Anthropic Expands AI Compute Deal with Google and Broadcom as Revenue Hits $30B Run Rate

Artificial Intelligence research firm Anthropic has dramatically escalated its commitment to AI infrastructure, announcing a threefold expansion of its compute capacity agreements with tech giants Google and Broadcom. The landmark deal—tripling the companies' previous 2025 agreement—secures 3.5 gigawatts of processing power by 2027, directly responding to an explosion in demand for Anthropic's Claude AI models across enterprise and defense sectors. This infrastructure push comes as Anthropic revealed its revenue run rate has skyrocketed to $30 billion, a more than threefold increase from the $9 billion recorded just six months prior, signaling the accelerating commercial adoption of generative AI technologies despite ongoing supply chain scrutiny.

• Anthropic's compute deal with Google and Broadcom expands to 3.5 gigawatts by 2027, tripling the original October 2025 agreement
• Company's annual revenue run rate surged from $9 billion to $30 billion in under six months
• Majority of new compute infrastructure will be U.S.-based as part of $50 billion commitment to domestic AI development
• Partnership involves expanded use of Google Cloud's tensor processing units (TPUs) for powering Claude models
• Expansion occurs amid enterprise demand growth and regulatory scrutiny over AI supply chain security

Why Anthropic's Infrastructure Expansion Matters for the AI Race

The AI industry's infrastructure requirements have reached unprecedented scale as model complexity and user adoption accelerate. Anthropic's latest compute deal represents one of the most substantial infrastructure commitments to date in the generative AI sector, where model training and inference demands now exceed terawatt-hour levels of electricity consumption annually. Google's TPUs—custom silicon designed specifically for AI workloads—and Broadcom's networking components will form the backbone of this new capacity, enabling Anthropic to train and deploy larger, more sophisticated versions of its Claude models.

The Technical Backbone: TPUs and Scale-Out Computing

Google Cloud's Tensor Processing Units represent a fundamental shift from traditional CPU/GPU architectures toward specialized AI accelerators. First introduced in 2016 and now in their fourth generation (TPU v4), these chips deliver up to 123 teraflops of performance per chip while consuming significantly less power than comparable GPUs. The 3.5 gigawatt commitment translates to approximately 3.5 million servers running at full capacity—roughly equivalent to the power consumption of a mid-sized American city. Broadcom's role centers on providing high-speed interconnect solutions that enable these TPU pods to communicate efficiently, with each pod containing thousands of individual chips working in parallel.

Revenue Surge Reflects Enterprise AI Adoption Wave

Anthropic's revenue trajectory offers a rare window into the commercial maturation of generative AI. The jump from $9 billion to $30 billion in run-rate revenue between December 2025 and April 2026 demonstrates how quickly enterprise customers have integrated AI into core business processes. More than 1,000 businesses now spend over $1 million annually on Claude models, with major deployments spanning customer service automation, software development assistance, and legal document analysis. This commercial momentum stands in contrast to the broader tech sector's cooling growth rates, with Anthropic's Series G funding round valuing the company at $380 billion—making it one of the most valuable startups in history.

Regulatory Tensions Complicate Infrastructure Push

Anthropic's infrastructure expansion occurs against a backdrop of increasing regulatory scrutiny over AI supply chains. In late 2025, the U.S. Department of Defense designated Anthropic as a "supply-chain risk," citing concerns over chip fabrication and software dependencies. This designation complicates the company's ability to secure government contracts while simultaneously pursuing massive infrastructure investments. Krishna Rao, Anthropic's CFO, framed the expansion as a strategic move to "define the frontier of AI development," but the regulatory environment adds layers of complexity to what would otherwise be a straightforward scaling operation.

The $50 Billion U.S. Compute Pledge: Geopolitical Dimensions

Anthropic's commitment to house the majority of its new compute capacity within the United States reflects broader federal efforts to decouple from foreign semiconductor supply chains. The $50 billion infrastructure pledge aligns with provisions in the CHIPS and Science Act, which provides $52 billion in subsidies for domestic semiconductor manufacturing. By localizing production, Anthropic aims to reduce latency for U.S.-based customers, comply with emerging export controls on advanced AI chips, and position itself favorably for federal AI research funding opportunities. This geographic strategy also responds to customer demands for data sovereignty guarantees in regulated industries like healthcare and finance.

“This groundbreaking partnership with Google and Broadcom is a continuation of our disciplined approach to scaling infrastructure: we are building the capacity necessary to serve the exponential growth we have seen in our customer base while also enabling Claude to define the frontier of AI development. We are making our most significant compute commitment to date to keep pace with our unprecedented growth.” — Krishna Rao, CFO of Anthropic

Broader Implications for the AI Infrastructure Ecosystem

Anthropic's compute expansion highlights broader trends reshaping the AI infrastructure landscape. The 3.5 gigawatt commitment represents one of the largest single deals in AI history, comparable in scale to major cloud providers' internal infrastructure builds. This deal also signals a shift toward "AI-first" data centers where power infrastructure, cooling systems, and chip architectures are optimized specifically for machine learning workloads. Industry analysts note that such mega-deals are becoming necessary as model parameters exceed 1 trillion and training datasets approach exabyte scales. The partnership with Broadcom—known for its networking expertise—underscores how AI infrastructure is becoming increasingly specialized, with compute, storage, and networking components all requiring custom optimization.

Historical Context: From Research Labs to Industrial-Scale AI

The generative AI revolution traces its roots to breakthroughs in transformer architecture first published in 2017, but the infrastructure demands only became apparent in 2022-2023 as models like ChatGPT demonstrated commercial viability. Early AI deployments relied on smaller GPUs and cloud instances, but the shift to trillion-parameter models has necessitated custom silicon and massive data center footprints. Anthropic's progression mirrors that of other AI pioneers: OpenAI's partnership with Microsoft for Azure supercomputing, and Google's own TPU v4 pods powering its PaLM models. The 3.5 gigawatt figure places Anthropic in the same league as major cloud providers, indicating that AI research labs are now operating at utility-scale infrastructure levels.

What Comes Next for Anthropic and Its Partners

With 3.5 gigawatts of capacity coming online in 2027, Anthropic's immediate focus will shift to training and deploying its next-generation models, rumored to exceed 100 trillion parameters. The company's Series G funding round—one of the largest in tech history—provides the capital to accelerate this expansion while absorbing the operational complexity of managing such vast infrastructure. Google Cloud will benefit from increased TPU utilization, potentially leading to new AI-as-a-service offerings, while Broadcom stands to gain significant networking revenue from high-speed interconnects. Analysts anticipate that Anthropic will need to raise additional capital within 18 months to sustain this growth trajectory, given the breakneck pace of demand.

The Competitive Landscape: AI Infrastructure as a Battleground

Anthropic's infrastructure push places it in direct competition with other AI heavyweights for limited semiconductor supply. Nvidia remains the dominant player in AI GPUs, but companies like AMD and Intel are rapidly developing alternatives. Meanwhile, hyperscalers Amazon (with Trainium chips) and Microsoft (partnering with Nvidia and AMD) are building their own AI infrastructure to reduce dependence on external providers. The Google-Broadcom-Anthropic deal specifically targets TPU optimization, creating a three-way ecosystem that could challenge Nvidia's near-monopoly in AI training hardware. This competitive dynamic is driving rapid innovation cycles, with new chip architectures expected every 12-18 months.

Frequently Asked Questions