Google Buys a Rival It Cannot Control
What happened
Google committed up to $40 billion to Anthropic on April 24, the largest single AI investment ever disclosed. The deal starts with $10 billion in cash at a $350 billion valuation, with $30 billion more contingent on performance targets. It expands an existing cloud partnership, giving Anthropic access to 5 gigawatts of Google TPU compute by 2027. The announcement came days after Amazon separately committed $25 billion to the same company, meaning two hyperscalers are now collectively pledging $65 billion to an AI startup that directly competes with their own AI products.
Google is not investing in Anthropic because it believes in Anthropic. It is investing because Gemini failed to become the default choice for serious AI work, and the only way to stay in the room where Claude runs is to own a piece of the house.
Prediction Markets
Prices as of 2026-04-26 — the analysis was written against these odds
The Hidden Bet
This is a financial investment with market-rate returns
The $30B tranche is tied to performance milestones Google helps define through its compute supply. Google controls the infrastructure Anthropic runs on, can set pricing, and influences which milestones are achievable. That is operational leverage, not passive capital.
Antitrust regulators will treat this like a normal investment because Google does not acquire a board seat or majority control
The EU is simultaneously demanding Google open Android to Claude, while Google funds Claude's development and hosts it on Google Cloud. The FTC has already shown it will challenge structural influence without formal acquisition; the Microsoft-OpenAI structure was the template, and regulators said they would not be surprised by it twice.
Anthropic's independence is structurally protected by its public benefit corporation status
Public benefit corporation status constrains shareholder primacy but not infrastructure dependency. If 90% of Claude's compute runs on Google TPUs and Google Cloud, independence on paper does not translate to independence in practice. Anthropic can resist a bad acquisition offer but cannot easily resist a compute price increase.
The Real Disagreement
The fork is whether this consolidation is inevitable or preventable. One side says: frontier AI requires capital and compute at hyperscaler scale, so every serious lab eventually becomes a subsidiary of one; fighting it is performative. The other side says: the compute dependency is artificial, sustained by hyperscaler pricing power, and regulators can break it by forcing compute interoperability just as the EU broke telecom monopolies. The second position has more analytical force, but the first one has more money. I would lean toward the inevitability argument in the near term and toward the regulatory disruption argument over the next decade, once the damage is documented.
What No One Is Saying
Anthropic is now co-dependent on two companies, Google and Amazon, who are competitors with each other. If those two hyperscalers ever diverge on technical standards or pricing, Anthropic becomes the party forced to choose a side, at which point its vaunted neutrality collapses instantly.
Who Pays
Smaller AI labs and independent developers
Already visible in pricing; structural lock-in solidifies by 2027 when the 5GW TPU build-out completes
When Google and Amazon are the compute substrate for the frontier labs they fund, independent labs cannot compete on price or access. The effective compute market becomes a three-entity structure: Microsoft-OpenAI, Google-Anthropic, and Amazon-Anthropic. Everyone else borrows from one of these stacks, on their terms.
Enterprise customers building on Claude
Medium-term, as performance milestone structures mature and Google's influence over product direction becomes visible in release patterns
Claude's independence is part of its brand premium. If Anthropic's safety decisions, model access policies, or pricing are increasingly shaped by Google's infrastructure needs, the enterprise trust premium erodes. Customers who paid a safety premium may find themselves on a Google cloud product with Anthropic branding.
EU tech regulation project
Immediate; the contradiction is already on the table in Brussels
The EU is demanding Google open Android to rival AI assistants in the same month Google commits $40B to one of them. If the Commission grants Android access to Claude as a concession while Google quietly controls Claude's infrastructure, the regulation produces the appearance of competition with none of the substance.
Scenarios
Antitrust intervention
FTC or EU opens formal investigation into the combined Google-Amazon-Anthropic investment structure. Google is required to divest either compute control or ownership stake. Anthropic raises independent infrastructure funding.
Signal An FTC second request on the deal, or EU opening a formal Phase 2 inquiry into the cloud dependency structure rather than just the investment percentage
Silent consolidation
Regulators investigate but do not act before the 5GW compute expansion locks in Anthropic's infrastructure dependency. Anthropic remains nominally independent while Google controls its operating costs. The market treats this as the new normal.
Signal Google's next earnings call references Anthropic TPU utilization in the same breath as Google Cloud revenue growth
Anthropic outgrows both
Anthropic's $30B run-rate continues accelerating. By 2027, it has enough revenue to self-fund significant infrastructure. It negotiates lower compute rates using a credible threat to build or buy its own capacity. The performance milestones are met early, removing the contingent leverage.
Signal Anthropic announces data center construction or a third-party compute agreement outside Google and Amazon clouds
What Would Change This
If Anthropic negotiates a public compute interoperability agreement guaranteeing it can migrate workloads to competing infrastructure without pricing penalties, the infrastructure dependency argument weakens substantially. That would suggest this is genuinely an investment rather than a capture.
Related
Google Just Paid $40 Billion for the AI Model That's Beating Its Own
powerGoogle Just Paid $40 Billion for a Competitor It Also Supplies
powerAnthropic Built a Cybersecurity Cartel. They're Calling It a Consortium.
powerOpenAI and Microsoft End Exclusivity: The Most Consequential Divorce in Tech Has No Villain