The AI trade is being repriced, but the claim that all AI stocks are falling misses an important distinction. The steepest damage is concentrated in companies whose growth depends on financing a costly buildout, delivering new capacity and converting ambitious AI plans into cash. Some profitable technology and AI-software names have held up much better. As of October 24, 2026, the latest completed US trading session is Friday, October 23. We compared September 23 and October 23 closing prices across 20 US-listed AI-related stocks. IREN, Cerebras, CoreWeave, Arm and Nebius suffered the five largest percentage declines in that defined screen. This is not a ranking of every AI-related stock worldwide, nor a claim that the largest percentage losers destroyed the most market value. The central question is changing from “How much AI capacity can companies build?” to “Who can finance it, and when will shareholders see a return?”
The five biggest casualties over the past month
Our screen covers IREN, CBRS, CRWV, ARM, NBIS, SMCI, MU, DELL, AMD, META, ORCL, AVGO, HPE, NVDA, TSM, GOOGL, AMZN, PLTR, MSFT and AI. It spans chips, servers, cloud infrastructure, hyperscalers and enterprise software. The ranking uses percentage price declines, not dollar market-cap losses.
| Rank | Company | Sep 23 close | Oct 23 close | One-month change |
|---|---|---|---|---|
| 1 | IREN ($IREN) | $47.05 | $24.91 | −47.1% |
| 2 | Cerebras ($CBRS) | $205.24 | $110.82 | −46.0% |
| 3 | CoreWeave ($CRWV) | $86.90 | $54.24 | −37.6% |
| 4 | Arm (Arm) | $332.56 | $210.33 | −36.8% |
| 5 | Nebius ($NBIS) | $226.61 | $158.74 | −30.0% |
Source: Yahoo Finance daily historical-price data for IREN, Cerebras, CoreWeave, Arm and Nebius, retrieved October 24. Returns are calculated from the provider's unrounded daily closes as (October 23 close ÷ September 23 close − 1) × 100. Displayed prices are rounded to cents. These are end-of-day, delayed prices and price changes, not dividend-reinvested total returns, October month-to-date changes or drawdowns from a peak.
The comparison also challenges the idea of an indiscriminate collapse: over the same interval, Nvidia fell about 6.5%, while Microsoft rose about 4.8%, Palantir rose 2.6% and C3.ai rose 14.7%. A weak AI infrastructure stock and a resilient software business can coexist.
The main driver: the price of financing the AI buildout
AI demand and AI funding are not the same thing. A company can have substantial customer demand yet struggle to finance GPUs, buildings, power connections and working capital on attractive terms. Debt-heavy businesses are particularly exposed when borrowing costs rise or lenders demand more protection.
CNBC reported on October 23 that CoreWeave's credit-default-swap spreads had widened past 11%, a level it described as distress territory, while credit concerns also spread to larger AI-related borrowers. A wider spread means investors are paying more for protection against default; it is a warning about perceived risk, not proof that default will occur. 1
The mechanism matters for equities: more expensive funding can reduce the returns from new projects, force slower expansion or increase the incentive to issue shares. New equity can protect the balance sheet while diluting existing owners. For CoreWeave, Nebius and IREN, the useful questions are how much funding is already secured, what it costs, when capacity begins producing revenue and how much additional capital is still needed.
Why strong AI demand is no longer enough
Investors are asking for evidence of payback, not just larger spending commitments. Reuters reported that Meta's proposal for a new share class, partly to help fund further capital spending, contributed to the market's concern about the cost of the next phase of AI expansion. Thursday's sell-off was followed by mixed trading on Friday rather than a uniform second collapse. 2
The distinction is between growth and profitable growth. Revenue may increase while free cash flow deteriorates because infrastructure spending rises faster. Even an earnings beat can disappoint investors if management raises spending without explaining the incremental revenue, margins or financing behind it.
For chip designers such as Arm and Cerebras, monitor customer adoption, revenue conversion, margins and valuation against realistic earnings expectations. For infrastructure operators, monitor utilization, contract economics and the timing of cash receipts. These are analytical risk channels, not a claim that one identical event caused each company's monthly decline.
War, oil and rates add a second layer of pressure
The sell-off is not solely an AI story. Al Jazeera reported that the Nasdaq fell 2.6% on October 22, while rising oil prices linked to the US–Israel war on Iran worsened the backdrop. Its October 23 report put Brent above $114.20 a barrel that morning. That is a dated reported observation, not a current oil quote. 3
Higher energy costs can keep inflation concerns alive, complicate the path to lower interest rates and increase operating costs. Higher yields also reduce the present value investors place on profits expected far into the future. That combination is uncomfortable for highly valued growth stocks and companies relying on external financing.
As the November US midterms approach, watch policy proposals on energy, export controls, tariffs and technology regulation. Election uncertainty may affect sentiment, but an election outcome does not mechanically determine stock returns. Oil, bond yields, company guidance and actual policy implementation are more useful evidence than campaign headlines alone.
What the five hardest-hit stocks need to prove
- IREN: that planned AI capacity becomes utilized, recurring revenue without expansion costs overwhelming cash generation. A large future campus is not the same as operating cash flow today.
- Cerebras: that customer commitments translate into revenue and sustainable margins. Watch customer concentration, deployment milestones, dilution and insider-share supply alongside demand.
- CoreWeave: that funding remains available on workable terms and contracted business can support debt service and capital expenditure. Credit-market stabilization would be a particularly relevant signal.
- Arm: that AI-related adoption translates into durable licensing and royalty growth sufficient to support its valuation. A lower share price alone does not establish a low earnings multiple.
- Nebius: that financing, construction timelines and customer ramp-ups align. Revenue contracted for future capacity should be assessed alongside the spending required to deliver it.
Analyst optimism can coexist with falling prices. MarketBeat's October 23 Cerebras report described a consensus Moderate Buy rating despite recent target reductions from TD Cowen and Citigroup. Treat those ratings as opinions with assumptions and differing publication dates, not a floor under the stock. 4
Six signals that matter more than a one-day bounce
- Credit spreads and refinancing: narrower spreads and credible funding plans would support stabilization; further widening would undermine an equity rebound.
- Cash flow relative to capital expenditure: improving operating cash generation or a clearer path to funding spending internally is stronger evidence than a bigger headline backlog alone.
- Earnings revisions: watch whether analysts cut future profits, not just price targets. A stock can fall sharply and still become more expensive relative to reduced earnings expectations.
- Oil and Treasury yields: sustained easing would reduce an external headwind; renewed increases could pressure valuations and financing again.
- Market breadth and relative strength: a recovery involving infrastructure, semiconductors and software is more convincing than a rally driven by one mega-cap. Compare stocks with the Nasdaq or a semiconductor benchmark.
- Price structure and volume: higher lows, successful retests and the recovery of commonly watched 50-day and 200-day averages can indicate improving momentum. These are signals to monitor, not a claim that today's chart has confirmed a bottom.
A sharp rebound after a large fall may reflect short-covering rather than improved fundamentals. No single indicator reliably identifies the low.
What to expect next: three plausible paths
Stabilization: upcoming earnings show credible AI monetization, funding plans become clearer and oil or yields ease. The market could then distinguish profitable platforms from more financing-dependent projects, allowing a selective recovery.
More downside: capital expenditure keeps rising, cash-flow forecasts weaken, credit spreads widen or financing requires greater dilution. In that case, a stock already down 40% can still fall substantially further.
A volatile trading range: demand stays strong but financing and payback remain unresolved. Large rallies and fresh sell-offs could alternate until several quarters of results provide clearer evidence.
These are conditional scenarios, not forecasts or probability estimates. Alphabet, Microsoft and Amazon's next results and spending commentary are important checkpoints; Reuters' October 23 report highlighted the coming earnings wave. The key is the relationship between AI demand, spending, margins and cash—not simply whether revenue beats consensus. 2
Bottom line: a sell-off is not automatically an entry signal
The evidence points to a reassessment of financing risk and returns on AI spending, amplified by oil, rates and geopolitical uncertainty—not proof that AI demand has disappeared. The five largest monthly percentage losers in our 20-stock screen are IREN, Cerebras, CoreWeave, Arm and Nebius, but their business models and risks differ.
A stronger entry thesis needs more than a large percentage decline: a defensible valuation, adequate funding, improving cash economics and evidence that expectations have become realistic. Patience until those signals improve is a valid alternative to buying the first rebound.
Explore the earnings calendar, compare upcoming reporters in the stock screener, and read our Cerebras analysis. All market observations are dated as of the October 23 close or explicitly attributed to a dated report. This article is general research, not financial advice or a recommendation to buy or sell any security.