
- What happened to AI stocks in the latest US market session?
- Why did strong Nvidia results fail to calm Wall Street?
- Why are AI cloud stocks such as CoreWeave and Nebius under pressure?
- What other risks are making Wall Street nervous?
- Author’s take: what should investors check before calling the AI selloff over?
AI companies are making more sales, but their shares are still falling. Wall Street is worried about how much it costs to keep the AI boom growing and whether that spending will earn enough profit.
Nvidia reported $115.8 billion in quarterly revenue, while Anthropic ended its stock-market debut 17.1% above its IPO price. Yet on November 19, CoreWeave fell 6.1%, Nebius dropped 9.4% and Nvidia lost 3.3%. Even these positive developments could not calm investors.
Behind the excitement about AI is an expensive business. Companies need powerful chips, large data centres and electricity to run them. Some borrow heavily to pay these costs before they earn money from customers. If projects run late or borrowing becomes more expensive, growing sales may still leave little profit.
Does that mean the AI bubble is bursting? The selloff alone cannot answer that. But it shows why investors are looking beyond impressive sales figures and asking which companies can turn AI demand into lasting profits.
What happened to AI stocks in the latest US market session?
AI stocks fell sharply on November 19 as investors questioned whether the industry's huge spending would generate enough profit. The selling spread across the wider market, with all three major US indices falling more than 2%.
| US index | November 19 close | Session change |
| S&P 500 | 7,175.83 | −2.02% |
| Nasdaq Composite | 23,879.20 | −2.25% |
| Dow Jones Industrial Average | 49,097.06 | −2.19% |
Sources: Associated Press and Reuters, November 19, 2026. Figures compare regular-session closing values.
The S&P 500 fell for the sixth straight session, leaving it approximately 8.2% below its October record. More than 90% of stocks in the index declined, showing that the nervousness extended beyond AI companies.
Within the AI industry, the selling hit companies that build computing facilities, rent out computing power and supply chips.
| AI-linked stock | November 19 close | Session change |
| Nebius (NBIS) | $96.92 | −9.39% |
| AMD (AMD) | $489.19 | −7.72% |
| CoreWeave (CRWV) | $44.91 | −6.10% |
| IREN (IREN) | $14.41 | −5.45% |
| Nvidia (NVDA) | $145.98 | −3.30% |
| Micron (MU) | $690.44 | −2.10% |
| Broadcom (AVGO) | $346.65 | −2.09% |
Sources: Stock Analysis, Investing.com historical prices, Yahoo Finance and Reuters. Prices exclude extended trading.
Why does this raise fears of an AI bubble bursting? AI stock prices depend partly on expectations of much larger sales and profits in the future. If investors become less confident about those profits, they may pay less for the shares today, even while the businesses continue growing.
Anthropic's successful debut showed that investors still wanted exposure to AI. But enthusiasm for one company could not answer concerns about how other companies would fund expensive infrastructure and earn a return on it.
The session showed weakening confidence in the AI investment story. It did not, by itself, prove that demand for AI was collapsing. That distinction matters: falling share prices can reflect expectations becoming more realistic before the full business impact is known.
Why did strong Nvidia results fail to calm Wall Street?
Nvidia's results showed strong chip demand, but Wall Street remained worried about the cost of sustaining the AI boom. Three concerns remained despite its 103% quarterly revenue growth.
Nvidia expects to earn less gross profit per dollar of sales
Management forecast gross margins of 71%-72% for the next quarter, down from 74.8%. This means direct costs would take a larger share of revenue. Sales could keep growing, but investors also needed to assess how much profit those sales would produce.
AI cloud companies still need money to expand
Nvidia disclosed $32 billion in maximum commitments under agreements with specialist AI cloud providers. These commitments depend on conditions, including those companies securing financing and bringing computing capacity into operation.
That matters to the bubble debate: demand for more chips does not automatically mean customers can afford every planned project. Delayed financing can delay infrastructure and the revenue expected from it.
A leadership change added uncertainty over future support
Following Jensen Huang's November 12 resignation, interim CEO Colette Kress signalled careful evaluation of future investments and guarantees. She did not announce a blanket withdrawal of existing support, but investors still had to assess how selectively Nvidia would fund further expansion.
Strong quarterly results could therefore confirm today's demand while leaving the industry's next stage of growth uncertain.
Sources: Nvidia CEO-transition announcement; MarketWatch coverage of the November 17, 2026 earnings call.
Why are AI cloud stocks such as CoreWeave and Nebius under pressure?
These companies must spend heavily on chips and data centres before earning back that money from customers. Investors are worried about whether they can fund this expansion and turn growing sales into lasting profits.
CoreWeave and Nebius illustrate two different concerns: CoreWeave's borrowing costs are weighing on profits, while Nebius faces delayed revenue alongside rising spending.
Why is CoreWeave's debt burden worrying investors?
CoreWeave earns an operating profit, but that profit is currently too small to cover its interest bill.
| CoreWeave financial measure | Q3 2026 |
| Revenue | $3.661 billion |
| Operating profit under US accounting rules, or GAAP | $207 million |
| Net interest expense | $846 million |
| GAAP net loss | $709 million |
Source: CoreWeave Q3 2026 financial statements, quarter ended September 30, 2026.
CoreWeave's biggest concern is that borrowing costs are outweighing its operating profit. In Q3, its net interest expense was approximately 4.1 times its operating profit. The business was generating more sales, but its earnings before interest and taxes were insufficient to cover that interest bill.
Expansion adds another challenge. CoreWeave must pay for chips and data centres before collecting all the revenue they are expected to generate. It therefore needs funding to keep building while managing its existing debt.
Customer payments could ease that pressure. In its November 10 update, management expected approximately $9 billion in prepayments and deposits over the next three months. These would help finance expansion, but the company must still deliver the services customers have paid for.
Why is Nebius falling despite strong revenue growth?
Nebius faces a different problem: it is spending more while some expected revenue is arriving later. Its 571% year-on-year revenue growth was impressive, but three details make that headline harder to interpret.
Delayed facilities mean delayed sales. Some computing capacity expected this year is now scheduled for 2027. Nebius consequently lowered its revenue expectations within its existing guidance range while increasing planned infrastructure spending. The concern is the longer wait to earn revenue from that investment, rather than evidence that those customers have cancelled.
Some revenue is less predictable. About one-third of quarterly revenue came from strategic partnerships that Nebius describes as non-recurring in nature. Investors therefore cannot assume the whole quarter's revenue will repeat regularly.
Its expansion still depends on funding arriving as planned. Management expects customer advances and loans backed by contracts with financially strong customers to fund much of next year's spending. The next test is whether that funding becomes available and facilities begin earning revenue on the revised schedule.
Source: Nebius Q3 2026 results, Nebius November 12, 2026 shareholder letter.
What other risks are making Wall Street nervous?
Tighter rules could slow AI-chip deliveries
New US measures require licences for advanced chips going to five countries, including Singapore and Malaysia, with exceptions for qualifying US-headquartered customers. The rules also restrict certain overseas access to computing power. Additional checks could complicate deliveries and expansion plans.
AMD faces separate questions about its China links
Hunterbrook alleged that AMD software appeared in servers made by a blacklisted Chinese company. AMD disputed the claims and said it follows US export laws, while Reuters could not independently verify the allegations. Hunterbrook’s affiliated fund was betting on AMD shares falling, so it could benefit from the decline.
Investors worry that further scrutiny could restrict sales or delay orders, making AMD’s future growth less certain. This adds to AI bubble fears because share prices supported by expectations of rapid growth can fall when those expectations weaken. The allegations alone do not prove wrongdoing or a collapse in AI demand.
Interest rates could keep funding expensive
Fed meeting minutes released on November 18 suggested that further rate increases could be needed under policymakers’ economic outlook. This matters because AI companies often borrow to buy chips and build data centres. More expensive borrowing means more money goes towards interest, leaving less profit from the same level of sales. New loans and refinancing can become costlier, although existing fixed-rate loans do not automatically become more expensive.
Higher rates can also reduce what investors are willing to pay for future growth. When bonds offer higher returns, investors may demand a better return from riskier AI shares, which can mean paying a lower share price today. That adds to AI bubble concerns: companies may face higher funding costs at the same time that investors become less willing to pay for profits expected years from now. Even growing demand may not fully offset those pressures.
Source: Federal Reserve minutes of the October 27-28, 2026 meeting, released November 18.
Author’s take: what should investors check before calling the AI selloff over?
Turn the bubble debate into a short list of assumptions you can test against the next disclosures. A falling share price alone cannot show whether a business is becoming better value or whether its original growth assumptions are weakening.
For CoreWeave, track whether operating profit is catching up with interest expense and whether expected customer deposits arrive. For Nebius, compare actual capacity deliveries and recognised revenue with the revised schedule. Separate recurring cloud activity from partnership revenue before projecting the latest quarter forward.
For Nvidia, distinguish demand for delivered chips from conditional financial commitments supporting future projects. Look for completed financing, operational facilities and customer payments rather than treating every announced commitment as funded demand.
Then check your own exposure. Several AI holdings can depend on the same customers, the same financing markets and the same construction cycle. Writing down those shared dependencies can reveal concentration that a list of different ticker symbols hides.
The evidence needed to rebuild confidence is specific: projects delivered, customers paying and a larger share of growth funded by the businesses themselves. That is a more useful test than waiting for either an alarming bubble headline or a sharp one-day rebound to make the decision for you.