The strongest verified signal: a memory maker spending far faster than its costs are rising
The clearest evidence of an AI-driven chip squeeze in our verified data is not a headline. It is a line in a regulatory filing. Micron Technology, a major memory-chip maker, spent $7.68 billion on capital expenditure (factories and equipment) in fiscal 2023 and $8.39 billion in fiscal 2024.1 In fiscal 2025 it spent $15.86 billion, almost double the year before.1 All three figures come from SEC filings and have been checked against them.
The pace has not slowed. In the first quarter of fiscal 2026, Micron's capital spending was $5.39 billion.1 The same quarter a year earlier it was $3.21 billion, and two years earlier $1.80 billion.1 That is roughly triple in two years.
The dossier has no revenue figures for Micron, so we cannot say how profitable this has been. It does let us compare spending with running costs. Micron's cost of revenue, meaning what it costs to produce what it sells, rose from $16.96 billion in fiscal 2023 to $19.50 billion in fiscal 2024 and $22.51 billion in fiscal 2025.2 Between fiscal 2024 and 2025, capital spending rose about 89% while cost of revenue rose about 15%. A company spending on capacity far faster than its current costs are growing is building for demand it expects, not demand it is already serving.
A balance sheet that has grown to match
Micron's cash, by quarter, tells a similar story. In the third quarter of fiscal 2025 it held $10.16 billion.3 In the same quarter of fiscal 2026 it held $24.995 billion, about two and a half times as much.3 The second quarter of fiscal 2026 stood at $13.91 billion, so most of the increase arrived between the second and third quarters.3 We have no filed explanation for that jump in the dossier, and we do not guess at one. For scale, the cash now held is more than Micron's entire capital spending for fiscal 2025 ($15.86 billion).1,3
Locking in supply: Nvidia and SK Hynix
The most direct evidence of a supply scramble comes from Nvidia. Raj Mirpuri said the planned Nvidia-SK Hynix expansion "will include a co-develop opportunity for us on the next-generation SK Hynix AI memory, and this will help us secure a stable supply of HBM memory."4 HBM is a stacked, high-bandwidth type of memory used alongside AI processors. The article reports that around $950 billion in new AI agreements were signed at a July 24 summit in San Francisco involving Jensen Huang, Sam Altman and the heads of Samsung, SK Group, Hyundai Motor and Naver.4 We cannot put that figure in context from our own data, and we flag that its source is only moderately reliable. Of 4,956 claims we checked from that outlet, 56% held up.4 Treat the headline total as a reported figure, not a verified one.
What analysts are saying about memory stocks
Lynx Equity Strategies rates Micron a clear buy with anticipated upside of 33%, and Sandisk a clear buy with anticipated upside of 49%.5 That report helped drive a rally of more than 4% in ASML, the Dutch company that makes chip-printing machines, on a Friday in early September.5 Two cautions apply. First, these are one firm's forecasts, not results. Second, the Motley Fool articles we draw on have a weak record in our checks: 22% of 18 checked claims held up, a small sample but a low figure.5 Our checks cannot tell you whether the forecasts will prove right, only how carefully the outlet reporting them has been shown to handle facts.
Consolidation and partnership across the chain
Deals are not limited to memory. Bull and Kalray, a French computing firm, announced a collaboration on networking for AI and high-performance computing, including compatibility with Ultra Ethernet, a new open standard.6 Kalray's Éric Baissus said the agreement confirms the relevance of Kalray's technological vision and the quality of its teams' innovations.6
In the United States, Skyworks and Qorvo announced the expected leadership team for their planned combined company, which takes effect once the pending transaction is complete.7 Bob Bruggeworth said the announcement reflects the strong partnership that has shaped integration planning.7 This is a merger of chip suppliers, but the dossier does not tie it specifically to AI demand, so we do not claim that link.
Government money is part of the picture. GlobalFoundries signed a letter of intent with the U.S. Department of Commerce for a $300 million award to advance silicon photonics, which uses light to move data inside AI data centers.8 A letter of intent is not a final award. AMD's Mark Papermaster said that moving data efficiently is as critical as increasing compute performance, and that silicon photonics and advanced packaging will be key.8
Newcomers and the custom-chip push
Zeta announced on October 8 that it has begun development of a specialized chip, the Athena Inference Model (AIM) chip, with partners, using what it calls a capital-light approach.9 The software model behind it was built with Fireworks using NVIDIA Nemotron.9 We have no technical detail on the chip itself. The "capital-light" wording is the company's own description, and it contrasts with the heavy factory spending visible at Micron.
At the research end, Hannah Earley, cofounder and chief technology officer of Vaire Computing, is working on chips that recycle energy normally lost as heat, an approach called reversible computing.10 She says she wants "to tackle every part of how computers are built, and rethink it in these terms."10 This is early-stage work, and the source needs a warning: none of the 11 claims we checked from that article held up against our verification.10 That sample is small. We cite Earley as stating her own aims and do not vouch for the technology.
Small companies show the pull-through
Solitron Devices, a small company traded over the counter, reported fiscal 2027 first-quarter net sales up 101% to about $5.44 million.11 Over the same period its bookings fell 48% to $1.45 million, while its backlog rose 28% to $23.34 million.11 The company said it expects revenue to stay at this level or higher for the rest of the fiscal year.11 The dossier does not link Solitron to AI chips, so it is a data point on semiconductor orders in general, not on AI.
Lattice Semiconductor's Esam Elashmawi reported selling 16,773 shares at a weighted average of $134.56 on Aug. 16 and 17, 2026.12 An executive share sale is not by itself a signal. Executives sell for many reasons, and we mention it only for completeness.
What the dossier could not verify
The research theme behind this piece also includes a 12-to-18-month shortage forecast, an FTC probe of AI labs, and US-China accusations over copying and espionage. We could not trace any of these to a source in our verified material, so we do not state them as fact. Nor does our data show whether Micron's buyback plans, which are limited under CHIPS Act conditions, will change.
What to watch
- Whether Micron's capital spending keeps growing faster than its cost of revenue. If the gap narrows, the build-out is being absorbed.
- Whether the $25 billion in cash is spent, returned or held. The filings show the balance, not the plan.
- Whether the GlobalFoundries letter of intent becomes a signed award.
- Whether the Skyworks and Qorvo transaction completes.
- How Zeta's chip effort progresses, given the claims that surround custom silicon.
We cannot tell from the dossier how much of any of these companies a typical index fund holds, so we make no claim about that.
