Google’s story in AI over the past few years hides a contradiction that’s hard to square at first glance.

On one side it is being sued by the U.S. Department of Justice for a search monopoly; on another (per a leaked memo) it has admitted that its own Gemini trails Anthropic’s Claude at writing code. And what does it do? It puts up to $40 billion into the very company behind Claude. In effect, it spends a fortune to make its strongest rival even stronger. If you want to get to know Google first, see What is Google.

This piece unpacks that contradiction: how big this money is, why Google deliberately wants no control, what its real play is, and why the DOJ is watching. Up front: this is explainer and information relay, not investment advice.

To set the tone in one line: this is not charity. The money goes out on the investment side and loops back through cloud orders and chip purchases — it’s a compute business dressed up as “feeding the enemy.”

How big the money is, and how it’s paid out

On April 24, 2026, Google announced an investment of up to $40 billion in Anthropic.

The structure splits into two parts: about $10 billion landed immediately, buying in at the then-valuation of roughly $350 billion; the remaining up to $30 billion will be paid out in tranches as Anthropic hits milestones. This isn’t Google’s first move. It started investing in Anthropic back in early 2023 — first about $300 million (for roughly a 10% stake), then a string of follow-ons, so that by the time this round was announced its cumulative holding was already in the low teens percent.

That scale would be staggering anywhere, all the more so when the target is a direct competitor.

Deliberately wanting no control

The most counterintuitive part of the design is that Google sank all this money in yet deliberately wants no say.

According to reports, Google capped its stake at 15% and took no voting rights, no board seat, not even an observer seat. There’s a very practical reason for this self-restraint: once a stake or degree of control crosses a certain line, it can trip the merger-review thresholds for antitrust and draw the regulatory magnifying glass. Keeping itself below that threshold is a design built to sidestep trouble.

Anthropic’s side has a layer of protection too. It uses a public-benefit-corporation (PBC) structure and has set up a long-term benefit trust that can step in if the company strays from its AI-safety mission. In other words, no matter who invested how much, no one gets operational control of Anthropic. What Google is buying is exposure and lock-in, not the driver’s seat.

The real play: compute

To make sense of this investment, you have to shift your gaze from “investment” to “compute.”

That $40 billion comes with compute commitments: Anthropic will make heavy use of in-house TPUs through Google Cloud, on the order of several gigawatts and chip counts in the millions. More crucial is the reverse cash flow — according to The Information, Anthropic has committed to buying as much as $200 billion of compute services from Google Cloud over the coming years, several times the size of Google’s investment. That $200 billion figure currently rests on a single media report, and neither Google nor Anthropic has formally confirmed it, so treat it as a strong but unconfirmed signal for now.

Connect that line and the logic becomes clear: Google takes a stake in exchange for a flagship customer willing to buy its own compute over the long term and in volume. For a company that has just commercialized its TPUs and is in a hurry to fill up chip capacity, locking in usage at Anthropic’s level is worth more than that low-teens stake. Money out, orders back — that’s the main thread.

On paper, it’s already making money

The paradox is that this “feeding the enemy” investment has already made Google a tidy sum on its books.

Anthropic’s valuation and revenue have grown explosively over the past two years. Its annualized revenue went from under $100 million in early 2024 all the way to about $30 billion by April 2026; its valuation neared $1 trillion after its latest private round in late May 2026, even surpassing OpenAI. The shares Google bought cheaply in the early days have risen in book value right along with it.

You can see this in the financials. A sizable chunk (around 40-something percent) of Alphabet’s net income in the first quarter of 2026 came from “unrealized” revaluation gains on equity investments — that is, the book appreciation of stakes like Anthropic as valuations rose. Two caveats here have to be spelled out: first, this is an unrealized paper figure, Anthropic has paid no cash to Google, and if the valuation falls the book value shrinks with it; second, this gain is calculated together with other holdings such as SpaceX, and how much Anthropic contributed on its own is not broken out and disclosed. Treating book appreciation as profit already in the bank would be a misread.

Not just Google: Amazon is betting too

This high-stakes bet isn’t Google playing alone.

Just days before Google’s move, Amazon also announced up to an additional $25 billion in Anthropic, likewise paired with compute commitments tied to its own Trainium chips and long-term cloud purchasing. The two cloud giants doubling down on the same AI company at once follow identical logic: use investment to lock in a flagship customer that can prop up your own compute platform. The difference is that Google’s contradiction is sharper, because it has Gemini going head-to-head with Claude, whereas Amazon’s ambitions for its own models are far smaller.

For Anthropic, Google in one hand and Amazon in the other means getting money and compute from both major clouds at once — and being able to play the two sides against each other. That’s a rather advantageous position.

The U.S. Department of Justice is watching

A relationship that is “investor and competitor” at once naturally can’t escape the eye of antitrust.

In Google’s search-monopoly case, the DOJ at one point proposed in its remedies to simply ban Google from holding stakes in any AI company. The proposal was later revised to allow Google to keep existing investments but require advance notice to the antitrust authorities before any future investment in an AI company. Interestingly, Anthropic itself stepped up to oppose this restriction in court, arguing that a prior-approval rule would actually hinder AI competition and harm rather than help the industry.

This line is still ongoing, with no final conclusion. This piece only describes the known claims and facts on each side and won’t predict how it will end. For the full status of Google’s antitrust case, see Where Google’s antitrust case stands.

The parts not yet laid bare

This deal has plenty of key points stuck behind confidentiality or single sources, so let’s flag them honestly:

  • The $200 billion in cloud purchases: comes from a single report by The Information; neither Google nor Anthropic has formally confirmed it.
  • The conditions on the $30 billion in milestones: what metrics and what timeline trigger the disbursements — all sides only call them “performance targets,” and the details are confidential.
  • Google’s exact stake: the “low teens percent” comes from media analysis and citations in court filings; neither Anthropic nor Alphabet has officially published a precise number.
  • How much cheaper TPUs are than Nvidia: the commonly cited 40-to-50-percent discount is an analyst estimate, varies with workload, and is not an official quote.

Penna’s take

Pull back from this investment and it’s actually a mirror, reflecting the real rules of this generation’s AI race: compute is the chip you bet with.

Google is willing to spend $40 billion — and even endure the awkwardness of “feeding the enemy” — because it has done the math: turning a rival into a major customer for its own compute is more worthwhile than starving that rival, and it conveniently turns a book profit along the way. This kind of multilayered relationship, where the competitor is at once a supplier, a customer, and an investment, will be the most common — and the least intuitive — structure in the AI industry over the next few years.

Further reading: What is Google, Even Google says it lags behind Claude, The cost moat of in-house TPUs.