By Connie · Last reviewed: April 2026 — pricing & tools verified · This article contains affiliate links. We may earn a commission at no extra cost to you if you sign up through our links.
Anthropic Signs 3.5 Gigawatt TPU Deal With Google and Broadcom — Revenue Hits $30B
April 7, 2026 · 8 min read
Anthropic signed its largest infrastructure deal ever on April 6, 2026: 3.5 gigawatts of Google TPU capacity via Broadcom, worth ~$21 billion, coming online from 2027. The same day, Anthropic revealed its revenue run rate has tripled to $30 billion annually. This is a signal that Claude's compute is about to scale dramatically — and that AI infrastructure is now a geopolitical asset.
On April 6, 2026, Anthropic posted a single tweet: "We've signed an agreement with Google and Broadcom for multiple gigawatts of next-generation TPU capacity, coming online starting in 2027, to train and serve frontier Claude models."
The understated announcement masked an extraordinary number. The deal covers approximately 3.5 gigawatts of computing capacity — about the output of three nuclear reactors, dedicated exclusively to running AI models. It is the largest infrastructure commitment Anthropic has ever made.
The announcement came on the same day Anthropic disclosed that its annualized revenue run rate has surpassed $30 billion — up from $9 billion just four months earlier. Both numbers tell the same story: demand for Claude is growing faster than anyone expected, and Anthropic is moving to match it.
What Exactly Was Signed
The deal has three parties: Anthropic (the AI lab), Google (the chip designer), and Broadcom (the intermediary that manufactures and supplies custom silicon at scale).
Under the arrangement, Broadcom supplies Google TPU v7p units to Anthropic under long-term contracts. Broadcom has separately agreed with Google to design future TPU generations and supply networking components through 2031.
The financial terms: the deal is valued at approximately $21 billion covering nearly 1 million TPU v7p units. Analyst firm Mizuho projects Broadcom may generate $21 billion in AI revenue from Anthropic in 2026 alone, potentially doubling to $42 billion in 2027 as the new capacity comes online.
Anthropic's CFO, Krishna Rao, described it as a "disciplined approach to scaling infrastructure" — noting that the majority of the new capacity will be located in the United States, consistent with Anthropic's November 2025 commitment to invest $50 billion in American AI infrastructure.
The Revenue Story Behind the Deal
Revenue context matters here. Anthropic's $30 billion annualized run rate represents a 3.3x increase in roughly four months. The company's post-money valuation stands at $380 billion following its Series G funding round in February 2026.
Enterprise adoption is driving the surge. The number of business customers spending over $1 million annually with Anthropic doubled — exceeding 1,000 customers — in less than two months. These are large-scale API users: software companies, financial institutions, healthcare systems, and government agencies running Claude at production scale.
Access Claude Opus 4.6, GPT-5.4, Gemini 3.1 Pro, and more from one subscription. Happycapy Pro starts at $17/month.
Try Happycapy FreeWhy This Deal Is Different from Past Compute Announcements
The AI industry has seen large compute deals before. What makes this one notable:
- Scale: 3.5 gigawatts is a meaningful fraction of US data center power consumption. Most AI compute expansions are measured in megawatts or hundreds of megawatts — not multiple gigawatts.
- Timing: The capacity comes online starting in 2027, signaling Anthropic is planning for Claude models that require substantially more compute than current versions.
- Multi-vendor design: Anthropic deliberately maintains relationships with Google (TPUs), Amazon (Trainium), and Nvidia (GPUs). This prevents any single vendor from controlling Anthropic's training pipeline — a lesson learned from the early OpenAI-Microsoft dependency.
- Power competition: The deal puts Anthropic in direct competition with Bitcoin miners and other large power consumers for grid connections. CoinDesk noted that Anthropic is now "a new rival for cheap power" as the 3.5 GW capacity requires securing electricity contracts at unprecedented scale.
Anthropic vs. OpenAI: The Infrastructure Race
| Metric | Anthropic (April 2026) | OpenAI (April 2026) |
|---|---|---|
| Annualized revenue | $30 billion | $25 billion |
| Valuation | $380 billion | $340 billion |
| Compute deal size | $21B (Google/Broadcom TPUs) | Stargate ($500B total commitment) |
| Primary chip supplier | Google TPU + Amazon Trainium + Nvidia | Nvidia + Microsoft Azure |
| Capacity timeline | 3.5 GW from 2027 | Multi-campus from 2026–2028 |
| Primary model | Claude Opus 4.6 / Claude 4 | GPT-5.5 (Spud) |
What This Means for Claude Users
For end users, the immediate impact is limited — the new capacity doesn't come online until 2027. But the deal signals several things:
- More capable models: The scale of compute being secured implies future Claude models will be substantially larger and more capable than current versions. Anthropic specifically cited "next-generation frontier Claude models" as the purpose of the deal.
- Pricing stability: Locking in multi-year supply contracts at scale typically reduces per-unit cost over time. This could support stable or even lower API pricing as Anthropic's scale advantages compound.
- Inference capacity: The deal covers both training and inference (serving). Users who have experienced Claude API rate limits may see those constraints ease as new capacity comes online.
For context on accessing Claude today: Happycapy provides access to Claude Opus 4.6, GPT-5.4, Gemini 3.1 Pro, and a library of 15+ specialized AI skills in a single subscription — without managing individual API accounts or rate limits.
Pro plan from $17/month. No API key. No rate limits. Switch between Claude, GPT, and Gemini in one workspace.
Start Free on HappycapyFrequently Asked Questions
What did Anthropic announce with Google and Broadcom in April 2026?
On April 6, 2026, Anthropic announced a multi-gigawatt agreement for approximately 3.5 gigawatts of Google TPU capacity via Broadcom, valued at ~$21 billion. Capacity comes online from 2027. The announcement coincided with Anthropic revealing its revenue run rate has tripled to $30 billion annually.
What is Anthropic's current revenue in 2026?
Anthropic's annualized revenue run rate surpassed $30 billion as of April 2026 — up from $9 billion at end-2025. The company has over 1,000 enterprise customers each spending more than $1 million annually.
Does this deal make Anthropic dependent on Google?
No. Anthropic operates a deliberate multi-vendor strategy: Google TPUs via Broadcom, Amazon Trainium, and Nvidia GPUs. This provides infrastructure resilience and prevents any single vendor from controlling Anthropic's training pipeline.
Will this affect Claude's performance and pricing for users?
Not immediately — the new capacity arrives from 2027. Long-term, the scale economics should support stable or lower API pricing and improved inference capacity. Next-generation Claude models requiring substantially more compute are implied by the deal's scale.
Sources
Get the best AI tools tips — weekly
Honest reviews, tutorials, and Happycapy tips. No spam.