Original author: Ada, Deep Tide TechFlow

On May 5, according to The Information, Anthropic committed to paying Google Cloud $200 billion over the next five years.

This multi-year agreement, starting in 2027, will account for over 40% of Google Cloud's revenue backlog—a metric that reflects contractual commitments from enterprise customers.

An AI company that didn't even exist five years ago has, with a single contract, consumed nearly half of Google Cloud's future revenue.

On the day the news broke, Alphabet's stock rose 2% in after-hours trading.

But another number is more intriguing. Alphabet simultaneously invested up to $40 billion back into Anthropic.

The money flows out of Google's accounts, circulates, and returns to Google's accounts, with just an additional line item reading "Anthropic computing expenditure" in the accounting books.

So, is this the largest cloud computing order in history, or the most elegant financial magic trick ever performed?

An "Exclusive Commitment" Not Just for Google

To understand the essence of this deal, first look at a set of data points that are not isolated.

On April 20, Anthropic announced an expanded collaboration with Amazon, committing to spend over $100 billion on AWS technology over the next 10 years in exchange for up to 5 gigawatts of computing power. In return, Amazon added up to $25 billion to its existing $8 billion investment.

And in November last year, Microsoft agreed to invest up to $5 billion in Anthropic, with Anthropic committing to purchase $30 billion in Azure computing power.

So, Google: invests $40 billion, receives $200 billion. Amazon: invests $33 billion, receives over $100 billion. Microsoft: invests $5 billion, receives $30 billion.

The three cloud giants together shell out about $78 billion, securing $330 billion in "contractual commitments," a net book inflow of $250 billion.

The essence of this game is washing capital expenditure as revenue. Investments in Anthropic are recorded under investing cash flow, while the computing fees paid by Anthropic count as main operating income. The same money goes out of one pocket and into the other, adding a pretty backlog to the financial statements.

While Alphabet pumps money into Anthropic, it books Anthropic's computing purchases as future revenue, creating a self-reinforcing closed loop for the AI infrastructure boom.

Wall Street is the real winner in this game. As long as the backlog numbers are large enough, the P/E ratio can hold up.

An Advanced Version of the Flywheel

The story of Strategy buying the top is not over yet; the AI circle has just magnified the same flywheel a thousand times.

Strategy's logic is to sell stock to raise money, buy Bitcoin; Bitcoin's rise boosts market cap; sell more stock, buy more Bitcoin.

The cloud vendors' logic is to invest in AI companies; AI companies pay for computing power; revenue grows; stock prices rise; capital markets double down; continue investing in AI companies.

The difference is that Bitcoin is a scarce asset, with each coin corresponding to real supply on the chain. Computing power is not. The "multi-gigawatt TPU capacity" that won't come online until 2027 doesn't even have racks installed today.

In other words, a significant portion of the $200 billion is Anthropic's advance commitment to purchase chips that haven't been manufactured yet, and Google uses this commitment to convince capital markets.

Isn't this a forward contract? The difference is that commodity futures have delivery dates and margin, but this contract doesn't. What if Anthropic can't pay by 2027? Who bears the default cost?

Not Google. It has already written the backlog into its earnings call PPT. In its April 29 earnings call, Alphabet disclosed that Google Cloud revenue grew 63% year-over-year, exceeding $20 billion, and cloud business backlog reached approximately $462 billion. This number underpins Alphabet's current market cap.

Nor Anthropic. It just needs to keep raising funds; its next valuation round is still climbing.

Those footing the bill in the end will likely be retail investors who thought they were buying into the "AI picks and shovels" story.

$5 Billion Leveraging $330 Billion

Does Anthropic's own size match these numbers?

According to media reports, Anthropic's annualized revenue grew from $1 billion to $5 billion in 2025.

A company with only $5 billion in annualized revenue signed contracts worth $200 billion over 5 years, $100 billion over 10 years, plus another $30 billion, totaling $330 billion across three contracts.

Even if Anthropic's revenue increased tenfold again, it couldn't accumulate $330 billion over 5 years.

So where does the money come from?

There's only one path: continue fundraising.

And the biggest potential investors are precisely these three cloud vendors themselves.

That's the whole secret of the cycle. Anthropic doesn't actually need to make money; it just needs to maintain a state of "always fundraising," using each new round of funds as the next year's computing bill. As fundraising valuations rise, it can raise even more.

Sound familiar?

Strategy. It doesn't need Bitcoin to generate cash flow either; it just needs to maintain a state of "always being able to issue stocks and bonds." The only difference is that Strategy's balance sheet holds Bitcoin, an asset with a globally public price.

The valuation logic for AI companies now closely resembles that of SaaS companies in 2021. Back then, everyone competed on ARR; today, it's computing power commitments. Essentially, both use the future to discount the present. The only question is whether the future will deliver."

What is OpenAI Doing?

In the same 8-K filing where Amazon increased its stake in Anthropic, OpenAI also committed to consuming approximately 2 gigawatts of Trainium computing power via AWS infrastructure, starting to ramp up from 2027.

Two months ago, Amazon invested $50 billion in OpenAI and signed a $100 billion cloud computing contract.

The script is exactly the same.

In other words, three major cloud providers, two major model companies—five players playing the same game multiple times. Each time accompanied by headlines like "largest in history," "strategic partnership," "computing power revolution."

Behind each time is the same money circulating.

So, who will be the first to stop?

Not the cloud vendors; their current market caps rely precisely on this narrative. Alphabet raised its 2026 capital expenditure guidance to up to $190 billion. An expenditure of this scale must be "hedged" into revenue by Anthropic and OpenAI; otherwise, Wall Street wouldn't approve.

Nor the model companies; stopping means failing to secure the next funding round, and thus death.

The first to be shown the door will likely be second-tier players that didn't pick the right side.

Will the Music Stop?

The fragility of it all is hidden in the word "delivery."

TPU capacity comes online in 2027. If Claude's commercialization doesn't keep pace with the expansion of computing power by then, how will Anthropic digest this $200 billion?

If a contract is renegotiated, reduced, or spread out, Google Cloud's $462 billion backlog table will immediately be exposed.

But today, no one wants to be the first to burst the bubble. CFOs are writing guidance, analysts are issuing buy ratings, CEOs are choosing their words carefully on earnings calls. Everyone is betting that they'll be standing closest to a chair when the music stops.

The issue now isn't whether there is a bubble, but how to dismantle it. Everyone knows this is circular trading, but everyone also knows that as long as the AI story continues, no one dares to short the backlog.

Contracts are on paper, money circulates among three companies, and valuations spin between primary and secondary markets. Everyone received a "promise for the future," and everyone treats this promise as a "current asset."

Until a future day when a company's earnings fall short of expectations. At that moment, $200 billion will suddenly have another name: contingent liability.

But until that day arrives, the party continues.