On Wednesday, President Trump claimed that Apple has agreed to work with Intel on designing and manufacturing chips in the United States — Intel shares surged 10.5% to ~$134, their largest single-day gain in years, on the announcement. Neither Apple nor Intel confirmed the deal directly; Apple did not respond, and Intel said only that it would not comment on "a potential Apple-Intel agreement." But reporting from Semafor and the WSJ placed the two companies in active talks for more than a year, with preliminary terms already reached — and Intel's 18A-P process, which entered risk production two days earlier, is the disclosed manufacturing vehicle.
If the relationship advances to volume production, it would give TSMC its first serious US-based competition for leading-edge Apple volume, validate the CHIPS Act's core industrial policy bet, and deliver the private-sector anchor for domestic advanced chip manufacturing that the Pentagon and Commerce Department have been seeking since Taiwan's fab concentration became a strategic flashpoint.
AMD acquired MEXT, a memory-tiering startup whose technology makes NAND flash appear as DRAM to applications through AI-driven predictive data migration — moving cold data from DRAM to lower-cost storage while maintaining DRAM-level performance for active workloads. The acquisition adds software-defined memory expansion to AMD's data center stack at a moment when HBM supply is constrained and DRAM cost per GB is rising with AI inference scale. Financial terms were not disclosed; MEXT's technology targets the TCO math that makes large model serving expensive — reducing reliance on HBM density per node in favor of tiered memory that responds predictively to model access patterns.
On Friday, Bloomberg reported that US Commerce Secretary Howard Lutnick briefed ASML's senior leadership with the US government's claim that one of ASML's EUV lithography machines reached China in violation of export controls in place since 2019. ASML denied it with specificity, presenting a global inventory of all 314 active EUV systems and 26 decommissioned units, none in China, and noting that the machines require continuous remote service and cannot be relocated without ASML's direct involvement. If the allegation is accurate, the foundational premise of the US chip control strategy has failed at its most critical single point; if inaccurate, it suggests the US is preparing diplomatic escalation ahead of the pending MATCH Act, which would extend allied-country export requirements to cover ASML's remaining DUV tools.
At VivaTech Paris on Wednesday, NVIDIA named Nscale — alongside Nebius — as its UK cloud partner for a 14,000 Blackwell GPU deployment across new UK data centres, part of a broader European sovereign AI infrastructure push in which France, Italy, and the UK are jointly backing regional cloud providers — Domyn, Mistral AI, Nebius, and Nscale — to deploy more than 3,000 exaflops of NVIDIA Blackwell systems. The NVIDIA designation converts Nscale's GPU supply access into explicit strategic endorsement at a moment when cloud providers across the EU are competing for Blackwell allocation priority; the "UK anchor cloud" framing by Jensen Huang at VivaTech positions Nscale and Nebius as the infrastructure backbone of the UK's sovereign AI buildout rather than as generic resellers. Watch the partnership terms for supply priority and co-marketing exclusivity, and whether the UK sovereign cloud model gets replicated in Germany or France under the same framework — NVIDIA's European push at VivaTech appears to be the template for a series of named national AI infrastructure commitments.
Odyssey closed a $310 million Series B at a $1.45B valuation this week, with Natural Capital leading and Amazon, AMD Ventures, GV, EQT, and In-Q-Tel participating. Founded by veterans of the autonomous vehicle industry, Odyssey builds world models: spatially and temporally coherent representations of environments that let AI systems reason about physics, causality, and agent dynamics beyond what transformer-only architectures can achieve. AWS is Odyssey's preferred cloud provider, with AWS Trainium as the disclosed compute platform.
The AI inference startup (formerly Recogni) disclosed its Napier chip earlier this month: manufactured on TSMC's 3nm process, co-developed with Broadcom and Juniper Networks, and using a proprietary logarithmic "Pareto" number format the company says cuts inference power consumption by up to 8x versus conventional floating-point approaches. Tensordyne reports over a dozen letters of intent and forecasts more than $200 million in demand, with AI infrastructure operators Cirrascale and BlueSky Compute among the named prospects; TSMC tape-out is confirmed. If the power efficiency claims hold under third-party benchmarking, Napier offers a structurally distinct point on the inference cost curve at a moment when energy costs have become the primary TCO constraint for large-scale model operators — watch for independent validation from early beta deployments.
Cerebras Q1 2026 Earnings — After market close; conference call at 2pm PT / 5pm ET. Cerebras' first quarterly results as a public company since its May 2026 IPO. Watch revenue growth trajectory, wafer-scale engine utilization rates, and management's commentary on the pipeline split between hyperscaler and enterprise customers — the latter being the segment where Cerebras most directly competes with GPU-based inference at scale.
Qualcomm Investor Day — New York City. First full disclosure of the Dragonfly data center platform: server CPU roadmap, AI inference accelerators (AI200, with AI250 to follow in 2027), and custom ASIC strategy. Management has confirmed it will name hyperscaler customers for the first time. Potential Tenstorrent acquisition announcement. Live stream available.