Cloud Computing Costs Shift as Meta Courts Anthropic
Anthropic wants to rent Meta’s spare hardware, and the arrangement under discussion could reach $10 billion across two years, according to CNBC’s July 17 report citing a person close to the talks. Anthropic reportedly raised the idea in June. Nothing is locked, payment would be spread across monthly installments, and both parties would keep the option to leave.
The dollar figure is not the story for founders. The seller is. Meta has functioned as a compute buyer for years, and a lease agreement would flip it into a vendor, reshaping the market that ultimately prices your AI infrastructure funding and your monthly inference bill.
What Is Actually on the Table
Start with the caveats. No agreement exists, Anthropic declined to comment, and Meta stayed silent when Reuters asked. The $10 billion should be read as the outer edge of a proposal rather than a signed number.
Size becomes clearer with a comparison. Anthropic committed to a much larger arrangement with SpaceX back in May, and the Meta proposal would be a fraction of it.
| Counterparty | Reported value | Term | Status |
|---|---|---|---|
| SpaceX | $45 billion | Three years | Signed in May |
| Meta | Up to $10 billion | Two years | In talks |
Meta’s reasoning is easy enough to follow. Advertising still funds almost everything the company does, and leasing idle capacity would open a second income stream while putting it head to head with specialist providers like CoreWeave and Nebius.
A Buyer Turning Into a Seller
Early-stage teams almost never sit across the table from a hyperscaler. The consequences still reach them. When fresh capacity arrives at the top of the market, availability improves and prices soften as the effect works downward.
Scarcity travels the same path in reverse. Large multi-year commitments tie up the newest hardware first, which leaves smaller accounts queuing longer and paying more. Anyone whose inference costs doubled last year should follow supply news with the same attention they give model launches.
The defensive move is optionality. A team that can shift lighter workloads onto local AI models negotiates from a stronger position, simply because leaving is a credible threat.
Where the AI Spending Curve Sits
Context for the deal size: Gartner expects global AI spending to pass $2.5 trillion in 2026, up 47% year over year. Against that backdrop, $10 billion is meaningful but hardly extraordinary.
Meta has also hinted at this move publicly. Speaking to shareholders in May, Mark Zuckerberg described cloud services as something the company was actively weighing, and mentioned that other firms kept approaching Meta about buying model access or surplus capacity.
Seen that way, July’s reporting confirms a direction rather than revealing one. The open question is whether Meta can spare enough hardware to make the economics work while its own models keep consuming more of it.
Questions to Put to Your Cloud Provider
Vendors rarely advertise their flexibility, so raise it yourself. Ask how your rate responds if usage lands well above or well below the forecast you committed to, because that clause decides whether growth rewards or punishes you.
Portability deserves a direct question too. Find out whether you can retrieve fine-tuned weights, embeddings and logs on your own, without opening a ticket and waiting a fortnight. Hesitation on that point reveals exactly how much leverage you will have at renewal.
Then establish your position in the queue. Contracts often push smaller customers behind enterprise accounts whenever supply tightens, and knowing that in advance lets you schedule launches around what is realistically available.
Three Markers Worth Tracking
Housekeeping first. Pull your compute contracts, note the commitment length and the exit terms, then calculate what you spend per active user each month. Without that figure you cannot tell whether a price cut helps the business or just subsidizes waste.
Few companies do this well. Budgets have climbed at plenty of firms with no matching evidence of returns, which explains the disappointing AI ROI reported across large enterprises. A lower unit price will never rescue a feature nobody wanted.
Over the coming quarter, three things will tell you where costs are heading. Does Meta formally launch a cloud product, do specialist providers adjust pricing in response, and do smaller customers start reporting shorter waits for capacity? Those answers matter far more to your 2027 budget than any headline valuation.
