AI Stocks Crash: Why Nvidia, Micron, AMD and CoreWeave Fell on Nov 5

On Thursday, November 5, 2026, the AI trade broke. The Nasdaq Composite fell 3.9% and entered correction territory, more than 10% below its record from last month, according to CNBC. Micron and AMD lost about 16%. Nvidia lost 8%. The semiconductor ETF SOXX fell as much as 11%, its worst day since March 2020.

The surprise is the cause. There was no failed funding round, no default and no collapse in demand. The trigger was politics: Democrats won both chambers of Congress, and investors now fear a slower data-center buildout. This post sets out what fell, why it fell, which claims on social media are accurate, and what it means for people who build with AI. It builds on our earlier fact-check of the October neocloud selloff and our long-running AI bubble analysis.

TL;DR: the questions people are asking

What fell on November 5?

Numbers vary by source and by minute, because the sell-off ran all session. The table below uses the closing-style snapshot published by OfficeChai, which matches the quote screens traders shared on social media, and CNBC's index figures.

Index level: CNBC reported the S&P 500 down 2.8% and the Dow down 1,153 points, erasing most of Wednesday's gain of more than 1,500 points. A widely shared intraday chart showed the S&P 500 at 7,484.93 near the lows of the session. Earlier in the day, OfficeChai had the Nasdaq down more than 5% and the Philadelphia Semiconductor Index down about 7%. Those were intraday figures that moved by the close, so use the CNBC numbers as the reference.

Asia moved first. South Korea's Kospi fell 5.28%, SK Hynix fell 9.03% and Advantest fell 11.69%, per CNBC. Hong Kong's Hang Seng Tech rose more than 5%, which shows this was a US-and-memory-chip story rather than a global tech rout.

Why did AI stocks fall? Six causes, ranked

1. Political risk to the data-center buildout

This is the trigger. Democrats won the House and the Senate, and data centers were an issue on ballots across the country. Fool.com reports that the incoming Texas governor campaigned on pausing new data-center construction. Texas is one of the largest data-center markets in the country.

The fear is concrete. UBS estimates that nearly a third of new US data-center capacity expected in 2028 could be delayed or lost if construction pauses spread, according to the same report. We tracked the ground-level version of this earlier: Ohio voters passed data-center bans in 14 municipalities, and our election-night check on Texas and the ballot results showed how strongly local sentiment ran.

One strategist told CNBC investors are "stepping away until they get a better sense of the implications of the midterms around the AI capex cycle."

2. Doubt that AI spending will pay back

Hundreds of billions of dollars are going into chips and buildings. The worry is that profits will not match. OfficeChai notes that Elon Musk acknowledged Grok's serving costs are high relative to revenue. Macro Research Board told CNBC that "investors are beginning to question excessive forward profit estimates" and that the AI shakeout "has much further to run."

3. The financing layer is fragile

Data-center builders borrow to buy GPUs, and neoclouds lend those GPUs out. Higher doubt means higher borrowing costs. We covered the structure in the neocloud shakeout analysis and the IMF's warning about circular financing. Neocloud stocks fell hardest on Nov 5: Nebius, Iren and CoreWeave all lost double digits at some point in the session.

4. Crowded positioning in memory and chips

Micron and AMD fell the most among large caps. OfficeChai describes the move as the memory-shortage trade reversing, with memory stocks falling hardest. Our history of memory prices explains why memory is the most cyclical corner of AI hardware: when the price outlook turns, the stocks overshoot both ways.

5. Weak guidance in specific names

Qualcomm fell about 11% after weaker-than-expected guidance. OfficeChai reports its handset revenue fell 26% year on year to about $5.14 billion versus a $5.25 billion estimate. Intel plunged 28.6% on October 30 after guiding fourth-quarter revenue to $15 to $16 billion against about $17.1 billion expected. These are real company problems, but they are not the main story. The same article notes that SiTime, MACOM and Coherent fell despite better-than-expected results.

6. Funding the build is hurting the spenders

Microsoft is reported to have offered buyouts to about 60,000 employees to fund AI spending, and Alphabet expanded its voluntary exit program, per OfficeChai. Treat the Microsoft figure with care: we flagged it as single-source in our October fact-check. The direction is clear, though. Even the strongest balance sheets are cutting costs elsewhere to pay for capex.

Was this about earnings? Mostly no

This is the part that makes the day unusual. The usual bear case is a bad quarter. Here, the best evidence against that is the stocks that fell on good news:

• AMD fell about 15% after recent record financial results.
• Viavi Solutions reported 57% revenue growth to $469.3 million, raised its outlook and approved a $500 million buyback. It still fell 18% the next morning, per Fool.com.
• Dell lost roughly $20 billion in market value on Nov 5, after falling more than 10% on Nov 4.
• Data-center builders and power-equipment makers fell as much as 20% or more on Nov 4, a day when the S&P 500 rose.

Fool.com's summary is the cleanest: these stocks trade on "fear about the whole AI build-out rather than the business in particular." That is a sentiment and policy shock, and sentiment shocks reverse faster than broken business models do. They also can run further than fundamentals justify, which is why the same stocks are the most volatile ones right now.

What do the viral takes get wrong?

Social posts on the day ranged from "game over" to "nothing happened." Both are off.

"The industry's day of reckoning has come." Gary Marcus, a long-time skeptic, posted a version of this and noted that SpaceX was 10% below its IPO price and CoreWeave was near its own IPO price. The price figures match the quote screens. The conclusion goes further than the data. A single session, even a brutal one, does not show that demand for AI has failed.

"Nvidia is still up 514%." A reply made this point. It is true over a long window, and Marcus answered that Nvidia is flat for the year, slightly down. Both can be right, because the starting date decides the answer. That is why we prefer to report the session's moves and the stated causes, not a cherry-picked window.

"Nothing actually happened, so this is just the beginning." Half right. There was no default and no new law. The article trail agrees that nothing has been enacted. But markets price expected policy, not passed policy. Congress does not take power until January, and any state pause would still need to be written and passed. The uncertainty can last for months.

"This was inevitable anyway." Possible, but unproven. Several outlets saw valuation fatigue before the election. Macro Research Board's quote supports that. The election gave the market a reason, not necessarily the only one.

Is the AI bubble bursting? Our read

Our long-running view in the AI bubble reality check is that "bubble" is not one thing. Split it into layers:

A burst would mean demand falls and financing fails together. On November 5, only the second condition showed stress, and only in the neocloud layer. The more accurate description is a repricing of data-center risk: the market just added a political discount to a build-out that already carried a debt discount.

What would change our mind: a major neocloud missing a debt payment, a hyperscaler cutting its 2027 capex guidance, or state pauses actually becoming law in large markets such as Texas. We saw Musk claim in October that most neoclouds will die. We checked that claim then. It is more plausible this week than it was last week.

What this means for what you build and pay

A finance story only matters here if it changes your decisions. Here is what it changes, and what it does not.

• API prices are not tied to share prices. Nothing in the Nov 5 data changes a provider's per-token price. Do not re-architect on a headline.
• Check where your inference runs. If you rent GPUs from a smaller neocloud, ask how it is financed. Prefer month-to-month terms and avoid large prepayments. Keep a second provider configured.
• Keep an open-weight fallback. Open models can run on several clouds or on your own hardware. Our guides on Kimi K3.1 Pro and running open models locally cover the options.
• Expect hardware prices to be noisy. Memory is the swing factor. Our report on PC hardware prices and data-center demand shows how much AI demand moved consumer parts. A memory sell-off could ease that, but not quickly.
• Plan for slower capacity, not zero capacity. If the UBS estimate holds, new US capacity in 2028 is the part at risk. That points to tighter rate limits and more regional pricing later, not an outage next week.
• Do not make decisions on one day's close. Check a live quote, read the filings, and wait for follow-through.

What to watch next

1. Texas. The incoming governor's data-center pause pledge is the clearest policy risk. Watch for actual draft language.
2. January. Congress is seated and committee agendas appear. Our look at the House "no AI bailouts" resolution shows what early drafts may look like.
3. Neocloud credit. Bond yields and any refinancing attempts at CoreWeave, Nebius and peers.
4. Hyperscaler guidance. Any change to 2027 capex plans at Microsoft, Alphabet, Meta or Oracle.
5. Follow-through. CNBC's note that Wednesday's gain of 1,500 Dow points was erased shows how fast sentiment moved both ways. A second weak day would matter more than the first.

Prices, percentage moves and quotes are accurate as of November 5, 2026, are drawn from press reports of intraday and closing snapshots, and may change. Figures differ between sources. This is analysis for builders, not investment advice.

Related reading

• The AI bubble in 2026: popping, deflating, or just starting?
• Has the AI bubble burst? Three rough weeks, fact-checked
• The neocloud shakeout: Gartner says 80% may vanish
• IMF warns an AI share crash could hit financial stability
• Ohio voters pass data-center bans in 14 municipalities
• Data-center election night: Texas and the ballot results, claimed vs verified
• Anthropic S-1 explained: is Anthropic profitable?
• Memory prices: DRAM, HBM and NAND history