On Friday, April 24, 2026, Anthropic posted a blog titled "Expanding our use of Google Cloud TPUs and Services." The headline read like a routine partnership update. The numbers underneath did not.
Google was wiring ten billion dollars in cash at a $350 billion post-money valuation.
Another thirty billion was promised if Anthropic hit a set of undisclosed performance milestones. The cash came packaged with 5 gigawatts of dedicated computing capacity on Google's seventh-generation TPU, the Ironwood, with access to as many as one million chips.
The blog post was 600 words long. It described the largest single check Google has ever written into a private company. It also described what is, in effect, a refund mechanism: most of Google's investment will return to Google as cloud spending on Google's own infrastructure.
Three days before, Amazon ran the same play. As covered in Amazon Just Put Another $25 Billion Into Anthropic, AWS committed an additional five billion dollars in cash with up to twenty billion more tied to commercial milestones, on top of eight billion dollars already invested.
The reciprocal: Anthropic pledged to spend roughly $100 billion on AWS compute over the next decade.
The math is now obvious. Anthropic's two largest investors are also its two largest landlords. The same capital that enters as equity exits as compute purchase orders.
The Circular Cash Flow
Look at what Google bought for its ten-billion-dollar check.
Anthropic gets dollars on its balance sheet at a $350 billion valuation. Google gets a roughly 14% ownership stake in the lab building Claude. Anthropic spends a meaningful share of those dollars renting Ironwood TPUs from Google Cloud. Google records that spend as Cloud revenue, which lifts Alphabet's cloud earnings, which lifts Alphabet's stock.
The structure is not new. Microsoft pioneered it with OpenAI in 2019, when ten billion dollars in cash was paired with Azure as the exclusive cloud provider. Nvidia has done a smaller version with CoreWeave and several other compute startups. Google and Amazon are now running it on the largest scale yet.
The difference: Anthropic is being capitalized by two cloud rivals at the same time. Google's deal does not include exclusivity. AWS remains Anthropic's primary training partner, and Claude will continue to ship across all three major clouds, including Microsoft Azure.
For data scientists, that matters. Claude is the only frontier model running on every major cloud's standard catalog, AWS Bedrock, Google Vertex AI, and Azure AI Foundry. The Friday deal does not change that. It hardens it.
What Anthropic Got
The compute terms in the Anthropic blog post are unusually specific.
| Resource | Amount | Timing |
|---|---|---|
| Cash investment | $10 billion immediate | Closed at $350B post-money |
| Contingent cash | Up to $30 billion | Tied to undisclosed performance targets |
| TPU capacity | 5 gigawatts | Over five years |
| Ironwood chips | Up to 1 million | Begins ramp 2026, scales 2027 |
| Capacity online in 2026 | Over 1 gigawatt | Confirmed in Anthropic blog |
The Ironwood chips are Google's seventh-generation TPU, announced earlier this month at Google Cloud Next 2026 in Las Vegas. Each chip delivers 4.6 petaFLOPS of FP8 compute and 192 GB of HBM3e memory, with 7.37 TB/s of memory bandwidth. A full Ironwood pod scales to 9,216 chips and 42.5 exaflops.
Anthropic says Ironwood will be used for both training and inference of Claude models, with the new capacity "starting to come online" through 2026 and accelerating into 2027. Google's earlier April 7 announcement with Anthropic and chip designer Broadcom locked in 3.5 gigawatts of TPU-based capacity beginning in 2027. Friday's deal lifts that to 5 gigawatts over five years.
For context, 5 gigawatts is roughly the peak summer electricity load of metropolitan San Francisco. The pledged compute build sits inside a broader story covered in Anthropic Tripled Its Revenue in 90 Days. Then It Signed for 3.5 Gigawatts, where the original Broadcom-and-Google partnership was announced.
How Anthropic Reached the $350 Billion Mark
Three primary rounds in seven months tell the story:
February 12, 2026: $30 billion Series G at $380 billion post-money, led by GIC and Coatue. The second-largest private financing round in tech history at the time.
April 24, 2026: $10 billion from Google at the $350 billion mark, with up to $30 billion contingent.
By April 23, a secondary tender offer for early investors and employees had already implied a valuation closer to one trillion dollars, as reported by Bloomberg and covered in Anthropic Passing OpenAI at One Trillion.
Friday's primary round came in at the $350 billion mark, roughly flat with February's pre-money level even as secondary markets traded the same shares above eight hundred billion dollars.
What anchors every one of those figures is revenue. Anthropic's annualized run rate hit $30 billion in April, up from one billion dollars in January 2025. Eight of the Fortune 10 are Claude customers. More than 1,000 businesses spend over one million dollars annually on the platform.
Why Google Is Funding the Lab Beating Gemini
Google sells Gemini. Anthropic sells Claude. In the consumer market, Gemini is taking share fast: ChatGPT's overall share fell from 87% to 68% between early 2025 and January 2026, while Gemini grew from 5.4% to 18.2% in the same period.
In the enterprise market, the picture is different. Claude holds an estimated 32% of the enterprise large-language-model API market by usage, ahead of OpenAI at 25%. Among companies buying frontier-model APIs for the first time, Anthropic now wins roughly 70% of head-to-head matchups against OpenAI.
Gemini barely registers in those head-to-heads. Google's own API business has lagged its Cloud platform's broader growth, even as Vertex AI expands. The lab beating Gemini at enterprise is the lab Google just bought 14% of.
The investment is therefore a hedge in two directions. If Claude wins the frontier API race, Google captures equity upside and TPU rental revenue. If Gemini catches up, Google still owns a major share of the runner-up. Either way, the alternative, where Anthropic drifts entirely into Microsoft's or Oracle's compute orbit, is closed off.
Anthropic CFO Krishna Rao framed the strategic logic in the lab's earlier April 7 announcement of the Broadcom and Google compute deal:
"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." — Krishna Rao, CFO at Anthropic (Anthropic press release, April 7, 2026)
How Big Tech's Anthropic Bet Has Escalated
The Other Side: A Defensive Bet With Strings Attached
Not everyone reads the announcement as a triumph for Anthropic.
The Motley Fool's Adria Cimino, writing on April 27, called the deal "a screaming bargain" for Google because the ten-billion-dollar check at a $350 billion mark prices in less than half of what secondary markets paid for the same shares a day earlier. If Anthropic's implied valuation reaches a trillion dollars in coming rounds, Google's latest tranche could effectively triple in value before any milestone payment is due.
Critics also flagged the contingency structure. The remaining thirty billion in follow-on capital is tied to undisclosed performance targets. Past contingent AI deals, including Amazon's $35 billion AGI-or-IPO clause with OpenAI, have rarely paid out on the schedule originally announced.
A separate question hangs over the cloud market itself. With Anthropic's biggest bills going to two competing cloud providers and most of Google's investment looping back into Google Cloud revenue, "AI revenue" as reported by Big Tech increasingly reflects the same capital changing hands inside a closed loop. Bernstein analyst Mark Moerdler has warned that this dynamic flatters near-term cloud growth in ways that may not reflect end-customer demand.
What It Means for ML Engineers and Data Scientists
Capacity comes first. Anthropic has been rate-limiting heavy Claude API customers throughout 2026, and Claude Code subscribers have repeatedly hit traffic caps during peak hours. A meaningful share of the new TPU capacity comes online this year. Expect rate-limit relief for Claude Sonnet and Opus customers in the second half of 2026.
Cost comes next. Anthropic has not announced price changes, but Google's own internal benchmarks put the all-in TCO per Ironwood chip at roughly 44% lower than an Nvidia GB200 server. That gives Anthropic room to undercut Opus and Sonnet pricing in late 2026.
Platform stability follows. Claude is now the only frontier model with three-way cloud redundancy across AWS, Google Cloud, and Azure. For enterprise teams running production workloads, that lowers single-vendor risk in a way Gemini and GPT-5.5 currently cannot match.
The last effect is the one nobody wants to discuss. Anthropic's two biggest investors and two biggest customers are now the same entities. If Claude becomes structurally dependent on AWS and Google Cloud spend, the lab's room to negotiate price, exclusivity, or revenue share narrows over time.
The Bottom Line
In one week, Anthropic took on roughly $65 billion in committed capital from its two cloud partners and added pledged compute commitments of more than one hundred billion dollars. The check from Google was the largest direct equity investment Alphabet has ever made into a private company.
The numbers describe a company growing faster than any frontier lab has before. They also describe a company whose financial life is now joined at the hip with two of the three companies it ultimately competes with on every layer of the stack: chips, cloud, models, and applications.
Sundar Pichai paid to make sure the most credible non-OpenAI frontier lab does not drift into a Microsoft or Oracle camp. The price of that insurance was ten billion now, thirty billion later, and a guarantee that most of the cash returns home as TPU rentals.
"We are moving to a compute-powered economy." — Greg Brockman, President at OpenAI, on the GPT-5.5 launch (April 2026)
Brockman was speaking about a different frontier lab. The line described all of them.
Sources
- Expanding our use of Google Cloud TPUs and Services (Anthropic Official Blog, April 24, 2026)
- Google to Invest Up to 40B in Anthropic in Cash and Compute (TechCrunch, April 24, 2026)
- Google Plans to Invest Up to 40 Billion in Anthropic (Bloomberg, April 24, 2026)
- Google to Invest Up to 40 Billion in Anthropic as Search Giant Spreads Its AI Bets (CNBC, April 24, 2026)
- Google's 40B Anthropic Move Is Big Tech's Latest Huge AI Bet (Axios, April 24, 2026)
- Anthropic Takes 5B from Amazon and Pledges 100B in Cloud Spending in Return (TechCrunch, April 20, 2026)
- Google Doubles Down on Anthropic With New 40 Billion Investment (PYMNTS, April 24, 2026)
- Google Is Getting a Screaming Bargain on Its New Anthropic Investment (The Motley Fool, April 27, 2026)
- Ironwood: The First Google TPU for the Age of Inference (Google Blog, April 2026)
- Google TPUv7: The 900lb Gorilla in the Room (SemiAnalysis, April 2026)
- Anthropic Ups Compute Deal With Google and Broadcom Amid Skyrocketing Demand (TechCrunch, April 7, 2026)