Zero One Daily Intelligence Brief — April 23, 2026

Memory Supercycle Extends to 2029 as TSMC Maps A13 and Intel Wins Tesla

Top Market Signals

SK Hynix sells out HBM for three years

SK Hynix (000660 KS) posted record first-quarter results with operating profit up roughly five-fold YoY, and management told investors HBM capacity is now sold out for three years as DRAM supply runs short. Reported operating margin reached 72%, underscoring the pricing power concentrated in the AI memory leader. The company flagged a prolonged chip supercycle with supply constraints expected to persist for another three years, and it is preparing to boost capex, build a new Korean fab, and purchase additional ASML EUV tools. Management is also exploring a US ADR listing.

The read-through for the broader memory complex is more nuanced. Micron (MU US) shares fell on the SK Hynix print, and Omdia raised its 2026 semiconductor forecast to 62.7% growth citing AI-driven memory tightness. Samsung Electronics (005930 KS) faces parallel profit disclosure pressure as tens of thousands of workers rallied outside the Pyeongtaek chip hub demanding a bigger cut of AI earnings, with three allied unions threatening a multi-week strike. The pay gap with SK Hynix is explicitly cited in the protests.

For the AI stack, the implication is that HBM and leading-edge DRAM remain a binding constraint through 2028E, underwriting continued pricing power at the memory incumbents and creating opportunity for adjacencies. Rambus (RMBS US) unveiled a SOCAMM2 chipset for LPDDR5X memory in AI servers today, and NEO Semiconductor completed proof-of-concept for its 3D X-DRAM technology with a strategic investment led by Acer (2353 TT) founder Stan Shih, aimed at filling Taiwan's AI memory design gap.

Linked stocks: 000660 KS, 005930 KS, MU US, RMBS US, 2353 TT Sources: Google News, Google News, Google News, Bloomberg, 宏碁施振榮投資記憶體設計廠NEO 力促3D X-DRAM技術

TSMC maps A13 and A12, sidesteps High-NA EUV

TSMC (2330 TT) unveiled a process roadmap extending to 2029 with new A13 "1.3nm" and A12 "1.2nm" nodes, and opted to sidestep ASML's (ASML NA) most expensive High-NA EUV tools for the current generation. The A13 node offers a 6% smaller chip area versus the prior generation. ASML shares lost nearly $17bn of market value on the news as investors reassessed the timing of High-NA adoption. TSMC is also planning a chip packaging plant in Arizona by 2029, extending its US footprint beyond wafer fabs into advanced packaging.

The ecosystem is already pulling in the new nodes. M31 (6643 TT) announced its eUSB2V2 IP has completed tapeout on TSMC's N2P process, presented at the foundry's North America Technology Symposium. Cadence (CDNS US) and Siemens (SIE GR) both announced expanded TSMC collaborations for AI chip design at the event. Bolt Graphics taped out its first Zeus GPU test chip on TSMC 12nm, touting a 17x lower cost of compute claim.

For investors, the A13/A12 disclosure fixes the foundry roadmap through 2029 and resets the High-NA EUV investment narrative. Taiwan's Financial Supervisory Commission also signaled confidence in TSMC's index dominance with a separate regulatory move detailed below.

Linked stocks: 2330 TT, TSM US, ASML NA, 6643 TT, CDNS US, SIE GR Sources: Google News, Google News, Google News, Google News, Tom's Hardware

Taiwan FSC lifts single-stock cap to 25% for TSMC exposure

Taiwan's Financial Supervisory Commission announced it will raise the single-stock investment cap for domestic equity mutual funds and active ETFs from 10% to 25%, effective upon tomorrow's order publication. Only stocks whose weighting in the TAIEX exceeds 10% qualify, which today captures only TSMC at roughly 44.3% of the index; Delta Electronics (2308 TT) at 3.48% and Hon Hai (2317 TT) at 2.54% are the next largest constituents. The total of single-stock holdings plus that company's financial debentures still cannot exceed 25% of fund NAV.

The regulator estimates roughly NT$200bn of fund capital could now increase TSMC allocations, though Deputy Director General Huang Chung-hao stressed this is not the policy's intent but rather removes a structural constraint on portfolio construction. The move follows a May 2025 easing of passive ETF single-stock limits and a concurrent approval for foreign-currency cash dividends. There are currently 138 domestic equity funds with NT$1.05tn of AUM and 14 active ETFs with NT$231bn.

This is the third regulatory tranche removing TSMC-specific constraints in under a year, reflecting the widening gap between TSMC's index weight and mutual fund mandates. The concentrated effect is a tailwind for TSMC flows even as broader valuation concerns surface; analyst commentary cited Moore Securities noting TSMC's 21x PE has arguably priced in 2026 earnings growth.

Linked stocks: 2330 TT, TSM US, 2308 TT, 2317 TT Sources: 金管會大鬆綁！台股基金、主動ETF買台積電上限放寬至25％, 台股基金、主動ETF投資台積電上限 金管會放寬至25%, 金管會鬆綁投資限制 估計近2千億基金活水將加碼台積電, Google News

Musk picks Intel 14A for Tesla's Terafab

Elon Musk used Tesla's (TSLA US) earnings call to reveal plans to build AI chips on Intel's (INTC US) not-yet-finished 14A process as the silicon anchor for the Terafab project. Musk said "14A seems like the right move" despite the process still being in development, citing the need for Tesla to own its silicon roadmap. Intel shares have rallied roughly 235% over the past 12 months to their highest level since the dot-com bubble, and the stock rose further into tonight's Q1 print as analysts frame the foundry business as the real earnings test.

Tesla's commitment is a meaningful external validation of Intel Foundry Services at a moment when TSMC's A13 roadmap caps the leading-edge premium. Tesla also disclosed plans to spend as much as $25bn on AI and robotics this year, materially above prior guidance. For Intel, a hyperscaler-grade customer using 14A for volume AI silicon is precisely the foundry proof point the stock has been pricing.

Linked stocks: INTC US, TSLA US, 2330 TT Sources: Theregister, Google News, Google News, Bloomberg

Google sharpens silicon stack against Nvidia

Alphabet's (GOOGL US) Google unveiled a new generation of AI chips designed to train large language models and power AI agents, along with an enterprise agent development platform. The line includes an inference-focused chip aimed at the rising share of workloads consumed by deployment rather than training. The timing puts additional competitive pressure on Nvidia (NVDA US), whose TPU-ASIC alternative narrative is now concretely paired with a Google-run agent platform targeting enterprise buyers.

Adoption of agentic AI is accelerating on parallel fronts. Microsoft (MSFT US) launched Agent Mode in Word, Excel and PowerPoint this week, extending what it previously called "vibe working." Microsoft also integrated Anthropic's Claude Mythos preview into its Secure Development Lifecycle tooling. Google's CEO separately disclosed that about 75% of new code at Google is now AI-generated and engineer-approved, a concrete productivity data point for the coding-agent thesis. Anthropic reached a $1tn secondary-market valuation on reports of a 233% revenue surge, overtaking OpenAI in the private-market benchmark.

For investors, the enterprise AI stack is consolidating around a handful of model/agent platforms, with custom silicon becoming table stakes for hyperscaler differentiation. The inference-chip focus suggests the next leg of AI capex intensity shifts from training clusters to deployment infrastructure.

Linked stocks: GOOGL US, NVDA US, MSFT US, AVGO US Sources: Google News, Google News, The Verge, Google 轉向代理式工作流程，執行長：75% 新程式碼已由 AI 產生, Google News

Nokia rides AI data-center wave as telecom capex diverges

Nokia (NOKIA FH) reported Q1 results that lifted shares to a 16-year high and prompted management to raise its outlook for AI and Cloud, with CEO Justin Hotard stating AI demand accelerated significantly since its November 2025 strategy revamp. The Finnish vendor said its AI data-center networking business is now a primary growth driver, while its mobile infrastructure business continues to lag. Nokia's performance contrasts with broader carrier capex softness, and Light Reading observed its "long-suffering mobile business desperately needs a revival."

The same AI-driven fiber and data-center demand lifted other telecom names today. Orange (ORA FP) raised full-year earnings guidance after Q1 revenue and EBITDA growth at home and in Middle East/Africa, which CEO Christel Heydemann again cited as the company's primary growth engine. AT&T (T US) reported record Q1 fiber and fixed-wireless-access customer additions, validating its convergence strategy. BT Group (BT/A LN) partnered with Nvidia-backed Nscale to build up to 14MW of UK sovereign AI data-center capacity using Nvidia's full-stack infrastructure. Zayo's long-haul business signed what it called a "monster" contract with an AI infrastructure anchor tenant. Comcast (CMCSA US) beat Q1 estimates on fewer broadband losses.

For investors, the message is that carrier AI-adjacent infrastructure (long-haul fiber, sovereign AI capacity, data-center networking) is where the upside is concentrated, while legacy mobile access remains commoditized. Nokia is now explicitly a levered play on AI networking rather than a pure-play RAN vendor.

Linked stocks: NOKIA FH, ORA FP, T US, BT/A LN, CMCSA US, NVDA US Sources: Telecoms, Lightreading, Telecoms, Telecoms, Mobileworldlive, Fierce-network, Bloomberg

Samsung labor risk meets China chip import friction

Three Samsung Electronics (005930 KS) unions rallied tens of thousands of workers outside the Pyeongtaek chip hub demanding profit sharing and removal of bonus caps, threatening a multi-week strike. The protests explicitly reference the widening pay gap versus SK Hynix, whose record margins have become a visible benchmark for rank-and-file compensation demands across the memory industry. A multi-week strike at Samsung's largest fab complex would be a material supply disruption at a moment when HBM and DRAM are already tight.

Separately, US Commerce Secretary Howard Lutnick disclosed that Nvidia still has not sold a single H200 AI GPU to China despite the export ban being lifted four months ago, with the Chinese government actively blocking imports to favor domestic silicon. Tencent (700 HK) used the backdrop to unveil a major upgrade to its foundation model, the first high-stakes test for its recently hired ex-OpenAI researcher. The dynamics point to a bifurcating global AI stack: China's domestic buildout is accelerating while its end-demand for US leading-edge accelerators has structurally stalled.

Wet-process equipment maker Gallant Precision (3131 TT) also denied media reports that core technology leaked to China and disclosed it has filed suit against its former CEO Huang Fu-yuan for alleged trade-secret infringement, with prosecutors now engaged. The case adds to the running theme of tightening Taiwan equipment-supplier IP perimeter.

Linked stocks: 005930 KS, 000660 KS, NVDA US, 700 HK, 3131 TT Sources: Bloomberg, Google News, Tom's Hardware, Bloomberg, 弘塑重訊公告 否認設備技術外流中國

先聲 First Word — Exclusives from Chinese-Language Sources

Pegatron commits NT$1.32bn to new Mexico plant for auto and AI servers

Pegatron (4938 TT) disclosed that its Mexican subsidiary is acquiring land and a factory building in Ciudad Juárez for US$42m (NT$1.32bn), with the new plant earmarked for automotive electronics and AI server production as existing Mexican capacity approaches full utilization. The expansion extends Pegatron's nearshoring footprint for its electric-vehicle customer and builds out a second AI server production base outside Asia. (23 Apr 2026)

Source: 和碩搶攻北美商機 砸13.2億元墨西哥購置新廠

Asia Optical partners with Frore Systems on AI data-center liquid cooling

Asia Optical (3019 TT) signed an MoU with US-based Frore Systems to co-develop liquid-cooling technology for AI data centers, with a plan to develop advanced liquid-cooling production at its Philippines factory. Frore's LiquidJet product uses a 3D short-loop jet-channel cold plate claimed to materially outperform traditional cold plates on GPU performance. The deal marks Asia Optical's formal expansion from optical components into thermal systems for hyperscale customers. (23 Apr 2026)

Source: 亞光攜手美國Frore Systems公司 合作AI資料中心液冷散熱技術

Artery Technology sees pull-forward demand lift 2026 revenue 60%

MCU vendor Artery Technology (6907 TT) Chairman Wang Kuo-yung told its earnings call that customers are pulling forward orders into 1H26 to hedge price-rise risk, with the revenue peak expected in 2Q and the full year tracking up roughly 60% YoY. Operating expenses will also step up on employee bonuses and expanded R&D, signaling margin dilution even as top-line accelerates. (23 Apr 2026)

Source: 客戶提前大拉貨 雅特力-KY今年營收估激增6成

Sinbon teams with Weijie Energy on hydrogen fuel cells for AI data centers

Cable-assembly maker Sinbon Electronics (3023 TT) announced a partnership with Taiwan hydrogen specialist Weijie Energy to develop hydrogen power systems for micro-EVs, commercial drones, mobile microgrids and site-level AI compute power supply. Sinbon posted 1Q26 EPS of NT$3.60 and guided 2Q better than 1Q as new products ramp. The deal opens a backup-power angle for Sinbon on the AI data-center buildout. (23 Apr 2026)

Source: 信邦攜手威杰能源合作氫能 鎖定算力中心、商用無人機等場域應用

Heran appliances doubles down on value brand Yamada

Heran (5283 TT) told its analyst meeting its second brand Yamada grew 50% YoY in 1Q26 and now accounts for roughly 10% of revenue, with future dividend policy shifting toward stock dividends to fund the brand push. After seven consecutive years of EPS above NT$10, 2025 EPS slipped to NT$5.63 on rising plastic-resin input costs and market-share competition, and 2026 price actions are being staged on new SKUs. (23 Apr 2026)

Source: 全力衝第二品牌 禾聯碩：股利政策未來以配股為主