CoreWeave (NASDAQ: CRWV) and Nebius (NASDAQ: NBIS) have each shed roughly a fifth of their market value over the past month, hit harder than almost any corner of the AI trade after three of the industry’s most prominent chief executives called for a deliberate slowdown in how fast AI is developed. The selling that began on September 14 has now given way to a stranger picture, because on Wednesday the neocloud names split three ways rather than moving together. Nebius jumped about 10% to roughly $229 after telling customers it would raise GPU rental rates, IREN (NASDAQ: IREN) rose about 6% to near $45 on the read-through, and CoreWeave slipped about 2% to around $82, according to on-chain data from TradingView.
The neoclouds are the leveraged GPU landlords whose business model assumes AI demand keeps compounding, and they are the names most exposed if the industry actually slows down, the risk FinanceFeeds mapped across the sector in its look at what a deliberate slowdown does to the $700 billion AI-capex trade. Days after being warned that slowing demand threatens their model, one of them raised prices, and the market rewarded the price-setter while punishing the most indebted operator.
Nebius fell with the neocloud group before rebounding about 10% on its GPU price increase. Source: TradingView.The Slowdown Call That Started the Neocloud Selloff
Anthropic chief executive Dario Amodei published an essay on September 12, “We Must Pace the Frontier,” arguing that AI companies should deliberately slow how fast they improve model capabilities, and within two days OpenAI’s Sam Altman and xAI’s Elon Musk had publicly agreed. Markets read the alignment of three fierce competitors on the same message as a genuine risk to the demand curve underpinning the entire AI buildout, and the selling spread from the labs to the infrastructure that serves them.
The neoclouds sit at the sharpest end of that risk, as these companies rent out GPU capacity on multi-year, take-or-pay contracts, funding enormous upfront buildouts with debt borrowed against the chips and the future revenue those contracts promise. The model works beautifully while demand compounds and breaks down if it stalls, which is why a call to pace the frontier landed hardest on the balance sheets built for a world where it never does.
The September 14 Selloff, Name by Name
The Monday sell-off broke a pattern that had held for most of the cycle, in which AI infrastructure stocks rose and fell together on demand headlines. CoreWeave fell about 7%, Nebius 8%, Applied Digital 6%, DigitalOcean 5% and IREN 4%, as Seeking Alpha reported, while the Nasdaq 100 dropped as much as 1.8% intraday. The damage ran through the wider AI supply chain, with networking makers Lumentum, Ciena and Coherent each falling 7% to 8% and chipmakers Marvell and Arm down 8%, though the move arrived alongside a firm oil spike and a hawkish Federal Reserve the same week, so no single catalyst explains the whole of it.
What made September 14 different was that one bank refused to treat the group as interchangeable. Bernstein Société Générale did not just flag sector-wide risk, it named the single stock it judged most exposed, and that call is why CoreWeave led the decline.
CoreWeave fell on Bernstein’s warning and lagged the group’s rebound, trading near $82. Source: TradingView.Why Bernstein Singled Out CoreWeave
Bernstein analyst Madison Rezaei estimated that about 70% of the 488-gigawatt US data-center pipeline sits in rural or Tier 3 markets built for latency-insensitive training workloads, the exact capacity most at risk if training demand cools. CoreWeave has roughly 25% of its active power and 74% of its contracted power in those markets, as reported by Yahoo Finance, which makes it the most exposed name in Bernstein’s coverage. The firm rates CoreWeave Underperform with a $74 price target, well below the roughly $139 average across Wall Street.
The balance sheet is what turns exposure into vulnerability. CoreWeave carried about $35.6 billion in debt as of the end of June, and its net interest expense more than doubled to $640 million in the second quarter, even as its revenue backlog swelled past $104 billion, a figure that converts to cash only over years, according to its SEC filing and its own results.
Nebius funds the same kind of buildout very differently, with roughly 70% of its second-quarter deals carrying customer prepayments that management expects to bring in more than $9 billion of upfront funding in 2026, covering 50% to 60% of the associated capex, as chief executive Arkady Volozh told investors, which is the structural reason it can raise prices from a position of strength while CoreWeave carries the leverage. The company has pushed back on the demand concern, with chief executive Michael Intrator saying at a recent conference that CoreWeave is “struggling to meet demand every day,” a claim FinanceFeeds weighed against the numbers in its AI-data-center bear case built on a $99 billion backlog and $50 billion of debt.
CoreWeave and Nebius each fell about a fifth over the month, while IREN held up. Data: TradingView, as of September 17 · Chart: FinanceFeeds.Investor Takeaway
The neoclouds are no longer one trade, and CoreWeave is the name carrying the most balance-sheet risk into any demand slowdown, with about $35 billion of debt and interest costs that doubled year over year.
The Price Hike That Split the Neocloud Trade
Nebius notified customers of higher on-demand GPU rental rates effective October 1, and the market treated it as the party setting the new price card, sending NBIS up about 10%, per 24/7 Wall St. IREN rose about 6% on the read-through that scarce GPU capacity can be re-rented at higher hourly rates, while CoreWeave slipped about 2%, weighed by its debt and by contracts locked at fixed prices it cannot raise. That the Global X Data Center ETF lagged the Nasdaq confirmed this was a neocloud-specific repricing rather than a broad infrastructure rally.
Days after Bernstein warned that a slowdown could erode neocloud demand and pricing, a neocloud raised prices, the opposite of what a cooling market would allow. Investor Michael Burry, who has pushed back on the slowdown calls, has argued the bear case can fail even if the safety warnings are right, because frontier labs redirecting spending toward inference and deployment would keep contracted neocloud capacity valuable. High short interest across the group, roughly 18% to 23% of Nebius’s float and about 24% of IREN’s, adds a squeeze dynamic to any move that reframes demand as intact.
What Would Confirm or Kill the Neocloud Slowdown Thesis
The next confirmations are dated and specific. Micron reports earnings on September 30 after the close, and its read on memory demand will signal whether the AI hardware chain is still tightening or starting to loosen, the same pressure that drove the memory names lower when SanDisk fell hardest among memory stocks on the slowdown call. Nebius’s price increase takes effect October 1, and whether it holds or draws customer pushback will test the pricing power the whole trade now hinges on.
Beyond the calendar sits the question Bernstein could not answer, whether AI demand slows outright or simply shifts from training toward inference, which would spare the contracted capacity the bank flagged. The scenario ranges capture how wide the outcomes are, from the Nebius $390 bull versus $130 bear case to the CoreWeave $192 bull, $113 base and $47 bear breakdown and the IREN $92 bull versus $16 bear range. Wednesday’s split showed the market beginning to sort the group by balance sheet and pricing power, and the coming prints will decide whether that sorting was the start of a repricing or a pause in the selloff.
Investor Takeaway
Nebius’s October 1 price increase is the live test of the whole thesis, because a slowing market cannot support higher rents, so whether the hike holds tells investors more than any analyst note.
