The constraint has moved from compute to memory and movement
For practitioners, the useful fact about this cycle is where the money is going. It is going to memory and the links between chips, not only to processors. Micron's capital expenditure was $7.676 billion in fiscal 20231 and $8.386 billion in fiscal 20241. It then jumped to $15.857 billion in fiscal 20251, which is roughly double in a single year. All three figures are canonical and come from SEC filings.
The quarterly figures show the same pattern. First-quarter capex was $1.796 billion in 20241, $3.206 billion in 20251 and $5.389 billion in the first quarter of 20261. That is about three times the 2024 level in two years. A company that triples its quarterly building budget is not guessing that demand is a blip.
The balance sheet is absorbing it
Micron's cash tells a parallel story. Its cash stood at $10.163 billion in the third quarter of 20252. It was $13.908 billion in the second quarter of 20262 and $24.995 billion in the third quarter of 20262, about two and a half times the year-earlier figure. Cost of revenue, the direct cost of making what it sells, rose more gently: $16.956 billion in fiscal 20233, $19.498 billion in fiscal 20243 and $22.505 billion in fiscal 20253, an increase of about a third over two years. Capex more than doubled over the same period while direct production costs rose by a third.
The dossier holds no revenue or margin figures for Micron, so we cannot say how much of the cash build reflects pricing power and how much reflects other factors. We flag that gap rather than fill it. Our narrative analysis of the wider reporting is that AI demand is driving a chip and memory shortage expected to last 12 to 18 months. That is Via News's synthesis, not a sourced forecast, and it should be read as one.
Locking in supply: Nvidia and SK Hynix
The clearest sign of how tight memory is comes from Nvidia's supply deals. Raj Mirpuri said of the Nvidia–SK Hynix expansion: "The 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 The surrounding report describes a July 24 summit in San Francisco where roughly $950 billion in new AI agreements were signed, and says Nvidia has spent much of 2026 signing deals across Asia, the Middle East and Europe to secure chips, memory and power4. The phrase "co-develop" matters here. Nvidia is not only buying a commodity. It is getting a hand in shaping the memory it will depend on.
On reliability: the source of this report, NewsEOD, has had 56% of 4,959 checked claims hold up4. That is a middling record, so treat the headline dollar total as reported rather than verified.
Interconnect and photonics: moving data is the other half
Bull and Kalray announced a collaboration on networking technologies for AI and high-performance computing infrastructure. They will also work together on compatibility with Ultra Ethernet, a new open standard that may define the future of networking for next-generation supercomputers5. Kalray's Éric Baissus said the agreement "confirms once again the relevance of our technological vision"5 (our translation from the French original). The source is weak: only 32% of 2,928 checked claims from this feed held up5. The announcement is a collaboration. The dossier does not support calling it a merger.
On the manufacturing side, GlobalFoundries signed a letter of intent with the U.S. Department of Commerce for a $300 million award to accelerate silicon photonics R&D6. A letter of intent is not a finished award. AMD's Mark Papermaster commented: "As AI systems scale, moving data efficiently is as critical as increasing compute performance."6 He added that silicon photonics and advanced packaging will be key to the bandwidth and energy efficiency that next-generation AI clusters need6. That is a competitor to Nvidia, per our entity graph, naming the same bottleneck. The source feed has the same 32% reliability record6.
The heat problem: reversible computing
A different line of attack treats waste heat as a design choice. Hannah Earley, cofounder and CTO of Vaire Computing, is building chips that recycle energy normally lost as heat, an approach known as reversible computing7. Her analogy: "Conventional computing is like racing through a city only to pump the brakes at every intersection: the car loses momentum and must burn more fuel to accelerate again."7 She also said of her resonator, "It's really a glorified pendulum."7 Her stated ambition is to "tackle every part of how computers are built, and rethink it in these terms."7
The caveat is large. The dossier's measured reliability for this source is 0% of 11 checked claims held up7. That is a thin sample, and it says nothing about whether Earley's technology works. It means our verification system could not confirm the article's claims. The technical case remains unproven in this evidence, and the quotes show her intent only.
What the market is saying
Lynx Equity Strategies rated Sandisk "a clear buy with anticipated upside of 49%" and Micron a clear buy with anticipated upside of 33%8. The report said the analysis sparked a sympathy rally that pushed ASML's equity up by over 4%8. These are one firm's price targets, reported by a Motley Fool source where 22% of 18 checked claims held up8. Read them as sentiment, not as verified forecasts.
The dossier also records that, under CHIPS Act restrictions, Micron could technically restart stock buybacks, though analysts were not necessarily expecting an announcement9. The event record carries a date later than this piece's publication, so we treat the item as an unconfirmed possibility. We do not treat it as a settled fact.
Custom silicon reaches enterprise software
Zeta Global's October 8 announcements show specialization reaching beyond chipmakers. Zeta introduced AthenaOS, an enterprise intelligence operating system10. It also introduced the Athena Inference Model, developed with Fireworks using NVIDIA Nemotron10, and Athena MCP, a headless way to bring its intelligence into agents and workflows10. Most notably, Zeta has begun developing a specialized AIM chip with partners through what it calls a capital-light approach10. We have no performance data on the chip, so we cannot judge it. The signal is the direction: even a software-led company now sees dedicated silicon as part of its inference strategy.
Consolidation in adjacent silicon
Skyworks and Qorvo announced the expected leadership of their combined company, pending completion of the previously announced transaction11. Bob Bruggeworth said the announcement reflects "the strong partnership that has shaped our integration planning efforts from the very beginning."11 This is a radio-frequency pairing and not directly an AI one. It does show that chip-sector consolidation is progressing on its own timeline. At the small end, Solitron Devices reported first-quarter net sales up 101% to about $5.44 million, with backlog up 28% to $23.34 million12. Net bookings fell 48% in the same quarter12. Both are small numbers from a company on the OTC Pink market, so they illustrate the point only loosely.
What we could not verify
The topic framing mentions further items: the onsemi–Synaptics deal, DeepSeek–Huawei ties, FTC scrutiny of AI agents, OpenAI's accusations against Moonshot, Chinese espionage targeting AI experts and YMTC's NAND expansion. None has a supporting fact, claim or source in the dossier behind this article. We therefore do not assert them here.
What to watch
- Whether Micron's capex keeps climbing from the $5.389 billion first-quarter 2026 level1, and whether its cash build continues beyond $24.995 billion2.
- Whether the GlobalFoundries letter of intent becomes a finalized award6.
- Whether Zeta's AIM chip gets named partners and benchmarks10.
- Whether the Micron buyback question gets an answer9.