Anthropic Secures Multiple Gigawatts of Google TPU Capacity as Claude Revenue Hits $30 Billion.

Anthropic has signed an agreement with Google and Broadcom to secure multiple gigawatts of next-generation TPU capacity starting in 2027, the company announced on LinkedIn. The deal comes as the AI company's run-rate revenue surged to $30 billion, more than tripling from $9 billion at the end of 2025.

The partnership will provide computing infrastructure to train and serve what Anthropic calls "frontier Claude models," the next generation of its AI assistant. The vast majority of the TPU capacity will be located in the United States, adding to Anthropic's November 2024 commitment to invest $50 billion in American computing infrastructure.

Tensor Processing Units, Google's custom-designed chips for machine learning workloads, represent a strategic alternative to Nvidia's dominant GPU architecture. While competitors like ChatGPT and Gemini have largely relied on Nvidia hardware, Anthropic's deepening relationship with Google Cloud positions it to leverage purpose-built AI accelerators at unprecedented scale.

The "multiple gigawatts" commitment is substantial. For context, a large data center typically operates at 50-100 megawatts; multiple gigawatts represents capacity equivalent to dozens of major facilities. The 2027 timeline suggests these are next-generation TPUs not yet in production, likely successors to Google's current TPU v5 chips.

Broadcom's involvement points to custom silicon development. The chipmaker has worked with Google on TPU design and manufacturing for years, and its inclusion in this announcement suggests continued co-development of specialized AI hardware optimized for transformer architectures like Claude.

Anthropic's revenue trajectory tells a story of rapid enterprise adoption. The jump from $9 billion to $30 billion run-rate in roughly four months represents one of the fastest growth periods in enterprise software history. The company has partnered with organizations ranging from pharmaceutical giant Eli Lilly to Dartmouth College, the first Ivy League institution to deploy Claude campus-wide.

Recent product launches have expanded Claude's capabilities significantly. The company shipped Claude Opus 4.7 with tripled vision resolution and self-verification for complex tasks, and introduced web-based code execution for browser-controlled development workflows. These features have made Claude increasingly competitive with coding-focused tools like Cursor and Windsurf.

The company has also moved into specialized domains. Project Glasswing, announced earlier this year, uses a specialized Claude Mythos preview model to audit and secure critical software infrastructure, signaling Anthropic's ambition beyond general-purpose AI assistance.

The Google partnership extends Anthropic's existing relationship with the search giant, which has invested billions in the company. By securing dedicated TPU capacity years in advance, Anthropic is effectively locking in the computing resources needed to maintain its position as demand for AI services continues to accelerate.

This infrastructure commitment also carries geopolitical significance. With the majority of capacity sited in the United States, Anthropic is aligning with growing pressure for domestic AI development and deployment. The $50 billion infrastructure pledge referenced in the announcement positions the company as a major contributor to American AI competitiveness.

For developers and businesses building on Claude, the TPU partnership offers implicit assurance of long-term availability and performance. As AI companies face ongoing questions about compute constraints and model access, Anthropic's multi-gigawatt commitment suggests it's planning for sustained growth rather than temporary scaling.

The 2027 timeline means the first frontier models trained on this new infrastructure are still years away. In the meantime, Anthropic will continue operating on its current compute allocation while preparing for what it clearly believes will be even larger-scale AI workloads ahead.