Nebius Lost $12 Billion in a Day. Meta's Cloud Plan Explains Why

Nebius Group closed at $229.18 on Wednesday, July 1, 2026 — down 17% in a single session, the worst one-day loss in the AI-infrastructure complex on a day full of bad ones. That is roughly $11.9 billion of market value, gone between the opening bell and the close. No contract was canceled. No guidance was cut. No customer walked.

What happened was a headline. Bloomberg reported that Meta is planning a cloud business: selling access to its AI computing power, and to its models, for outside developers — a direct lane change into territory held by Amazon Web Services, Microsoft Azure and Google Cloud. Meta has not confirmed a product, a name or a timeline. For Nebius, none of that mattered. The customer that agreed in March to buy as much as $27 billion of Nebius capacity over five years just signaled it may one day rent out chips instead of renting them.

The violence of the reaction makes more sense than it first appears, and the reason sits in the fine print of a contract signed 107 days earlier.

What Bloomberg reported, and why it landed so hard

The Bloomberg report is thin on specifics: Meta is exploring ways to commercialize its AI infrastructure, offering customers raw computing capacity as well as hosted access to AI models. There are two ways to read that, and neither is friendly to the companies supplying Meta.

The first reading is that Meta expects to have surplus — after committing tens of billions of dollars to outside capacity, it now anticipates more compute than its own models need. The second is platform ambition: an AWS-style business built on infrastructure Meta already owns. Either way, the assumption that carried neocloud stocks through the past year — Meta as a bottomless, price-insensitive buyer — took direct damage.

The scale of that buying spree is the context that matters. Meta signed a $10 billion deal with Google Cloud in August 2025. It expanded its CoreWeave commitment to $21 billion through 2032 in April. It has poured money into its own data centers, including the $27 billion Louisiana build with Blue Owl. A company that assembled that much capacity, telling reporters it may resell the extra, is a different kind of tenant than the one everyone underwrote.

CoreWeave and Nebius both slid within minutes of the report and kept falling into the close. Meta shares rose 8.8% the same day — investors handed Meta credit for exactly the optionality they subtracted from its suppliers.

The March contract that changed meaning overnight

The agreement Nebius announced on March 16 has two parts, and the distinction between them is the whole story of Wednesday's selloff.

The first part is conventional: Nebius will deliver $12 billion of dedicated capacity to Meta over five years, built on one of the first large-scale deployments of NVIDIA's Vera Rubin platform, with delivery starting in early 2027. That tranche is firm, and nothing in the Bloomberg report touches it.

The second part is the one that just changed meaning. Meta committed to purchase up to $15 billion of additional compute across certain upcoming Nebius clusters — but, per Nebius's own announcement, "Nebius currently intends to sell this capacity to third-party customers of its AI cloud business, with remaining capacity to be purchased by Meta."

Read that slowly. It is demand insurance. Nebius builds gigawatt-scale clusters, tries to fill them with enterprises and AI startups, and whatever doesn't sell, Meta absorbs. It was a structure that let Nebius borrow and build aggressively, because the downside scenario — empty racks — had a Fortune-10 buyer attached to it.

Insurance like that works only while the insurer stays out of the market. A Meta that sells its own compute is a Meta far less likely to need someone else's remainder capacity in 2028. Worse, the third-party customers Nebius intends to sell that $15 billion of capacity to are the same developers a Meta cloud business would court. One report put both sides of the tranche at risk at once: the backstop got weaker, and the demand it was backstopping got a new competitor.

There is history here, too. The March agreement was already the second between the companies — Nebius disclosed an initial $3 billion, five-year Meta deal in November 2025 and delivered its tranches on schedule. Meta kept coming back for more. That track record is precisely why the market treated the $27 billion figure as bankable rather than aspirational — and why repricing it hurt this much.

The arithmetic of an $11.9 billion haircut

Nebius has 253,898,194 shares outstanding, per its most recent quarterly release. Multiply by Wednesday's $46.99 decline and the market removed about $11.9 billion of value — close to 80% of that entire $15 billion option tranche, erased over one trading session on a report about a business that does not yet exist.

That is not as irrational as it sounds. At roughly $58 billion, Nebius still trades at 17 to 19 times its own 2026 revenue guidance. A multiple like that doesn't price what the company bills this year; it prices the assumption that $27 billion contracts keep arriving and keep renewing. Cut the probability on the optional half of the newest one, trim the odds on the renewal after that, and $12 billion of present value disappears without a single invoice changing.

Perspective helps, though. Nebius had gained 219% from its first close of 2026 to its June 18 record of $286.69. Even after Wednesday, the stock is up about 155% for the year — the selloff took it back roughly five weeks, to late-May prices, not to some crisis level.

The pain also wasn't Nebius's alone. CoreWeave fell 13.9% on the same logic applied to its own $21 billion Meta relationship and heavier debt load, and the selling bled outward into memory and equipment names that feed the same buildout. When one tenant anchors a whole district, a rumor about that tenant moves every landlord's price.

The business underneath didn't change on Wednesday

None of Wednesday's damage shows up in Nebius's operating results, which is exactly what makes the setup interesting rather than terminal.

The first quarter, reported May 13, was close to flawless: group revenue of $399.0 million, up 684% year over year, with the core AI cloud growing 841% year over year and 82% sequentially. Adjusted EBITDA turned meaningfully positive at $129.5 million. Operating cash flow came in at $2.3 billion, swollen by customer prepayments — the quiet benefit of those anchor contracts.

The shareholder letter filed alongside it showed annualized run-rate revenue of $1.9 billion exiting March, up from $1.25 billion at year-end, against a reiterated full-year target of $7 billion to $9 billion. Contracted power passed 3.5 gigawatts, management raised its year-end target to more than 4 gigawatts, over 75% of that capacity is owned rather than leased, and a second U.S. gigawatt-scale site in Pennsylvania joined the Missouri build. Cash stood at $9.3 billion. Live pricing and fundamentals for the stock sit on TECHi's Nebius quote page.

Growth like that is why the multiple got paid. Concentration is its mirror image. Add the announced deals — Microsoft at $17.4 billion, Meta at up to $30 billion across two agreements — and Nebius carries contracted future revenue more than ten times its current-year guidance, held overwhelmingly by two logos. A stock priced off that book is, functionally, a bet on the intentions of two counterparties. Intentions are the one thing that changed this week.

The financing loop is the transmission channel

There is a second-order effect that deserves more attention than it got in Wednesday's coverage: how a headline about Meta's product roadmap eventually touches Nebius's cost of capital.

Nebius funds its buildout from three sources — customer prepayments, convertible debt (it raised $3.75 billion in notes in March, days after the Meta announcement), and borrowing secured against the contracts themselves. The Microsoft filing spelled the model out: debt issued against the agreement "at terms enhanced by the credit quality of the counterparty."

Meta's credit quality did not move an inch on Wednesday. Its incentives did. A lender pricing the next facility against Vera Rubin clusters — clusters whose remainder capacity is supposed to flow to Meta — now has to consider that the counterparty may be marketing its own surplus compute into the same market by the time those clusters energize. The $6.3 billion Nebius raised in the first quarter, including NVIDIA's $2 billion equity investment, bought a long runway. It did not remove the dependence; it collateralized it.

Three ways this plays out

The scenarios below differ on one variable that has nothing to do with Meta's press office: whether AI compute is scarce or abundant in 2027, when Nebius's dedicated Meta capacity and its Pennsylvania and Missouri phases all come online. Scarcity makes the Bloomberg report a footnote. Abundance makes it a preview.

What to watch from here

Meta reports second-quarter earnings in late July; any capex commentary that references external capacity, resale or hosting revenue moves this story from report to plan. Nebius reports in August, and the single most telling disclosure will be progress selling option-tranche capacity to customers that are not Meta — every enterprise dollar signed against those clusters makes the backstop, and Wednesday's fear, less relevant. Between the two, watch whether Bloomberg's reporting hardens into a named product with a launch window.

The tape will keep treating CoreWeave as the read-across trade: more Meta exposure, more leverage, same question. Nebius enters that comparison with $9.3 billion of cash, positive EBITDA and the industry's stranger, safer contract structure — a backstop that may now matter less than the momentum of the business it was designed to protect.