Google's $40B Anthropic Bet: Cash, Compute, and Claude

TL;DR

On April 24, 2026, Google committed up to $40 billion to Anthropic in a two-tranche deal — $10 billion upfront in cash at a $350 billion valuation, with $30 billion more conditional on Anthropic hitting performance and computing-power milestones¹²³. The package is paired with 5 gigawatts of dedicated Google Cloud TPU capacity rolling out to Anthropic over five years¹⁴. The deal lands four days after Amazon announced it will invest up to $25 billion in Anthropic — $5 billion now at a $380 billion valuation, plus $20 billion tied to commercial milestones — in exchange for Anthropic committing more than $100 billion in AWS spend over a decade and bringing roughly 1 gigawatt of Trainium2 and Trainium3 capacity online by the end of 2026⁵⁶. Anthropic said earlier in April that its annualized revenue run rate has crossed $30 billion, up from about $9 billion at the end of 2025⁷⁸. The two parallel deals together add up to one of the largest single-company infrastructure carve-outs in AI history.

What You'll Learn

• The exact terms of Google's two-tranche $40B commitment, including the $350B valuation and the 5 GW compute figure
• How the Google deal compares with Amazon's parallel $25B investment four days earlier
• Why Anthropic is taking compute and capital from competing hyperscalers simultaneously
• What Anthropic's $30B run-rate revenue tells us about why the deals are this size
• What the Broadcom-built TPU expansion in 2027 adds to the picture
• The open questions: valuation discrepancies, milestone-gated tranches, and regulatory exposure

The Deal: $10B Now, $30B Later, 5 Gigawatts Over Five Years

Bloomberg first reported the deal on April 24, 2026, with TechCrunch, CNBC, and Reuters confirming the structure within hours¹²³⁹. The terms break down cleanly:

The $10 billion cash portion alone makes this one of the largest single private-market AI investments on record. The $40 billion ceiling — if both tranches close — would make Google one of Anthropic's two largest backers, alongside Amazon²⁹.

The structure is notably different from the typical equity round. Most AI lab funding rounds are pure cash for equity. This deal is cash plus compute capacity reserved on Google Cloud, with the second tranche gated on milestones rather than time. That gives Google two simultaneous returns: equity upside if Anthropic continues compounding revenue, and recurring infrastructure revenue from Anthropic's TPU consumption.

The $350B vs. $380B Valuation Question

One detail worth flagging: Google's $10 billion was reported at a $350 billion valuation¹². Amazon's $5 billion four days earlier was at $380 billion⁵⁶. Anthropic's Series G in February 2026 priced the company at a $350 billion pre-money / $380 billion post-money valuation¹⁰.

The most plausible reading is that the two deals are using different reference points from the same Series G price stack — Amazon's at the post-money mark, and Google's reported at what looks closer to the pre-money. Bloomberg's reporting on the Google deal noted that the $350 billion figure matches the valuation Anthropic carried at the February funding round¹. Neither company has published a fully reconciled per-share economics. For observers, the practical takeaway is that within the same calendar month, Anthropic's shares are pricing in a tight band around the Series G price stack rather than across a wide range.

The Compute Angle: 5 GW From Google, ~5 GW From Amazon

The cash is the headline; the compute is the substance. Five gigawatts of TPU capacity is a serious number. For context, a single gigawatt of always-on data-center load is roughly the peak power draw of a mid-sized US city the size of San Francisco — so the 5 GW Google has reserved for Anthropic represents power on the scale of a major metropolitan area like Miami¹¹.

The 5 GW commitment from Google sits on top of two earlier deals:

• October 23, 2025: Anthropic-Google Cloud expansion. Anthropic announced access to up to 1 million TPU chips and "well over a gigawatt of capacity coming online in 2026" — a deal Google described at the time as worth tens of billions of dollars¹²¹³.
• April 6-7, 2026: Anthropic, Google, and Broadcom announced an expanded TPU deal. A Broadcom SEC filing showed 3.5 gigawatts of next-generation TPU capacity scheduled to come online starting in 2027¹⁴¹⁵.

The April 24 commitment of 5 GW over five years appears to subsume and extend those earlier numbers rather than stack on top of them, though Anthropic and Google have not published a fully reconciled compute roadmap. What is unambiguous is that Anthropic's TPU capacity on Google Cloud will scale from "well over a gigawatt" in 2026 toward several gigawatts by the end of the decade.

On the Amazon side, the picture is similar in scale. Anthropic's April 20 deal with Amazon commits roughly 1 gigawatt of Trainium2 and Trainium3 capacity online by the end of 2026, with the broader AWS partnership reaching toward roughly 5 GW of compute capacity over time⁵¹⁶.

Two parallel ~5 GW commitments is a multi-cloud bet that — if both deliver — gives Anthropic compute optionality that few AI labs can match.

Amazon's $25B Counter-Move Four Days Earlier

Amazon's April 20 announcement is the necessary context for the Google deal⁵¹⁷. The terms:

• $5 billion now at Anthropic's $380 billion valuation
• Up to $20 billion in additional investment tied to "certain commercial milestones"
• $100+ billion of AWS spend committed by Anthropic over the next decade
• This is on top of the $8 billion Amazon had previously invested in Anthropic

Amazon's deal arrived less than two months after the company committed to invest up to $50 billion in OpenAI on February 27, 2026⁵. The OpenAI-and-Anthropic dual commitment is Amazon's hedge: AWS revenue scales whether the leading frontier lab is Anthropic, OpenAI, or both.

The strategic logic is similar for Google. Google is one of Anthropic's two largest cloud partners (alongside AWS, which Anthropic has named its primary training partner) and runs Gemini — its own competing frontier lab. The $40 billion bet is partly defensive (lock in TPU revenue and Anthropic's compute continuity), partly opportunistic (equity upside if Claude continues outpacing internal models), and partly insurance (ensure that if Anthropic does win the frontier race, Google is on the cap table).

Why The Money Is This Big: The $30B Run-Rate

These numbers do not make sense without Anthropic's revenue trajectory. On April 7, 2026, Anthropic disclosed that its annualized run-rate revenue had crossed $30 billion, up from about $9 billion at the end of 2025 — roughly a tripling in four months⁷⁸.

A few revealing data points:

• Claude Code hit $2.5 billion in annualized revenue by February 2026⁷
• The number of customers spending more than $1 million per year on Anthropic products doubled from roughly 500 in February 2026 (at the Series G) to over 1,000 by April 2026⁷
• Anthropic's $30B ARR in April 2026 is the first time the company has out-paced OpenAI's reported run-rate of approximately $25 billion⁸¹⁸

When a private company is growing run-rate revenue from $9B to $30B in four months, the binding constraint is no longer demand, customer acquisition, or model quality. It is compute supply. Anthropic itself has framed it that way, citing "inevitable strain" on infrastructure that has impacted reliability and performance⁵.

That framing explains both deal structures. Google and Amazon are not just buying equity at a high multiple — they are selling Anthropic the one input it cannot manufacture: power-and-silicon at gigawatt scale.

What This Means for the AI Infrastructure Race

Three takeaways for builders, investors, and observers:

1. Multi-cloud is a strategic posture, not just a technical convenience. Anthropic now has gigawatt-scale capacity reserved on Google's TPU stack and AWS's Trainium stack. Either supplier could under-deliver — power, chips, networking, or all three — without taking the company offline. OpenAI has built its own multi-cloud footprint over the past year — Microsoft Azure, a $38 billion AWS deal signed in November 2025, and the multi-hundred-billion-dollar Oracle Stargate program — so this kind of diversification is becoming the norm rather than the exception at the frontier¹⁹.

2. Hyperscalers are willing to fund their competitors at unusual scale. Google runs Gemini. Amazon does not have a flagship frontier model but competes on AWS as the substrate for everything. Both are putting tens of billions into a company whose flagship product can take revenue from their own AI offerings. For Google in particular — which competes with Anthropic in the enterprise market via Gemini and the recently-rebranded Gemini Enterprise Agent Platform (formerly Vertex AI) — the deal signals that TPU revenue and equity exposure to Claude outweigh the strategic risk of strengthening a direct competitor.

3. The cost of compute is now an explicit milestone. Both deals tie a large majority of the headline figure to performance and commercial milestones — Google's $30B-of-$40B is 75% conditional, Amazon's $20B-of-$25B is 80% conditional. That structure protects the hyperscaler from over-paying if growth slows, and gives Anthropic an aligned incentive to keep capacity utilized. Expect this two-tranche pattern — modest cash upfront, large conditional follow-on tied to milestones — to become more common as AI lab valuations push past $300 billion.

The Bottom Line

The Google deal is not, on its own, a surprise — Anthropic and Google Cloud have been deepening their partnership for over a year. What makes April 24 a milestone is the scale and timing: $40 billion committed within four days of Amazon's $25 billion, both tied to gigawatt-scale compute commitments, both at valuations between $350 and $380 billion.

The simplest read is that the 2026 frontier-AI race is no longer primarily about model quality. Anthropic's $30B run-rate is direct evidence that demand for frontier capability has outrun the supply of compute. The companies with capital and infrastructure — Google, Amazon, and on the OpenAI side, Microsoft, AWS, and Oracle — are now writing checks structured less like venture investments and more like long-term supply contracts.

For Anthropic specifically, the bet is that with 10 GW of combined Google + AWS capacity disclosed across the next several years, plus the Broadcom-built TPU expansion arriving in 2027, the compute ceiling that has been throttling reliability and feature velocity will lift before competitive pressure forces a model-quality reckoning. For Anthropic's revenue trajectory, the next 12 months will determine whether the run-rate continues compounding into a number that justifies $40 billion of follow-on capital — or whether the second tranche stays on the shelf.

What is no longer in doubt: the price of admission to the frontier-AI race has cleared $50 billion per relationship. Tomorrow's question is who is left to write the next one.

References