The selloff and the spending point in opposite directions
In early November 2026 the semiconductor sector had its worst session since March 2020, on 5 November. The Nasdaq entered correction, and the AI and data-center trade unwound after the Democratic sweep of Congress and a rotation out of tech.1 For anyone tracking AI infrastructure, the useful question is whether the builders changed course. The verified filings and company reports in our dataset say they did not, at least not by the time of their latest reports.1
This is a story about what the dossier can and cannot show. It holds hard data on Micron's balance sheet and Infineon's results, and thinner material on everything else. We mark the thin parts.
Micron: capital spending four times its FY2023 level
Micron Technology's capital expenditure, which is money spent on factories and equipment, is verified directly from SEC filings. It was $7.676 billion in fiscal 20232 and $8.386 billion in fiscal 2024.2 It rose to $15.857 billion in fiscal 20252 and $30.712 billion in fiscal 2026.2 That is four times the FY2023 figure and nearly double the prior year. Micron's own cost of revenue, the direct cost of making what it sells, grew far more slowly. It went from $16.956 billion in FY20234 to $25.684 billion in FY2026, an increase of roughly half.4 The company is investing in capacity much faster than its production costs are rising.
The quarterly figures show the same pattern. First-quarter capex was $1.796 billion in fiscal 2024,2 $3.206 billion in fiscal 20252 and $5.389 billion in fiscal 2026, about three times the 2024 quarter.2
Cash grew too. Micron held $8.577 billion in FY20233 and $9.642 billion in FY2025.3 The FY2026 figure is $38.364 billion, about four and a half times the FY2023 level.3 The quarterly path shows a build rather than a one-off. It was $13.908 billion in Q2 20263 and $24.995 billion in Q3 2026.3 The dossier does not say what produced the cash or what the capex buys. We will not guess, but the numbers show a company funding a very large expansion from a very large cash position.
Our entity graph links Micron to the AI supply chain. It develops High-Bandwidth Memory, which it supplies to Amazon Web Services and Meta Platforms.12 Micron is headquartered in Boise, Idaho, and the graph also places it in Taiwan.12 It is held by the Fidelity Select Technology Portfolio and covered by Goldman Sachs.12 If you own a technology fund, you may already own Micron indirectly. The dossier gives no portfolio weightings, so we cannot say how much.
Infineon: record revenue, AI revenue named, and a disclosure that went away
Infineon reported fiscal Q4 results on 11 November. Revenue was a record EUR 4.706 billion, up 13% from the previous quarter. Adjusted gross margin was 45.1% and free cash flow was EUR 657 million.7 For AI readers, the notable line is that Infineon put FY2026 AI revenue at EUR 1.7 billion.7 Its FY2027 outlook has data center revenue close to EUR 4 billion, which would be more than double that AI figure. It also has silicon-carbide revenue well above EUR 1 billion and total revenue around EUR 20 billion.7 Q1 revenue is guided to about EUR 4.6 billion with a segment margin near 23%.7
The seam is the backlog. Infineon said it will give only qualitative colour on its order backlog and stop disclosing specific numbers. The previous quarter had disclosed close to EUR 30 billion.7 Backlog is the figure that lets outsiders test whether demand is real, so ending it is a reduction in what we can verify. The dossier does not give Infineon's reason, and we do not speculate about one. Readers should treat the AI and data-center guidance as the company's own outlook. They can no longer check it against a published backlog.
Siltronic: record shipments, but a weaker earnings picture
Siltronic supplies silicon wafers, the base material chips are made on. It reported record 300mm shipments, with third-quarter sales up 10.5% from the previous quarter to EUR 355 million.6 EBITDA rose from EUR 69 million to EUR 81 million, and the net loss narrowed from EUR 63 million to EUR 55 million.6 EBIT was still negative EUR 42 million.6 The company said its new Singapore fab ramp was progressing well and was no longer dilutive to group EBITDA margin in Q3.6
Guidance moved modestly. Siltronic cut its 2026 capex range to EUR 180-200 million from EUR 180-220 million,6 and its depreciation range to EUR 480-500 million from EUR 490-520 million.6 It left unchanged its view that 2026 EBIT would be significantly below the previous year.6 A supplier that is shipping record volumes while still making an operating loss is not a picture of strength. Our dossier supports the phrase "solid results" only for volumes and the narrowing losses, not for profitability.
The analyst call, and how old it is
Lynx Equity Strategies rates Micron a clear buy with anticipated upside of 33%, and Sandisk a clear buy with anticipated upside of 49%.5 The source is a Motley Fool article whose header carries a 4 September date, two months before the rout. So the ratings predate the selloff, and we have no evidence Lynx has re-affirmed them since. The article's reliability should also be weighed. In our measurement, 31% of 58 checked claims from this publication held up.5 That is a small sample and a low rate, so we treat the ratings as one firm's opinion rather than as a verified outlook.
Market reaction: a partial rebound, and strategists stepping back
The rebound was real but partial. On 6 November the Nasdaq rose 1.9%, the S&P 500 1% and the Dow 0.1% after the US added 165,000 jobs in October.8 US indices were higher again by midday on Monday 9 November.8 The same day, strategists led by Beata Manthey cut US stocks and the global technology sector to neutral and raised European stocks outside the UK to overweight.8 The strategists did not call for selling tech, only for holding less of it than before.
Other signals from the AI hardware layer
Several smaller items in the dossier point to where AI infrastructure money is going, though none of them touches the selloff directly. Bull and Kalray announced a collaboration on networking for AI and high-performance computing, including compatibility with Ultra Ethernet, an open networking standard.10 GlobalFoundries signed a letter of intent for a $300 million award from the US Commerce Department to accelerate silicon photonics research, the optical technology used in AI data centers.11 A letter of intent is not a completed award. Both items come from NewsEOD, where 36% of 3,660 checked claims held up, so the announcements are better read as company statements than as confirmed deals.10, 11
A more distant idea comes from Hannah Earley, cofounder and CTO of Vaire Computing, which is building chips that recycle energy normally lost as heat. She compares conventional computing to racing through a city and pumping the brakes at every intersection, so the car loses momentum and must burn more fuel to accelerate again.9 She describes the resonator in these chips this way: “It's really a glorified pendulum.”9 This is relevant to AI because data-center energy use is a central constraint. But the source's measured reliability is only 6%, and that is from 16 checked claims. That is too small a sample for a firm number, and too low to treat the article as established fact.9
What to watch
First, Micron becomes eligible to restart share buybacks on 9 December under CHIPS Act restrictions. Analysts are not necessarily expecting an announcement that day.7 Second, whether Infineon's qualitative backlog commentary holds up against its FY2027 data-center target of close to EUR 4 billion.7 Third, whether Micron's capex, now at $30.712 billion for FY2026, keeps rising or is trimmed after the market's reaction.2 That would be the clearest sign of whether the selloff changed anyone's plans. Fourth, whether the strategists' neutral stance on global tech is reversed.8
The evidence we have says the market and the companies are telling different stories. The market says the AI trade was overextended. The companies' verified numbers say the investment cycle has not paused. Prices can be right about the future while company reports are right about the past, and the dossier cannot settle which view is correct. It also holds no demand data from the AI customers themselves, such as the buyers of chips and memory.

