is more than a technical transaction—it's a strategic move. It signals that compute is shifting from a bottleneck to an accelerator of innovation. Anthropic's rise also demonstrates that startups can thrive through smart alliances amid competition among tech giants.