From October 6 into November 2026, the AI trade went through its roughest stretch in years. It came in two waves, the first driven by money and the second by politics, and both arrived while the largest AI customers were reporting some of the strongest cloud demand on record. A falling share price and a narrowing moat are different things, and this autumn demonstrated the gap unusually clearly.

What actually happened

October. The Philadelphia Semiconductor Index fell for eight straight sessions through October 16, losing about 16% from its October 6 close, its longest losing streak since October 2008. By month-end, Yahoo Finance calculated, 47 of the largest AI and chip stocks had lost about $5 trillion of market value while the other 625 large US stocks in its tally added about $1.3 trillion: a rotation, not an exit.

November. The second wave followed the November 3 midterm elections. On November 5 the semiconductor index fell 12.25% to 9,011.14, which a Reuters columnist called its second-largest one-day decline on record, and its worst since March 2020. The Nasdaq-100 fell more than 6% that day, the Dow lost more than 1,800 points, and the Nasdaq Composite closed in correction territory. The chip index recovered 4.2% on November 6, though the Nasdaq still posted its fourth straight weekly decline, and the major indices rose broadly on November 9.

For proportion: at the November 5 close, the semiconductor index stood about 32% below its October 6 close and about 38% below its June 22 record. Yet the iShares Semiconductor ETF had ended October still up more than 50% for the year, so the autumn's losses were taken out of a very large run.

The drivers: money, financing, supply and politics

Five forces were cited, and only some speak to competitive position.

Rates and oil. West Texas crude reached $101.91 on October 14, and the 10-year Treasury yield hit 5.10% the next day. After strong October payrolls, traders put the chance of a December Federal Reserve rate hike at about 78%, according to LSEG data.

The capex bill, and how it is being paid. In late-October earnings, Alphabet reported negative free cash flow of $9.07 billion for the quarter and Meta negative $2.12 billion, while Amazon's trailing-twelve-month free cash flow swung to an outflow of $42.4 billion. Microsoft, at a positive $18.63 billion, was the only one of the four in the black. Financing moved with it: Alphabet's long-term debt reached $125.1 billion, against $21.6 billion a year earlier, and on October 26 Meta announced plans to raise up to $95 billion in new equity.

A supply-side shock in memory. On October 13, China's CXMT said it would launch a denser "4F²" server memory chip by year-end. It published no yield data, yet memory-chip shares in Seoul and New York fell sharply that day.

Politics and permitting. Data centers were a campaign issue in the midterms, mostly over local costs: electricity bills, water and siting. (A 2026 Dallas Fed working paper estimated that data centers had raised wholesale electricity prices by 6.9%.) Voters were on track to reject development in all but one of 41 local ballot measures on data centers, according to Al Jazeera. Democrats won both chambers of Congress, and Texas, home to 27% of planned US data-center capacity according to Wells Fargo, elected a governor who campaigned on a moratorium on data-center approvals. Calls for a national moratorium are contested even among Democrats, and the backlash over local costs has been described as bipartisan. None of it is law yet; Goldman Sachs expects construction delays in Georgia and Texas.

Positioning. Nomura's Charlie McElligott said the bigger driver on November 5 was hedge funds selling the year's biggest winners to lock in gains.

Against all of that, demand was strong. Azure grew 48%, AWS grew 39% to $46 billion, and Google Cloud revenue more than doubled to $30.7 billion.

A falling price is not a narrowing moat

A share price compresses two things into one number: what a business will earn, and what those future earnings are worth today. A moat concerns the first, specifically whether a company can keep earning returns above its cost of capital for a long time. A higher discount rate cuts the present value of distant cash flows whether or not the moat protecting them has changed.

That is why MoatScan AI builds moat analyses from business evidence rather than prices (see How to Identify Economic Moats). A rating that moved because a stock fell would be measuring sentiment; a pullback is informative only where it lines up with business evidence.

One trade, five different moat questions

"AI stocks" covers businesses whose advantages have little in common. Here is how MoatScan AI's analyses rated each layer as of November 10, 2026, and how the autumn's news reached it:

  • Equipment and foundry. ASML (Wide, 83/100) and TSMC (Wide, 82/100) rest on efficient scale and decades of process know-how. They sell into worldwide demand, so a slower US buildout reaches them through total spending, not any single site.
  • Chip designers. NVIDIA (Wide, 81/100) and Broadcom (Wide, 76/100) lean on switching costs built into software ecosystems and custom design work; AMD (Narrow, 58/100) is still building its equivalent. They share a dependence on a handful of hyperscaler customers.
  • Memory. Micron (Narrow, 47/100) sells a product that, high-bandwidth AI memory aside, customers largely buy on specification and price. In a near-commodity a rival's cost curve is the threat, which is why one unverified claim could move the group, and storage and memory names were among the hardest hit on November 5.
  • Neoclouds. CoreWeave (Narrow, 53/100) and Nebius (Narrow, 60/100) rent out GPUs bought from the same suppliers as everyone else, largely with borrowed money. Their business is adding capacity, which puts them where permitting delays and financing costs meet. Gartner predicts 80% of neoclouds will be absorbed or cease operations by 2030.
  • Hyperscalers. Microsoft (Wide, 82/100), Alphabet (Wide, 79/100), Meta (Wide, 78/100) and Amazon (Wide, 77/100) hold moats built long before the AI buildout, as the AI capex arms race analysis argued. The open question is how much cash those moats can divert, and for how long.

These ratings measure business quality, not price. Each page lays out the evidence and risks behind its rating; the wide-moat and switching-costs hubs group companies by source of advantage.

Where permitting risk lands

Permitting rules raise the cost of adding capacity, not of owning it, so their effect is uneven. Harder entry is the classic efficient scale condition: if new sites become slower to approve, capacity already built, powered and permitted grows scarcer, which can favor whoever holds it even as it slows whoever is still building. Builders running on borrowed money are the most exposed, because a delay pushes back revenue and raises financing costs at once.

For the hyperscalers, siting rules leave the funding moats (search, advertising, enterprise software, retail) untouched and change the cost and timing of the capacity those moats pay for. Proposals that make data centers pay for their own grid upgrades would show up as a higher cost of building, not a weaker franchise. How much of this becomes policy is genuinely uncertain; it is a cost-and-timing variable to watch, not a settled outcome.

Four tells worth more than the price chart

Whether the funding moat is still compounding. Older moats are paying for the buildout. In October, Alphabet's Search revenue grew 12% to $63.35 billion, its third straight quarter of slowing growth, and Amazon's online-store sales rose just 2%.

Free cash flow and the financing mix. Capex funded from operating cash is a strategic choice; capex funded with new debt and shares is also a bet that capital markets stay open, and free cash flow conversion is where that shift registers first.

Circularity. Nvidia, AMD and Broadcom indicated non-binding interest in buying up to $5 billion of Meta's share offering: suppliers helping to fund a customer that buys their products. In the late-1990s fiber boom, equipment makers lent heavily to their own customers, which deepened the damage when demand reset. One instance proves nothing, but the pattern is worth tracking.

Returns on capital as depreciation arrives. ROIC separates an investment from an entry fee, and it tends to slide before revenue growth does when spending is not earning its keep.

What a pullback can't tell you

The autumn did not show that AI demand is fading; the cloud numbers say otherwise. Nor did it show that every AI business is equally fragile: layers with hard-to-replicate advantages and layers selling near-commodity capacity were marked down together, for different reasons. What it did show is that a business funded by capital markets and a business funded by its own moat behave differently when money gets more expensive and new capacity gets harder to build. It is why the framework in How AI Is Reshaping Economic Moats starts by asking what each company's advantage actually rests on.

MoatScan AI is a free educational research tool. This article explains how a business and its competitive position work; it is not investment advice or a recommendation to buy, sell or hold any security. Figures are as of the dates stated and will change.