Silicon Nexus

NVIDIA CEO Jensen Huang met SK Group Chairman Chey Tae-won in Taipei on June 1 to mark SK Hynix crossing $1T market cap and discuss AI memory cooperation. Huang also hosted a 'Korean trillion-dollar dinner' with executives from Samsung, SK Hynix, LG, Hyundai Motor and Doosan Robotics, reaffirming Korea as a critical part of NVIDIA's chip, DRAM, robotics and AI factory ecosystem, and signaled openness to holding a GTC event in Seoul.

Why it matters: Headline-grabbing relationship signaling and confirmation of SK Hynix's $1T milestone and ongoing HBM/DRAM tie-up with NVIDIA, but no new contract, capex or specific commercial terms disclosed.

Phison (TW: 8299, not in universe) unveiled a system-level AI strategy at Computex, pairing its Pascari aiDAPTIV cache memory with Intel Core Ultra to run 26B-parameter models on 16GB DRAM PCs (vs 32GB normally required). It also launched the 245.76TB Pascari D206V enterprise SSD (Computex Gold Award), a 14.9 GB/s E37T PCIe Gen5 DRAM-less SSD, a 10 GB/s UFS 5.0 controller, and PCIe 6.0 retimer/redriver chips, positioning Phison as an end-to-end AI infrastructure-to-edge supplier alongside Asus, MSI, Acer and Ollama.

Why it matters: Computex product showcase and strategy pivot for Phison (not in tracked universe); no direct capex, contract, or earnings catalyst for tracked TW/KR names, though it touches NAND controller and AI PC supply-chain themes relevant to memory peers.

At Computex, Nvidia CEO Jensen Huang declared the shift to Agent AI, positioning the PC for its biggest reinvention in 40 years as an 'Agentic Computer.' Nvidia unveiled RTX Spark, a new AI PC architecture co-developed with Microsoft and MediaTek (2454) over nearly two years under codename N1X, using unified memory with CPU, GPU and Tensor Cores. Huang also pitched Vera CPU for agent workloads and argued Agent AI will expand—not cannibalize—software demand.

Why it matters: MediaTek (2454) is named as Nvidia's nearly two-year co-development partner on the RTX Spark / N1X AI PC platform, a concrete product reveal with direct revenue implications.

NVIDIA CEO Jensen Huang said N1X is a long-term platform architecture, with N2X and N3X already in development plus a smaller N1 variant coming, framing the lineup as a full AI PC software-stack play rather than single hardware products. Huang also outlined a disaggregated/distributed AI compute model spanning cloud, enterprise and edge devices like DGX Spark, reinforcing on-device inference as the future direction for AI Agents.

Why it matters: Roadmap commentary on NVIDIA's AI PC platform family with no specific capex, contract, or named supplier — sector/roadmap color rather than a direct stock-moving catalyst, though TSMC as fab partner has indirect read-through.

Intel disclosed more specifics on its next-gen Diamond Rapids Xeon at Computex 2026: built on Intel 18A-P with chiplet/Foveros packaging, up to 192 cores (+50% vs prior gen), PCIe 6.0, 16-channel DDR5 delivering ~1.2 TB/s bandwidth, but SMT is removed this generation. Launch slips to 2027, trailing AMD's Zen 6 Venice EPYC, with follow-on Coral Rapids targeted for 2028. Full architecture/IPC details come at Hot Chips in August.

Why it matters: Server CPU roadmap update with implications for the AI/data-center supply chain (foundry, advanced packaging, HBM/DDR5 memory) but no direct contract or capex disclosure tied to listed names.

Nvidia CEO Jensen Huang confirmed the US has approved chip sales to China and pledged to serve the market once local demand is ready, while spotlighting Taiwan partners TSMC, ASE, Wistron, and Foxconn for their US investments. He also disclosed a new 4,000-seat Taipei HQ on top of ~1,000 current staff, positioning Nvidia as one of Taiwan's largest employers, and backed high worker pay when asked about Samsung's labor deal.

Why it matters: Sector/supply-chain commentary with confirmation of US export approval for China and a concrete Taiwan headcount plan, but no specific capex numbers or contract awards that would move stocks directly.

At a global press conference, Jensen Huang said Nvidia will sell the upcoming Vera CPU in China if export and import rules permit, framing Vera as a purpose-built AI-agent processor with 3–6x performance vs mainstream CPUs, ~3x per-core memory bandwidth and 2–3x I/O/interconnect bandwidth. Huang positions Vera as complementary to Nvidia GPUs (following Grace) rather than a share-grab from AMD/Intel, implying tighter Nvidia GPU+CPU bundling in AI data centers and incremental TSMC advanced-node and HBM demand.

Why it matters: Roadmap/strategy commentary on Vera CPU positioning and China access — directionally bullish for Nvidia's TSMC and HBM supply chain but no contract, capex or timeline disclosure.

Bloomberg's review of procurement records found at least 7 Chinese military-linked universities — including blacklisted Beihang and Northwestern Polytechnical — seeking to rent H200 compute hosted offshore, sidestepping US export controls. Over 25 PLA-affiliated institutions are also using or pursuing older A100/H100/H20 chips, raising fresh congressional scrutiny that could tighten future Nvidia China access.

Why it matters: Geopolitical/export-control story with no direct revenue impact on tracked TW/KR names, though potential tightening of US-China chip policy is a sector overhang worth flagging.

Phison (8299-TWO, not in universe) unveiled an 'AI Enabler' strategy at COMPUTEX 2026, expanding from SSD controllers into an integrated AI infrastructure, edge AI and software platform play anchored by its aiDAPTIV tech that uses NAND Flash as a memory-extension layer to reduce DRAM/HBM dependency. The Pascari AI20EH SSD claims up to 102x faster inference, 67% less memory usage and 53% lower on-prem AI deployment cost, and Phison flagged deep tie-ups with Intel (Core Ultra/OpenClaw) and MediaTek's Dimensity 9500 for hybrid UFS 4.1 on-device AI.

Why it matters: Roadmap/strategy story centered on Phison (not in tracked universe); MediaTek appears only as a peripheral Dimensity 9500 partner, with no near-term earnings or contract impact disclosed.