AI Stocks Selloff, October 2026: What Broke and What's Next

AI stocks sold off in October 2026 because higher interest rates and weak free cash flow reached the money paying for the AI buildout. A hot inflation report pushed Treasury yields above 5%, Alphabet's Search revenue missed for a second quarter, and Meta and Amazon both reported negative free cash flow. The Nasdaq had its worst month since March 2025.

On Monday, Palantir reported one of the strongest quarters any software company has posted. Revenue grew 96%. US commercial revenue grew 148% and, for the first time, overtook US government revenue. Net dollar retention reached 165%, and its Rule of 40 score came in at 158.

The stock fell about 7% after hours.

That reaction says a lot about where AI stocks are today. The market has stopped debating whether AI is real. It is now asking who gets paid, who pays, and at what interest rate. October was the month those questions started to cost money.

This post compares October with the two AI drawdowns before it, looks at the forces likely to decide what happens next, and sets out five scenarios with rough odds. It closes with what each scenario could mean for anyone running the wheel. The scenarios and the wheel section are our AI-assisted view. They are neither a forecast nor advice.

---

How the October 2026 AI selloff compares with DeepSeek and the Leopold blowup

Each AI selloff so far has broken a different link in the chain. For a premium seller the distinction matters, because the link that breaks shapes how quickly volatility settles and how painful an assignment becomes.

DeepSeek (January 2025): the demand scare

DeepSeek said it had trained a near-frontier model for under $6 million on export-compliant chips. Nvidia fell roughly 17% in a day, about $600 billion in market value and, at the time, the largest one-day loss for any US stock.

The worry was that the world needed far less compute than the market had priced in. The opposite happened. Cheaper intelligence led to more usage, and the narrative reversed within months.

What broke: a story. Spending, earnings and financing were untouched.

The Leopold blowup (July 2026): the positioning scare

Situational Awareness, Leopold Aschenbrenner's AI fund, had grown to as much as $45 billion by early July. It was long AI infrastructure (SK Hynix, Micron, Sandisk, Nebius, CoreWeave) and short software. Running on heavy borrowing, it lost 67% in a month. Its prime brokers issued margin calls, and Citadel bought the public portfolio. (We priced the other side of the fund's put book in what it would pay to sell Aschenbrenner his puts.)

The backdrop was Korea's memory-stock crash. KOSPI fell more than a quarter between late June and mid-July, SK Hynix dropped 38% and Micron about a third. Geared single-stock ETFs, record foreign outflows and an index where two memory makers made up over half the market value did the rest.

What broke: market plumbing. Forced selling stopped once the forced seller had finished.

A detail worth remembering: part of what hurt the fund was its short position in software. Software had started to outperform well before October made it obvious.

October 2026: the financing scare

October came in two waves.

Mid-month: interest rates. Broadcom's Hock Tan warned that financing could become a bigger constraint on AI than chips or power. We covered his remarks, and the chipmakers' lending to their own customers, in AI circular financing. China's CXMT showed progress on a denser DRAM design, and SK Hynix fell 10% in a session. A hot inflation report then pushed Treasury yields above 5%. Twenty of the largest AI and semiconductor stocks lost more than $1 trillion in a week.

Late month: cash. Earnings turned the rate story into a cash story.

• Alphabet fell about 12%, its worst post-earnings day on record, after missing on both revenue and earnings. Search missed for a second straight quarter, and Search is the cash engine that funds Alphabet's buildout.
• Meta raised its 2026 capex range to $135–145 billion and reported its first negative free cash flow quarter since 2012. The stock rose on the day and has fallen about 12% since.
• Amazon reported negative free cash flow of roughly $42 billion.
• Microsoft cut its full-year capex guidance to $170 billion. Neocloud stocks fell, because Microsoft is one of their biggest customers.
• The Fed held rates on 28 October. Two members voted for a hike, and futures now price high odds of one in December.

The Nasdaq finished its worst month since March 2025.

Side by side

	DeepSeek (Jan '25)	Leopold (Jul '26)	October '26
Link that broke	Demand: do we need this much compute?	Positioning: borrowed money and crowding	Financing: who pays, and at what rate?
Fundamentals hit?	No. Cheaper AI drove more demand	No. Mechanical forced selling	Partly. Search pressure, negative free cash flow, higher rates
Epicentre	Nvidia, power names	Korea, memory, neoclouds	Global semis, then the megacaps
How it resolved	Narrative faded	Seller finished selling	Still open

October is the hardest of the three to dismiss because it reaches the funding source. In mid-October, Dan Ives said he would change his view only on "a demand signal" such as hyperscaler capex cuts, and he treated rates as a separate issue. Two weeks later, Microsoft cut its capex guidance.

Whether that counts as the demand signal, or simply one company being disciplined, is the question the rest of this post works through.

---

What drives AI stocks next: consumer agents and capex

For three years, AI money flowed in one direction: from hyperscaler balance sheets into chips, memory, power and data centres. The bet was that revenue further down the chain, in software and consumer products, would eventually arrive to justify it. Software is covered in the scenarios below, where it matters most. This section looks at the other two forces: consumer agents and the shape of capex growth.

Consumer agents are both the prize and the threat

October was the month consumer AI agents moved from demos to products.

• Google launched Universal Cart across Search and Gemini.
• Meta said on its earnings call that Muse, its personal shopping agent, will take a cut of the sales it drives. That is a new revenue line on top of advertising.
• TikTok launched an AI shopping assistant and native in-app checkout, Walmart opened its store to third-party shopping agents, and Shopify opened checkout to browser-based agents in late September.
• ChatGPT ads now run in 89 countries, with more than 100,000 advertisers.

Alphabet's Search miss and Google's Universal Cart belong to the same story. Agents open a new way to make money, and they also eat into the high-margin ad model that pays for the capex. The likely winners own both the agent and the checkout. The likely losers are businesses that depended on someone clicking a link.

This holiday season is the first real test. Adobe expects AI-referred traffic to US retail sites to rise 130% year on year. Trust is still low: about 31% of shoppers say they would let an AI complete a purchase for them, and comfort drops sharply for expensive items. Black Friday and Cyber Monday data will be worth reading closely.

We look at which companies are best placed to win consumer adoption, why small merchants are likely to move first, and which stocks are exposed in Consumer AI adoption: who wins, and what it means for stocks.

Capex, and why the growth rate matters more than the total

Hyperscaler capex in 2026 is running at roughly $700–800 billion, up about 80%. Forecasts for 2027 vary widely:

Forecaster	2027 capex
Morgan Stanley (cloud tracker)	About +29%
Consensus	About $1.1 trillion
Goldman Sachs	About $1.2 trillion, +54%
Goldman Sachs, 2028	+12%

Every forecast points up, and that can mislead. AI supplier stocks tend to follow the rate of change in spending more than the total. Spending can keep rising while growth slows, and that slowdown alone is usually enough for Nvidia, memory makers and neoclouds to trade at lower multiples.

Two more points. Capex now exceeds hyperscaler operating cash flow, so the gap is being filled with debt and leases. That is why a 5% Treasury yield hurts. And no hyperscaler has yet given 2027 capex guidance. That arrives in the January–February earnings season.

---

Five scenarios for AI stocks after October 2026

Software and consumer agents add up to one question: does AI revenue show up further down the chain? Capex asks the other: does its growth hold, or slow? Crossing the two gives four scenarios, with a fifth that can sit on top of any of them.

The horizon runs through Q4 earnings and 2027 capex guidance, roughly to March 2027. The odds are our judgement and should be read as rough.

	AI revenue shows up	AI revenue lags
Capex growth holds	The AI rally resumes: 25%	Spending outruns revenue: 20%
Capex growth slows	Software takes the lead: 40%	AI demand falls short: 15%

1. Software takes the lead: 40% (base case)

Capex keeps growing, but more slowly, closer to Morgan Stanley's +29% than Goldman's +54%. Meanwhile, revenue keeps building in workflow software and consumer agents. Market leadership passes from the companies building AI capacity to the companies earning money from it.

• Likely laggards: high-beta infrastructure such as Nvidia, Micron and SK Hynix, and neoclouds such as CoreWeave and Nebius. The more likely path for them is a long sideways stretch while valuations compress.
• Likely leaders: workflow software such as Palantir and ServiceNow, agent and checkout owners such as Meta and Shopify, and cash-generating platforms such as Microsoft.
• Why it is the base case: Microsoft's capex cut, Palantir's quarter and Meta's take-rate model already point this way. October looks more like the start of a rotation than the end of a cycle. It is also the path we sketched in August in the AI buildout's move from capex to ROI capture.

Which software thrives, and which is at risk

Palantir's quarter is the clearest evidence for this scenario so far. US commercial customers grew 29%, existing customers are spending 65% more than a year ago, and 2026 adjusted free cash flow guidance sits at $4.8–5.0 billion. Two cautions apply. The stock fell anyway, which shows how much was already priced in. And total contract value grew only 6% year on year, a number that tends to lead revenue.

This scenario will not lift all software equally. Alex Karp's sharpest line on the call was aimed at "software companies that are reselling tokens and it counts as revenue." That point, and two others, do most of the sorting:

1. Does it own the data, or display it? Agents need a system of record to act on. A screen that sits on top of that record is the part an agent can replace.
2. Does pricing grow with work done or with headcount? If an agent does the work of five people, usage-based and outcome-based pricing grows, while per-seat pricing shrinks.
3. Is the AI revenue its own? Revenue that is mostly model cost passed through tends to flow back to the model provider, and the market usually reprices it once customers notice.

Likely to thrive:

• Workflow and operations platforms built on proprietary data, such as Palantir and ServiceNow. Agents add value here because they have to run through these systems.
• Data and usage-based platforms, such as Snowflake, Datadog and MongoDB. Every agent that queries, logs or stores something generates billable usage, around the clock.
• Commerce and payment rails that agents check out through, such as Shopify. If consumer agents take off, whoever processes the order earns from it, whichever agent wins.
• Vertical software in regulated industries, such as Veeva. Compliance requirements, proprietary data and cautious buyers make these businesses hard for a general-purpose agent to displace.

Somewhat at risk:

• Per-seat horizontal apps sold against a headcount budget (customer support, basic CRM, project management), such as Monday.com, Asana and HubSpot. The product may hold up well. The risk sits in the number of seats.
• Creative and content tools where models make the output cheap, such as Adobe. The brand and distribution are strong, but the core output is what generative models produce most cheaply.
• Point solutions that larger platforms can bundle, such as standalone note-taking, scheduling or analytics tools that Microsoft or Google can add to an agent suite at no extra cost.
• Thin AI wrappers whose revenue is mostly model cost. Many are private, but some listed companies carry this exposure inside their "AI revenue" lines. Gross margin trends are the place to check.

A caution on the second list: "at risk" means slower erosion over years, and these names can still rally hard in the meantime. Leopold's short position in software, Adobe included, was part of what sank his fund in July. For wheelers, the list is more useful for deciding where to avoid selling puts than for picking shorts.

A deliberate omission: cybersecurity. It may be the most important software category in a world full of agents. Every agent holding a payment card and API keys widens the attack surface, and agent identity and permissions remain unsolved. The group is not immune to selloffs either: Fortinet fell 25% in October on a narrow margin miss and pulled the sector down with it. Cybersecurity deserves its own analysis, so we are intentionally leaving it out of this post.

2. The AI rally resumes: 25%

In this scenario, October turns out to have been a rate scare that looked like a demand scare, much as DeepSeek was an efficiency story that looked like one. Inflation cools, markets stop pricing a December hike, Nvidia's next report shows growth picking up again, and 2027 capex guidance lands nearer Goldman's forecast than Morgan Stanley's.

• Likely leaders: the stocks that fell hardest, semis and memory such as Nvidia, AMD and Micron, along with the hyperscalers.
• What has to go right: both the rate path and the capex guidance. Needing two things to break the right way at once is why this is not the base case.

3. Spending outruns revenue: 20%

Capex keeps climbing on borrowed money, but downstream revenue falls behind. This follows the telecom pattern of 2000: the infrastructure is real and the usage eventually arrives, but the people who financed the buildout do not get repaid in full.

• Where it shows up first: in credit markets, before equities. Neocloud and data centre debt reprices, lease terms tighten, and a GPU-backed loan or two gets restructured. What actually breaks in the AI value chain sets out the credit data to watch.
• Likely laggards: debt-heavy infrastructure such as CoreWeave, Nebius and IREN, and the suppliers whose order books depend on them.
• The link to July: Leopold's blowup was borrowed money concentrated in one fund. This scenario is the same mechanism spread across the whole system.

4. AI demand falls short: 15%

This is the bear case. Capex slows because demand turns out to be thinner than expected, and the Fed hikes into the slowdown. 2027 capex guidance comes in flat. Search keeps losing share and agent revenue does not replace it.

• Likely laggards: almost everything linked to AI, including the most richly valued software, such as Palantir, where growth stops supporting the valuation.
• Why only 15%: nothing in the October reports looked like a collapse in demand. Azure grew 48%, and Google Cloud revenue more than doubled. The warning signs Ives listed, such as cancelled contracts or falling GPU rental prices, have not appeared.

Wildcard: Private labs keep the profits

This one can sit on top of any of the other four. We put about a one-in-three chance on it becoming the main story by end-2027.

AI revenue arrives, but most of it lands at OpenAI and Anthropic, which are private. Listed hyperscalers carry the capex, the debt and the depreciation, while the margin goes to labs that public investors cannot buy directly.

In this world, public AI stocks can lag even while AI succeeds commercially. It is the scenario the market seems least prepared for. Signs to watch include the labs raising API prices, or moving from renting compute to owning it.

---

What the AI selloff could mean for selling puts

This section is an AI-assisted view and is not financial advice. It translates the scenarios into the questions a wheeler weighs: where put premium may be worth the assignment risk, where covered calls may fit, and where assignment risk looks highest. The names are worked examples only.

For any cash-secured put, the maximum loss is the strike minus the premium received, per share, if the stock goes to zero. The breakeven is the same number: strike minus premium. Every scenario below changes how likely you are to get near either.

In every scenario:

• Rates work in two directions. Higher rates weigh on AI valuations, but the cash behind a cash-secured put is earning close to 4% in T-bills. That raises the bar each trade needs to clear.
• Size positions for the gap. October's single-day moves (SK Hynix down 10%, Alphabet down 12%) went straight through 20-delta strikes. Earnings and macro dates carry more risk than usual.
• Check the balance sheet. Being assigned a cash-generating megacap is manageable. Being assigned a debt-heavy neocloud if spending outruns revenue is a much larger risk.

By scenario:

Scenario	Where put premium may be worth the risk, such as	Where covered calls may fit, such as	Where assignment risk looks highest
Software takes the lead	Pullbacks in workflow software and agent platforms, such as Palantir, ServiceNow and Meta	Covered calls on range-bound semis, such as Nvidia and Broadcom	Debt-heavy neoclouds
The AI rally resumes	Beaten-down semis and memory, such as Micron and AMD	Strikes further out of the money, since high-beta names get called away quickly	Lower here; the bigger risk is missing the rebound
Spending outruns revenue	Cash-rich platforms, such as Microsoft	Debt-heavy names already held	Neoclouds such as CoreWeave, Nebius and IREN
AI demand falls short	Smaller size, longer dated, further out of the money	Soon after any assignment	Richly valued software

---

What to watch next, in order

1. Nvidia earnings (late November). The growth rate and guidance matter more than whether it beats.
2. Black Friday and Cyber Monday. AI-referred traffic and agent checkout data give the first real read on consumer agents.
3. Software earnings (early December). These will show whether Palantir's pattern is spreading.
4. The December 8–9 Fed meeting. A hike could give the financing scare a second wave.
5. Neocloud credit spreads. The earliest warning that spending is outrunning revenue.
6. Q4 earnings and 2027 capex guidance (late January to February). This is where the scenario is likely to be decided.
7. Monetisation checks: Palantir's total contract value, Meta's disclosures on sales through Muse, and any pricing moves from the private labs.

October did not settle whether AI is a bubble. It raised a narrower and more useful question about who earns a return on the buildout. For premium sellers, that question matters more than the headline debate, because it decides which shares are worth owning if assigned. The levelbox screener updates each name's trend and premium daily.

Method and sources

• Market moves and earnings: CNBC, CNN and Yahoo Finance reporting from 15 to 30 October 2026. Monthly Nasdaq figures vary by index and source, so we give the ranking (worst month since March 2025) rather than one number.
• Palantir: the Q3 2026 press release filed with the SEC on 2 November 2026 and the earnings call transcript.
• Situational Awareness and Korea: CNBC (30 July 2026), Yahoo Finance, ChinaTalk.
• Capex forecasts: Goldman Sachs, Morgan Stanley and Moody's, as reported in September and October 2026. They cover different company sets, so the growth rates are not directly comparable.
• Consumer agents: Retailgentic's October 2026 roundup (which collects the Adobe, Shopify and CivicScience survey figures) and TechCrunch.
• Fed decision: the 28 October 2026 hold and the 10–2 vote, as reported by HackerNoon.
• Scenario odds: our judgement, made with AI assistance. They are not model outputs.

Links:

• CNBC: AI stocks tumble in global tech rout on AI funding worries (15 Oct 2026)
• CNN: Chipmaker sell-off sends investors to other stocks and bonds (15 Oct 2026)
• CNBC: Alphabet faces record post-earnings loss (29 Oct 2026)
• CNN: Nasdaq suffers worst month in over a year (30 Oct 2026)
• Yahoo Finance: Alphabet's 12% plunge, Meta and Microsoft
• CNBC: Aschenbrenner forced to unwind public positions (30 Jul 2026)
• Yahoo Finance: How 4x borrowing wiped out 67% of a $45B AI fund
• ChinaTalk: Korea's stock market crashed $1 trillion
• Tiger Brokers: CXMT's 4F² DRAM and the memory selloff
• DCD: Nvidia's record $600bn DeepSeek loss
• Investing.com: Palantir Q3 2026 earnings call transcript
• SEC: Palantir Q3 2026 press release
• 24/7 Wall St: Palantir commercial tops government
• Retailgentic: October agentic commerce roundup
• TechCrunch: Shopify opens checkout to browser-based AI agents
• Yahoo Finance: Goldman says 2027 hyperscaler capex consensus is too conservative
• Seeking Alpha: Morgan Stanley sees 29% cloud capex jump in 2027
• DCD: Moody's sees hyperscaler capex closing in on $1trn by 2027
• HackerNoon: Fed holds at 3.75–4.00% in 10–2 vote

This article is analytical and educational and is not financial advice. It names companies as worked examples only. Figures are a point-in-time snapshot as of 3 November 2026 and can contain errors. This article was generated with AI assistance. If you spot a mistake, tell us.