Google commits up to $40 billion to Anthropic, doubling down on its top AI rival

Anthropic and Google confirmed on Friday that Alphabet will pour up to $40 billion into the AI lab behind Claude, in a staged deal that converts compute, cash, and conditional milestones into one of the largest single investments in the history of the AI industry. The check arrives as Anthropic's annualized revenue run-rate has rocketed past $30 billion this month, more than triple the roughly $9 billion the company was running at the end of 2025.

Under the terms disclosed by the two companies, Google is committing $10 billion immediately at a $350 billion valuation — below the $380 billion post-money mark Anthropic's February Series G closed at — with an additional $30 billion to follow if the startup hits a series of performance targets. Google Cloud has separately agreed to provide 5 gigawatts of computing capacity over the next five years, according to Bloomberg and TechCrunch reporting on the deal terms, almost all of it expected to be delivered on Google's in-house Tensor Processing Units, with the option to scale further.

The announcement caps a remarkable week of capital formation for Anthropic. Just days earlier, Amazon disclosed a new $5 billion investment in the company, with the door open to as much as $20 billion more. Earlier this month, Anthropic, Google, and Broadcom unveiled a separate infrastructure agreement for 3.5 GW of TPU-based capacity beginning in 2027, according to a subsequent Broadcom securities filing. Layered together, the deals add up to tens of billions of dollars of cash and well over a decade's worth of historic data-center build-out — all aimed at one company.

Frenemies, formalized

Google's relationship with Anthropic has always been the industry's most awkward open secret. Alphabet first put $300 million into the lab in 2023 for a stake reported at roughly 10%, then quickly added another $2 billion. Before today's deal, Google's cumulative investment had crossed $3 billion for a reported ~14% stake.

Google now finds itself in the strange position of being the single largest outside financial backer of the company whose flagship product, Claude, competes most directly with Google's own Gemini — particularly in the high-margin developer and coding markets where Anthropic has built a commanding lead.

Asked to characterize the partnership in past briefings, Google executives have leaned on the word "complementary." Investors will be entitled to ask whether $40 billion of complementarity still passes the laugh test.

Compute is the deal

Strip away the headline number and the structure of the agreement says something specific about where the AI race is now being won: at the substation, not the model card.

Anthropic's recent product cadence has been throttled by infrastructure. Customers spent much of the spring complaining about Claude usage limits, prompting CEO Dario Amodei to acknowledge the squeeze in a statement tied to the Amazon deal earlier in the week:

"Our users tell us Claude is increasingly essential to how they work, and we need to build the infrastructure to keep pace with rapidly growing demand."

Five gigawatts is not a marketing number. For context, that is roughly the steady-state output of five large nuclear reactors, and it dwarfs the total power footprint of most hyperscale operators just a few years ago. Delivered on TPUs rather than Nvidia GPUs, it also represents the most aggressive single endorsement to date of Google's custom silicon as a viable substrate for frontier-scale training and inference.

For Google Cloud, that is the prize. The unit has spent two years arguing that TPUs can stand toe-to-toe with Nvidia's H- and B-series accelerators on cost and throughput. Anchoring Anthropic's next training runs on TPUs gives that pitch a reference customer no competitor can match.

Spreading the bets

For Alphabet, the investment is also a portfolio move. With OpenAI tightening its alignment with Microsoft and a wave of new entrants — xAI, Mistral, Meta's open-weights push — competing for the next decade of enterprise contracts, Google appears unwilling to let any single horse, including its own, define its exposure to generative AI.

That logic cuts both ways. Anthropic now has compute commitments from Google, Amazon, and (via Broadcom) a third silicon path. The diversification reduces the company's dependence on any one cloud provider, but it also means Anthropic is increasingly the connective tissue holding together the infrastructure plans of the two largest cloud businesses on Earth.

Valuation gymnastics

The $350 billion mark Google paid is, by the standards of this market, already stale. Reports earlier this month indicated that a new Anthropic primary round is being marked at valuations north of $800 billion, with bankers sounding out an IPO window as early as October 2026.

If those numbers hold, Google's $10 billion check is being struck at less than half the price the next investor is expected to pay — a generous discount that reflects both the depth of the existing partnership and the strategic value of locking in five gigawatts of demand for TPUs.

The conditional $30 billion tranche is the more interesting line item. Neither company disclosed the specific milestones, but the structure mirrors the increasingly common pattern in late-stage AI deals: capital that converts only when the recipient hits revenue, model-capability, or capacity thresholds, giving the funder optionality and the founders a forcing function.

What to watch

• The TPU narrative. Five gigawatts of Anthropic workloads on TPUs is the strongest validation Google's silicon team has ever received. Expect Google Cloud to lean on it relentlessly in enterprise sales.
• Amazon's response. AWS is now the junior partner in compute terms, despite holding a larger reported equity position. Watch for matching infrastructure commitments and possibly a re-pricing of its own preferred terms.
• Antitrust posture. Three of the four largest U.S. tech companies are now financially intertwined with two of the four leading frontier AI labs. The pattern is increasingly difficult for regulators in Washington and Brussels to ignore.
• The IPO clock. If Anthropic does file in October, today's deal will be Exhibit A in the prospectus — both as the largest single commitment on the cap table and as the clearest signal that the company's compute roadmap is funded through the end of the decade.

For now, the message from Mountain View is unambiguous. Whatever Google thinks about the long-run competitive threat from Claude, it would rather own a piece of the upside than cede the field. Forty billion dollars is a lot of money to spend hedging your own bet — and a lot of money to take from a company whose chatbot you are trying to beat.