NVIDIA builds Taiwan campus as chip CEO convergence makes COMPUTEX an AI summit
Zero One Daily Intelligence Brief — May 25, 2026
NVIDIA builds Taiwan campus as chip CEO convergence makes COMPUTEX an AI summit
Top Market Signals
NVIDIA breaks ground on permanent Taiwan campus
NVIDIA (NVDA US) is establishing a permanent Taiwan headquarters, with a groundbreaking ceremony for the "NVIDIA Constellation" campus on T17 and T18 land parcels in the Taipei area scheduled for May 27, the first time NVIDIA has committed to a physical headquarters presence inside its primary supply-chain ecosystem.
Jensen Huang arrived in Taiwan on May 23, a week ahead of COMPUTEX, which opens June 2. The Constellation campus is not a manufacturing facility: it is designed to house engineers, researchers, and partner-coordination teams as NVIDIA deepens integration with Taiwan's supply chain spanning TSMC (2330 TT), Quanta (2382 TT), Foxconn (2317 TT), and the packaging and ODM ecosystem around them. NVIDIA has previously operated through representative offices; this is a structural upgrade to a permanent footprint.
Intel CEO Lip-Bu Tan also arrived in Taiwan ahead of COMPUTEX and announced that Intel and NVIDIA are jointly developing new products combining NVIDIA's NVLink interconnect with Intel's Xeon server processors. A volume NVLink-on-Xeon product would make Intel a direct component supplier inside NVIDIA's AI infrastructure stack, adding a revenue line from the AI data-center market that Intel's own GPU program has not yet delivered.
The June 2 COMPUTEX roster is the densest in recent memory: AMD's Lisa Su, Marvell's Matt Murphy, Qualcomm, Arm, and NXP have all confirmed keynote or major presentation slots alongside Huang's keynote. Taiwan's chip exhibition has functionally become the industry's primary AI roadmap event, where supply-chain depth and manufacturing context are available in ways that US-only venues cannot match.
Linked stocks: NVDA US, 2330 TT, 2317 TT, 2382 TT
Sources: 經濟日報, COMPUTEX 成為全球 AI 競爭最前線
Huawei claims 1.4nm chip process to rival TSMC
Huawei (unlisted) announced it has developed 1.4nm chipmaking technology and intends to use it to compete directly with TSMC, which is currently ramping its 2nm N2 process for NVIDIA, Apple, and AMD.
The claim, reported by Taipei Times and Notebookcheck, has not been verified by third-party process analysis. Huawei operates under US export controls that restrict access to ASML's EUV lithography, the primary tool for achieving leading-edge density at 3nm and below without workarounds across multiple process steps. Chinese foundries have previously announced node equivalences that, on closer examination, reflected different process characteristics and density metrics than the TSMC nodes they cited.
The announcement is significant regardless of exact process parity, because it compresses the perceived leading-edge gap between TSMC and Chinese foundries at the precise moment TSMC is introducing 2nm and planning its A14 (1.4nm-class) node. AMD's next Zen 7 CPUs are reportedly bound for TSMC A14 (see next signal). Intel's 18A node, which is approximately 1.8nm-equivalent, is already running Apple's first external foundry production. Huawei claiming the same node class as these Western supply-chain leaders adds urgency to the geopolitical risk calculus around advanced foundry concentration.
Linked stocks: 2330 TT, INTC US, AMAT US
Sources: Google News, Google News
AMD's next flagship CPUs target TSMC's 1.4nm node
AMD (AMD US) is reportedly planning to produce its Zen 7 "Grimlock" CPUs on TSMC's A14 (1.4nm) process node with advanced packaging, according to Notebookcheck and TweakTown, a commitment that would extend AMD's sole-source foundry dependency on TSMC through the next CPU generation.
AMD has not officially confirmed the node selection for Zen 7, which follows the current Zen 5 (on TSMC 3nm) and the in-development Zen 6. TSMC's A14 node is the successor to N2, with volume production expected from 2027 onward. Committing the flagship CPU die to A14 at this stage extends AMD's advanced-node relationship with TSMC by at least three to four years.
From our May 11, 2026 Blade, Apple's first external foundry win went to Intel 18A, establishing that Intel could attract large-volume, process-sensitive customers. AMD's reported A14 commitment indicates the most demanding x86 server CPU workloads still default to TSMC for leading-edge process performance. The sub-2nm landscape now shows three declared positions: TSMC A14 (AMD confirmed for Zen 7), Intel 14A (in roadmap, no volume customers disclosed), and Huawei's just-announced 1.4nm claim. TSMC's backlog advantage for volume production is the market's working assumption, but today's signals show the race is actively contested at every front.
Linked stocks: AMD US, 2330 TT, INTC US
Sources: Google News, Google News
Samsung 900-layer NAND sets density world record
Samsung Electronics (005930 KS) has demonstrated a 900-layer NAND Flash prototype using a Cascaded Merging Bonding (CMB) technique that doubles achievable stack height versus current production methods, setting a new NAND density record and extending Korea's lead in Flash memory technology.
CMB works by separately fabricating two 450-layer stacks and bonding them into a single 900-layer structure after individual processing. This sidesteps the key physical constraint of direct stacking: as layers accumulate in a single build sequence, wafer stress and warpage become progressively harder to control, capping practical stack heights. Samsung's approach separates the manufacturing challenge into two tractable sub-problems and then combines the outputs, a similar logic to chip-on-wafer-on-substrate (CoWoS) packaging that separates compute and memory fabrication.
Kioxia is planning its 10th-generation BiCS 10 NAND for 2027 volume production, with Samsung and SK Hynix (000660 KS) both compressing their own next-generation timelines in response. The 900-layer Samsung result is a research demonstration, not a confirmed production schedule, but it sets the density ceiling that Kioxia and SK Hynix now plan against. The financial context: Samsung's equipment subsidiary SEMES reported 1Q26 revenue up 106% YoY, reflecting the internal investment intensity behind Samsung's NAND technology program.
Linked stocks: 005930 KS, 000660 KS, MU US
Sources: Google News, DigiTimes
TSMC offers faster bonuses after Samsung-style strike threat
TSMC (2330 TT) has committed to accelerating bonus growth for its workforce after employees began coordinating grievance submissions under an anonymous escalation mechanism, a labor dynamic that reflects the structural leverage of a workforce responsible for generating more than 15% of Taiwan's total export value.
TSMC workers cited the company's growing economic centrality as justification for a larger profit share. TSMC generates a disproportionate share of Taiwan's exports relative to its headcount, giving any credible strike threat consequences well beyond the company's own operations. Management's commitment to faster bonus growth signals its assessment that a work stoppage would cost more than the wage concession.
From our May 21, 2026 Blade, Samsung Electronics resolved its own worker dispute through a stock-based bonus pool tied to AI memory profits, establishing a new benchmark for how semiconductor conglomerates handle labor pressure in a tight AI-supply cycle. TSMC's workforce is watching the same pattern. The difference is that TSMC operates without a union structure comparable to Samsung's, which means management retains more flexibility on the terms of any commitment, but also faces harder-to-predict coordination dynamics if negotiations stall.
Linked stocks: 2330 TT, 005930 KS, 2303 TT
Sources: Google News, TSMC 擴大社會永續計畫員工紅利將增加
ASML builds Taiwan factory for next-gen EUV
ASML (ASML NA) is adding approximately 200 more staff in Taiwan by end-2026, bringing its Taiwan headcount to roughly 1,000 employees, and building a local factory for maintenance and servicing of its high-numerical-aperture EUV (extreme-ultraviolet) systems, the next-generation lithography equipment required for chipmaking at the 1.4nm node and below.
ASML does not build local service infrastructure in markets where it expects short or uncertain equipment cycles. Taiwan absorbs a disproportionate share of ASML's EUV shipment volume because TSMC is the only foundry globally producing at 2nm and below at high volume. High-NA EUV tools, which will be required for A14 and beyond, demand on-site technical support that justifies permanent local manufacturing infrastructure rather than fly-in service teams.
From our May 21, 2026 Blade, ASML CEO Christophe Fouquet stated that the chip shortage is a multi-year phenomenon. The Taiwan factory buildout provides a tangible operational anchor for that position: a 1,000-person manufacturing and service organization in Taiwan is a decade-scale commitment, not a hedged one. Analyst targets on ASML have been revised higher on the Taiwan hiring news, per AD HOC News reporting, as the physical investment provides a demand-cycle backing that CEO commentary alone does not.
Linked stocks: ASML NA, 2330 TT, INTC US
Sources: ASML 計畫在台灣新增約200人，加碼年底前完成新廠及工廠人員招募, Google News
先聲 First Word — Exclusives from Chinese-Language Sources
CATL bids to join DeepSeek's first financing round
CATL, China's largest battery manufacturer, is planning to participate in DeepSeek's first external financing round, according to The Information, as DeepSeek targets a valuation of approximately US$45bn ahead of a June 2026 close. CATL's entry signals that China's industrial majors are treating DeepSeek's AI capabilities as operational infrastructure for battery management and energy systems, not just a technology investment. (25 May 2026)
Source: 寧德時代擬參與 DeepSeek 融資首輪
Taiwan's TAIEX hits 43,644 as NVIDIA drives record foreign inflows
TAIEX closed at 43,644 on May 25, with three major institutional investors net-buying NT$64.76bn, of which foreign investors accounted for NT$49.45bn, a 3-session high. The five-session cumulative net foreign buy reached its highest level since late 2025 as Jensen Huang's Taiwan campus announcement and pre-COMPUTEX momentum pushed semiconductor names higher. TSMC (2330 TT) rose NT$55 to NT$2,310. (25 May 2026)
Source: 台股漲幅達近5月來最大，外資大買1,376億元…主力股最強5檔
MediaTek target lifted 32% to NT$5,922 on Google TPU confirmation
An analyst raised MediaTek (2454 TT) from NT$4,500 to NT$5,922 (+31.6%) with a "High Conviction Outperform" rating after confirming the company secured the core compute center role in Google's next-generation customizable TPU. MediaTek's stock stood at NT$4,245 on May 25, implying 39% upside to the new target. (25 May 2026)
Source: 經濟日報
Micron closes Virginia DDR4 fab, pivots to next-gen DRAM node
Micron Technology has confirmed the closure of its Manassas, Virginia facility, which produced DDR4 and LPDDR4 DRAM, replacing output with its 1-alpha node DRAM, the company's most advanced production process. The closure is the industry's clearest signal yet that legacy DDR4 economics can no longer justify retaining older process nodes alongside the next-generation ramp. (25 May 2026)
© 2026 Zero One Investment Research Pte Ltd. For informational purposes only. Not investment advice.