Why Neocloud Stocks Fell in October 2026: Prices, Dates, Causes

Roughly 18% to 50% of their value in about three weeks, if you hold one of the listed GPU clouds or the bitcoin miners now leasing data centers to them. The market as a whole did not fall: the S&P 500 ETF ended October 23 within 0.3% of its September 30 close. So this was a sector selloff, not a market crash. The data below are Exent market data, daily closing prices, with October 23, 2026 as the last session of the October window (prices in this table and chart; the rest of the page reuses them). What happened next, to the November 5 close, is in the section below.

Scope: US-listed neoclouds (GPU clouds) and the data-center landlords grouped with them, as defined in what is a neocloud, compared with NVIDIA, Oracle and broad index funds. Excluded: private neoclouds and intraday prices. This page records what happened and why it may have happened; it does not say whether to buy, sell or hold any of these stocks.

Change from the September 30 close to the October 23 close. Averaged with equal weights, the three GPU clouds fell 36.5% and the seven landlords 31.2%. IREN and Core Scientific have no stock page in our dataset.

Most of the group fell further, and this time the market fell too. From the October 23 close to the November 5 close CoreWeave lost 19.5%, Nebius 24.2% and IREN 27.6%, against 3.5% for SPY and 8.1% for QQQ. Riot, up 17.8%, was the only stock in the table that rose. So the “sector selloff, not a market crash” description fits the October window; it does not fit November 5.

That day was a very large market drop: the S&P 500 fell 4.36%, its worst day since April 4, 2025, and the Nasdaq-100 fell 6.36%, its worst since March 16, 2020. The path was not a straight line. Most of these stocks rose through October 28, their highest close of this stretch, then fell; November 5 was the lowest close of the stretch for every stock here except Riot.

Closing prices, Exent market data. Averaged with equal weights, the three GPU clouds fell 23.8% since October 23 and the seven landlords 13.6%, Riot included. Measured from September 30, $10,000 in CoreWeave was worth about $5,011 on November 5 and needed a 99.5% gain to get back; the same $10,000 in SPY was worth about $9,611.

What the press said, which is attribution and not proof: CNBC reported that voters sided against data centers in 40 of 41 local ballot measures on November 3, and that Baird analysts called the results a “sentiment headwind” for neoclouds, which CNBC said “got whacked” on November 4 with CoreWeave down about 10%, on a day the S&P 500 rose. On November 5 CNBC tied the fall to concern that the AI buildout could slow, plus a weak forecast from Qualcomm. None of the company filings we read gives a cause for these share price moves.

As of November 11, 2026, eight of the group have reported the quarter to September 30; Nebius has not. CoreWeave reported after the close on November 10: revenue of $3.66 billion, up 168% from $1.37 billion a year earlier and up $1.09 billion from the June quarter, with a revenue backlog of about $132.5 billion against about $104 billion at June 30. TeraWulf reported before the open on November 6 and IREN after the close on November 5. Nebius has scheduled its results for November 12, before the open. Applied Digital’s fiscal year is different: its latest report, on October 7, covers the quarter to August 31.

Company releases and filings, unaudited; stock moves are closing prices from the same data as the tables above. “Backlog” is each company’s own definition (CoreWeave and IREN add estimated amounts under committed contracts to their performance obligations; TeraWulf’s figure is contracted lease revenue from its investor presentation), so the column is not comparable across rows; see backlog, RPO and ARR explained.

CoreWeave on the measure this page watches most, interest against operating income: operating income turned positive at $207 million, from a $49 million loss in the June quarter, but net interest expense was $846 million, about four times as much, and the company guided fourth-quarter interest expense to $1.05–1.15 billion. Net loss was $709 million. It spent $9.44 billion on property and equipment in the quarter, raised about $4.2 billion of convertible notes and set up an at-the-market program for up to 35 million shares. It also raised its 2026 revenue guidance to $13.0–13.3 billion. TeraWulf’s HPC lease revenue was $63.0 million, about 83% of the total; its operating loss was $88.2 million against $53.6 million of interest expense, and its $709.9 million net income came mainly from a non-cash change in the value of its warrants.

What the reports show on the three things this page watches. Interest against operating income: at CoreWeave, TeraWulf, Cipher, Riot, Hut 8 (before its digital-asset gains), IREN and Applied Digital, operating income did not cover interest in these quarters, while Core Scientific’s $72.4 million exceeded its $35.0 million of net interest. Financing: Hut 8 closed a $1.07 billion revolving facility, Galaxy completed a $3.5 billion senior secured notes offering, and on November 4 Google agreed to backstop certain of Anthropic’s lease obligations to TeraWulf. Delivered capacity: IREN says Microsoft requested a network design change across its Horizon deployments, and that delivery dates, fees and credits are under discussion (10-Q); Core Scientific was billing for 485 MW at September 30, up from 395 MW at June 30.

Results did not move these stocks in one direction. On the first session after their reports Galaxy rose 9.6% (Oct 27), Riot rose 9.6% and Hut 8 fell 23.2% (both Nov 4), Core Scientific fell 10.8% on a day the whole market dropped (Nov 5), and IREN rose 20.0% and TeraWulf 12.8% (both Nov 6, SPY +0.3%). CoreWeave’s first session after its report, November 11, had not closed when this page was updated. These are dates and closes, not causes. We found no lost customer or missed payment in the eight Q3 releases.

A drop in percent hides how much harder the way back is. Worked example with CoreWeave’s closing prices, reproducible from the table above:

The asymmetry is general: a 50% fall needs a 100% gain to recover (see loss recovery math). It is why the size of a position in stocks this volatile matters more than the choice between them, a point covered in the risks of investing in neoclouds.

The documented sequence centers on the bond market, not on a company filing. Every row is either a closing price from the table above or an event with a primary source or a named news report; the columns do not claim that the event caused the move.

The bond-market levels behind these dates, and how to read a 15% yield or a 1,000-point CDS spread without turning it into a default probability, are in CoreWeave’s debt explained. Gartner’s wording matters: “cease independent operations” includes being acquired, so it is not a forecast that 80% go bankrupt.

Three things are on the record. First, the cost of credit for this group rose sharply and in public: CoreWeave’s unsecured bonds fell to prices that imply yields near 15–16%, with a new closing high on October 13 according to an investor’s chart reported by CNBC, and protection against its default became more expensive through October 21. Second, financing news dominated company statements: a withdrawn IPO (Firmus), a convertible offering (Cipher), a 364-day bank facility (Nebius) and two CEO memos about funding (CoreWeave). Third, demand statements did not change: NVIDIA’s CEO, Nebius’s chief revenue officer and IREN’s co-CEO all told CNBC that customer demand was intact.

What is not on the record: none of these companies announced a cancelled contract, a missed debt payment or a covenant breach in October in the filings and releases cited here. During the October window the latest filed financials were those for the quarter to June 30 (Applied Digital: to August 31), which are unaudited quarterly figures. CoreWeave, for example, reported $640 million of net interest expense against a $49 million operating loss in that quarter and $10.6 billion of principal due by the end of 2027 (10-Q). Those numbers were known in August; what changed in October was the price investors demanded to finance them. Its third-quarter report on November 10 showed the same gap at a larger size: $846 million of net interest against $207 million of operating income.

Four explanations circulate. Each is plausible; none is proven by the data, and two are partly contradicted by it. Labels: “supported” means the timing and the cross-section fit; “partly” means some evidence points the other way; “not shown” means nothing new was disclosed in October.

One more reading fits the data and is often missed: much of the fall started before October. On September 30 Applied Digital was already 50.9% below its 12-month closing high, TeraWulf 48.9%, IREN 46.5%, Cipher 45.8% and Core Scientific 45.2%, most of them from peaks in May and June 2026. October was the third leg of a decline, not the first. For the debate behind the depreciation row, see how long AI GPUs last; for concentration, see who are CoreWeave’s customers.

Treasury yields fell over this window, weakening a simple rising-yield explanation; that does not identify what caused the stock declines. The rise in yields came earlier, in September, around the Fed’s increase and before the stocks broke. Daily par yields from the US Treasury:

Interpretation (ours): the rate level matters for companies that must keep borrowing, because new debt is priced off it, and a 10-year yield around 5% is high by recent standards. But a rate story alone would have hit long-duration Treasuries too, and the iShares 20+ Year Treasury Bond ETF (TLT) rose 4.6% from September 30 to October 23 and another 0.7% to November 5 (closing prices, same data as the table above). The October move looks more like a repricing of these companies’ own credit than of interest rates in general. For how rate changes reach a whole portfolio, see portfolio rate sensitivity and Treasury yields.

Counterexample: a selloff driven by financing fears can reverse quickly if financing arrives on acceptable terms, or deepen if a report shows slower delivery. The same data fit both a temporary repricing and the start of a longer one; five weeks of prices cannot tell them apart. Large falls in this group are not new either: in the year to September 30, 2026, CoreWeave’s largest peak-to-trough fall in daily closes was 57.5% and Cipher’s 51.4% (Exent market data, same daily-close series as the table above; every name is in neocloud stocks compared).

Who should not use this page to act: anyone looking for a signal to buy a dip or to sell. The page explains the record; it does not estimate the chance of a rebound or of a default.

The big tech reports this page was waiting for are now dated outcomes, and they did not settle it. Microsoft and Alphabet reported on October 28 and Amazon on October 29. On the next session Microsoft moved −0.16% and Alphabet −12.64% (SPY −0.38%), its worst day since its 2004 listing, and Amazon −1.41% (SPY −1.77%). Quarterly spending on property and equipment roughly doubled at all three: Microsoft $39.4 billion vs $19.4 billion a year earlier, Alphabet $51.4 billion vs $24.0 billion, Amazon $77.3 billion vs $35.1 billion. The neoclouds still fell on those two sessions (CoreWeave −5.0% and −9.7%). The record is in how big tech stocks move after earnings and mega-cap stocks after their worst day; the spending plans are tracked in big tech AI spending in 2026.

What would change the picture from here, with dates: the market’s first full session after CoreWeave’s report on November 11, Nebius’s Q3 report on November 12, NVIDIA’s results on November 17 and the next Federal Reserve meeting on December 8–9, after it held its policy rate at 3.75%–4.00% on October 28.