Happy Sunday.

The holiday-shortened week handed Wall Street the kind of macro print that usually steals the room: the June jobs report showed payrolls up just +57,000, unemployment at 4.2%, and April-May payrolls revised down by a combined 74,000. But the deeper market story was not the labor tape. It was the question now hanging over the AI buildout: What happens when the companies buying all the compute start trying to rent some of it back out?

That is the subject of today’s deep dive.

Artificial Intelligence

For two years, the AI trade ran on a simple rule: whoever controlled the GPUs controlled the tollbooth.

That rule still matters. But this week, it got messier. Meta Platforms (META) is building a cloud business to sell excess AI computing capacity, according to a Reuters writeup of Bloomberg’s reporting, while Nvidia (NVDA) is backstopping some younger AI clouds by agreeing to rent back unused GPUs in exchange for a cut of cloud revenue. Put those together and the hottest market in tech starts to look less like a clean shortage and more like a real estate cycle: everyone needs space, everyone is building space, and suddenly everyone is asking who is stuck with the lease.

Meta is the useful case study because it has been one of the AI boom’s most aggressive tenants. The company raised its 2026 capital-spending outlook in April to $125 billion to $145 billion, up from a prior $115 billion to $135 billion range, citing higher component pricing and more data-center costs for future capacity. In the same quarter, Meta reported $56.31 billion in revenue, up 33% year over year, and $19.84 billion of capital expenditures including principal payments on finance leases.

Management’s public line has been that this is not optional wallpaper for a shiny AI lobby. Susan Li, Meta’s chief financial officer, said on the Q1 call that the company has “continued to underestimate our compute needs” even as it ramps capacity, and that compute will become more central to the business.

The reported cloud plan does not prove Meta has overbuilt. The Reuters writeup noted the plans are still in development and could change. But it does change the market’s question. Investors are no longer only asking whether Meta can get enough compute. They are asking whether compute can become a product, a hedge, or both.

That matters because Meta is not just another hyperscaler in a hoodie. It is also a major customer of the companies Wall Street has treated as pure-play AI landlords.

CoreWeave (CRWV) announced in April an expanded agreement to provide Meta with AI cloud capacity through December 2032 for approximately $21 billion. Nebius (NBIS) announced in March a five-year agreement under which it will provide $12 billion of dedicated capacity to Meta starting in early 2027, with Meta also committed to buy additional available compute capacity up to a total of $15 billion; Nebius put the total contract value at up to roughly $27 billion.

Those contracts are validation until they start looking like concentration risk. Reuters reported that CoreWeave and Nebius fell 10.8% and 12.4%, respectively, on concerns Meta’s move could both reduce future spending on their services and add a new competitor. That is the unpleasant arithmetic of the AI sublet: the customer that fills your building can eventually decide to build one next door.

Nvidia’s new wrinkle makes the cycle even more circular. Data Center Dynamics reported that Nvidia is supporting neocloud customers by acting as a financial backstop in exchange for a revenue share, including an arrangement to rent back unused GPUs at a fixed rate. The first adopters include Firmus and Sharon AI, with Firmus deploying 170,000 GPUs in Batam, Indonesia, and Sharon AI deploying 40,000 GB300 GPUs, according to the same report.

For Nvidia, that is elegant. It sells the chips, helps the buyer finance or derisk capacity, and then participates if that capacity generates cloud revenue. For the market, it is a little more complicated. If the chipmaker is also the backstop, the AI infrastructure trade starts to depend not just on demand for models, but on utilization rates, depreciation curves, customer quality, and who carries the downside when capacity sits idle.

That is why the week felt like a turn in the tape. The AI boom is not suddenly over because Meta may sell compute or Nvidia may share in cloud revenue. If anything, both moves show how large the market has become. But they also show that the easy part of the story is ending. The first phase rewarded scarcity. The next phase will reward occupancy.

This is the question that will follow the sector into the second half: can AI infrastructure earn its rent fast enough to justify the buildout?

The spending base is enormous. A Reuters-cited Bridgewater estimate put expected 2026 AI-related investment by Alphabet, Amazon, Meta, and Microsoft at about $650 billion. Meta alone is now guiding to as much as 145 billion of 2026 capital expenditures including finance-lease principal payments.

The bull case says this is how the next computing platform gets built: spend first, monetize later, and let scale become the moat. The bear case says the moat may get crowded, especially if hyperscalers, neoclouds, chipmakers, and model labs all end up selling access to the same scarce resource at the same time.

The best answer is probably less dramatic and more Wall Street: the winners will be the ones with real demand, high utilization, cheap financing, and customers who keep showing up after the hype cycle leaves the lobby. The losers will be the ones that confuse a signed lease with a full building.