Qualcomm pushes into data-center AI as SK Hynix races Samsung to ship HBM4E early
Zero One Daily Intelligence Brief — June 16, 2026
Qualcomm pushes into data-center AI as SK Hynix races Samsung to ship HBM4E early
Top Market Signals
Qualcomm pushes into data-center AI inference
Qualcomm (QCOM US) is making its largest move yet toward the data center, in talks to acquire AI chip startup Tenstorrent while AWS is reported to be lining up Qualcomm's AI200 inference accelerator. The stock rose more than 4%.
According to The Information, Qualcomm is negotiating to buy Tenstorrent, which designs accelerators for AI training and inference. The deal would hand Qualcomm a chip team and intellectual property aimed squarely at the data-center workloads it has so far sat out, beyond the smartphone modems and SoCs it is known for.
Separately, DigiTimes reports that Amazon's AWS plans to adopt Qualcomm's AI200, an inference chip approaching mass production, to lower the cost of serving AI models. The timing fits a broader shift from model training toward large-scale inference deployment, where cost per query matters more than peak training throughput.
Both moves point Qualcomm at the cost-sensitive inference layer rather than NVIDIA's training stronghold. It is the same opening AMD and a wave of custom-silicon startups are chasing: inference is where the addressable market broadens beyond a single dominant supplier.
Linked stocks: QCOM US, NVDA US, AMZN US
Sources: Google News, AWS傳導入高通AI200 搶攻低成本AI推論商機
NVIDIA returns to the bond market after five years
NVIDIA (NVDA US) has filed to raise at least US$20bn through corporate bonds, its first debt issuance since 2021, joining a wave of Big Tech borrowing to fund AI infrastructure. CNBC first reported the plan.
The scale tells the story of NVIDIA's growth. Its last bond, in 2021, raised US$5bn. Revenue has since climbed from about US$27bn in FY2022 to US$216bn in FY2026. The company carries roughly US$7.5bn of long-term debt, and management says the proceeds are for general corporate purposes including refinancing existing debt.
This is not a cash problem. NVIDIA's most recent quarterly free cash flow was US$49bn, up from US$35bn a year earlier, against an US$80bn buyback it announced in May 2025. The borrowing mirrors Alphabet, which raised more than US$55bn in new bonds plus an US$85bn equity-linked plan, Amazon at about US$54bn issued in 2026, and Super Micro at US$7bn.
The read-through: even the most cash-generative company in AI is using debt to pre-fund the buildout, a measure of how capital-hungry the GPU cycle has become. NVIDIA shares rose 3.5% and are up about 14% year to date.
Linked stocks: NVDA US, GOOGL US, AMZN US
Sources: 時隔 5 年輝達將再發行公司債,預計籌資 200 億美元, Google News
SK Hynix races Samsung to ship HBM4E early
SK Hynix (000660 KS) has pulled forward its HBM4E sample timeline, readying shipments to key customers including NVIDIA as soon as June to July, according to TrendForce. HBM4E is the next generation of high-bandwidth memory, the stacked DRAM that feeds AI accelerators. The stock jumped 7% to lead Korean equities.
The acceleration comes as Samsung (005930 KS) is reported to be moving ahead on its own HBM4E samples for NVIDIA, sharpening a two-way race to qualify next-generation stacks first. Korean brokerages have raised long-term profit estimates for both names on the AI memory super-cycle.
Our 9 June 2026 Blade flagged Nanya deepening its LPDDR tie-up with SK Hynix. Today's pull-forward shows the same dynamic one level up: memory suppliers are now competing on qualification timing, not only on capacity, to hold their share of NVIDIA's allocation.
HBM remains the tightest link in the AI supply chain, and an earlier qualification window is a direct lever on share of wallet with the dominant buyer.
Linked stocks: 000660 KS, 005930 KS, NVDA US
Sources: Google News, Google News
TSMC builds panel-level packaging line against Samsung
TSMC (2330 TT) is building a mass-production system for panel-level packaging (PLP), which packages chips on rectangular panels instead of round wafers, and has reportedly secured a major global AI chip customer. South Korea's ETNews first reported the plan, which sets up a direct contest with Samsung.
The economics favor the switch. A mainstream 600x600mm panel yields five to six times the chip output of a 300mm wafer and removes the edge waste left when a round wafer is diced into square dies. That matters most for the increasingly large, complex AI accelerators TSMC now packages. TSMC plans a pilot line this year and mass production as early as next year.
Samsung holds the lead, having run PLP since it absorbed Samsung Electro-Mechanics' PLP unit in 2019 for mobile processors and power-management chips, and is now extending it to high-performance computing and AI. The next battleground is glass substrates, where both intend to compete.
For Taiwan, a TSMC-led push lifts the back-end chain. ASE Technology (3711 TT) already runs a 310mm PLP automation line targeting volume in 1H2027, Powertech reports fan-out PLP yields at 95% with customer validation in 2H, and substrate makers Unimicron (3037 TT), Nan Ya PCB and Kinsus stand to gain on larger board area per chip.
Linked stocks: 2330 TT, 3711 TT, 005930 KS, 3037 TT
Sources: 韓媒:台積電面板級封裝PLP晶片明年量產 與三星競爭領先地位, 台積電拚面板封裝 原料、零組件、設備廠等入列 相關類股喊衝
China's Nexchip overtakes Vanguard as No.8 foundry
China's Nexchip (Hefei) overtook Taiwan's Vanguard International Semiconductor (5347 TT) for the first time to become the world's eighth-largest foundry in 1Q26, as the global foundry market set another record, per DigiTimes.
The move reflects China's continued mature-node expansion, with Nexchip scaling CMOS image sensor (CIS) and display-driver output, while front-loaded TV, PC and notebook orders lifted overall foundry demand into the quarter.
This is a lasting shift in the mature-node ranking, not a one-quarter blip. Chinese capacity is climbing the table even as advanced-node leadership stays with TSMC. For Vanguard and other Taiwanese commodity-node makers, the competitive pressure on legacy processes is set to build rather than fade.
Linked stocks: 5347 TT, 2303 TT
Sources: 合肥晶合首度超越世界先進 躋身全球晶圓代工第8
Foxconn and Schneider partner on AI data center design
Foxconn (2317 TT) and France's Schneider Electric will jointly develop a reference architecture for next-generation AI data centers, pairing Foxconn's compute platforms and rack integration with Schneider's power, cooling and energy-management systems. Joint production starts later this year, and Foxconn shares rose 2.7% to NT$267.5.
The partners are targeting closed-loop energy optimization, modular power and cooling, and standardized designs, aiming for a repeatable "AI factory" blueprint as data-center power draw and heat density climb.
Power and cooling have become the gating constraint on AI buildout, and the tie-up pushes Foxconn beyond server assembly into full infrastructure delivery. It follows the company's broader move into AI data-center systems for hyperscaler and sovereign customers.
Linked stocks: 2317 TT, SU FP
Sources: 加速次世代 AI 資料中心發展,鴻海與施耐德宣布策略合作, 鴻海與施耐德攜手策略合作 加速次世代AI資料中心發展進程
先聲 First Word — Exclusives from Chinese-Language Sources
Hon Precision to sample CPO test gear in 4Q26
Hon Precision (7769 TT) told its annual meeting that its co-packaged optics (CPO) test equipment will go to customers for validation in 4Q26, with existing AI and high-performance-computing test-handler orders filling capacity; it plans to expand capacity by about 50% next year. Keeps a Taiwan back-end equipment maker on the CPO ramp as optical interconnect moves toward volume. (15 Jun 2026) Source: 鴻勁明年產能拚增5成 CPO設備Q4送樣驗證
Zhen Ding lifts capex to chase AI substrate orders
Zhen Ding-KY (4958 TT), the PCB leader, bought NT$1.225bn of exposure equipment and raised 2026 capex above NT$80bn to expand high-end PCB, optical-module and IC-substrate capacity; it targets doubling server and optical-module revenue and over 80% growth in IC-substrate revenue this year. Signals confidence that next-generation AI platform ramps will sustain substrate and optical demand. (15 Jun 2026) Source: 臻鼎砸12.25億元採購設備 加速高階 PCB、IC 載板擴產
PowerColor deepens AMD tie, targets 600k boards
PowerColor (6150 TT) told shareholders it will deepen its partnership with AMD and expects total graphics-card and board shipments of about 600,000 units in 2026, after swinging to a profit last year on edge-AI demand and an industrial-PC recovery. It also flagged rising component costs from the memory, PCB and packaging squeeze, a read on supply pressure reaching consumer hardware. (15 Jun 2026) Source: 撼訊深化AMD合作關係 今年板卡整體銷量上看60萬片
© 2026 Zero One Investment Research Pte Ltd. For informational purposes only. Not investment advice.