The European footprint in this story is small but real. Grenoble-based Bull and Kalray, a company listed on Euronext Growth Paris, announced a collaboration on networking for AI and high-performance computing on 28 July 2026.1 They will also work together on compatibility with Ultra Ethernet, an open standard the announcement says could define the future of networking for coming generations of supercomputers.1 The main money and the main deals, however, sit outside Europe, and this piece says so where the evidence runs out.
The European angle: networking, not memory
Kalray's chief executive Éric Baissus said the agreement with Bull confirms the relevance of Kalray's technological vision and the quality of its teams' innovations. He added that the collaboration aims to develop solutions for the next generation of AI and HPC, and to create long-term value for customers worldwide.1 That is a company's description of its own deal, and it gives no financial terms. Our dossier contains none, so we cannot say how large the arrangement is.
The Netherlands appears through ASML, which the Motley Fool describes as the Netherlands-based company that prints circuits onto microprocessors. Its stock rose by over 4% after the research firm Lynx Equity Strategies published a highly bullish analysis of a top chip-sector name, which set off a sympathy rally in related shares.2 Lynx rated Micron a clear buy with anticipated upside of 33%, and Sandisk a clear buy with anticipated upside of 49%.2 These are one firm's opinions, not forecasts we have verified. The source is also weak on our measure: only 22% of 18 claims we checked from it held up.2 Treat the percentages as a mood indicator, not as data.
If you hold a broad European index fund, ASML may be one of your holdings. The dossier does not give its weighting, so we will not guess at one.
The verified backbone: what Micron's filings show
The strongest checked evidence in the dossier comes from Micron's SEC filings, which our system treats as the source of truth. Micron makes memory chips, which AI systems need in large quantities. Its capital expenditure, meaning money spent on factories and equipment, was $7.676 billion in fiscal 2023 and $8.386 billion in fiscal 2024.3 In fiscal 2025 it reached $15.857 billion, roughly double the year before.3
The quarterly figures show the same trend. First-quarter capex was $1.796 billion in fiscal 2024, $3.206 billion in fiscal 2025 and $5.389 billion in fiscal 2026.3 That is about three times the level of two years earlier in a single quarter. We have no price tag from the dossier to compare these totals with, such as a national budget or a company's market value, so the clearest yardstick is Micron's own history. It is now spending in one quarter close to what it spent in a whole year in fiscal 2023.
Cash has grown too. Micron held $10.163 billion at the end of the third quarter of fiscal 2025 and $24.995 billion at the end of the third quarter of fiscal 2026, about two and a half times as much.4 The intermediate readings are $13.908 billion in the second quarter of fiscal 2026 and $9.731 billion in the first quarter.4 The cash pile is rising even as capex climbs.
Costs have risen more slowly. Cost of revenue, meaning what it costs Micron to produce what it sells, was $16.956 billion in fiscal 2023, $19.498 billion in fiscal 2024 and $22.505 billion in fiscal 2025.5 Quarterly cost of revenue rose from $4.761 billion in the first quarter of fiscal 2024 to $5.997 billion in the first quarter of fiscal 2026.5 Our dossier holds no revenue figures, so we cannot calculate profit margins and will not imply them. What the data does show is a company that is spending heavily on capacity while its cash also grows.
One related event in the dossier: under CHIPS Act restrictions, Micron could technically restart stock buybacks, though analysts were not necessarily expecting an announcement.6
Alliances: securing supply
Memory is scarce enough that the biggest buyers are tying themselves to suppliers. Raj Mirpuri said the 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."7 HBM is a high-bandwidth type of memory used alongside AI processors. The same report says roughly $950 billion in new AI agreements were signed at a 24 July summit in San Francisco attended by South Korea's president and the heads of Samsung, SK Group, Hyundai Motor and Naver.7 That is our source's figure and we have not independently verified it. The same article says Nvidia has spent much of 2026 signing deals across Asia, the Middle East and Europe to secure chips, memory and power.7
The topic summary we work from also lists other consolidation and alliances: onsemi–Synaptics and DeepSeek–Huawei. It also names regulatory and geopolitical friction: FTC scrutiny of AI agents, OpenAI's copying accusations against Moonshot, Chinese espionage targeting AI experts and expanded NAND capacity at China's YMTC. Our source documents do not contain the detail behind these items, so we report them only as the themes in the underlying story, not as verified events.
The interconnect: moving data is the bottleneck
Mark Papermaster of AMD said: "As AI systems scale, moving data efficiently is as critical as increasing compute performance." He added that silicon photonics and advanced packaging will be key to the bandwidth, energy efficiency and system-level connectivity that next-generation AI clusters need.8 He made the remark when welcoming a letter of intent under which GlobalFoundries expects a $300 million award from the U.S. Department of Commerce's CHIPS Research and Development Office.8 The Kalray–Bull deal is a European version of the same bet: a better network between chips, not just faster chips.
Custom silicon and the heat problem
On 8 October 2026 Zeta announced it has begun work with partners on a specialized AIM chip, which it describes as a capital-light route to faster intelligence.9 The same announcement says its Athena Inference Model was developed with Fireworks using NVIDIA Nemotron.9 This is a company announcement, not an independent assessment.
A more radical approach comes from Hannah Earley, cofounder and chief technology officer of Vaire Computing, which is building chips that recycle energy usually lost as heat, a strategy called reversible computing.10 She says conventional computing is like racing through a city only to brake at every intersection, so the car loses momentum and must burn more fuel to accelerate again.10 She also said: "I want to tackle every part of how computers are built, and rethink it in these terms."10 We flag a reliability warning here: none of the 11 claims we checked from this source held up (0%).10 We quote Earley because she said it, not because the technology is proven.
Where the evidence is thin
Three limits are worth stating plainly. First, the dossier says the shortage is expected to last 12 to 18 months, but it names no source for that estimate, so treat it as an unattributed expectation. Second, the Kalray–Bull announcement comes from a source where 32% of 2,928 checked claims held up, so the strategic language should be read as the companies' own description.1 Third, the dossier contains little on EU regulation, so we cannot say how European policy bears on any of this.
What to watch
- Whether Micron's capex keeps climbing from the $5.389 billion first-quarter level, and whether cost of revenue follows.35
- Whether Micron announces buybacks now that it is technically eligible under the CHIPS Act restrictions.6
- Whether Kalray and Bull publish terms or customers for their Ultra Ethernet work.1
- Whether the GlobalFoundries letter of intent becomes a final award.8
