Susana SchwartzSK hynix moves from components to broader AI architecture
Last Friday in Silicon Valley, South Korea’s SK hynix launched a new corporate venture capital (CVC) brand, SK hynix Ventures, which complements last month’s high-profile $29 billion (USD) share buyback program.
Designed to allay Wall Street analysts’ fears that its core memory business had become too cyclical and vulnerable to any AI infrastructure slowdowns, this new venture is part of a dual-track strategy to stabilize stock price volatility while also seeding new infrastructure technologies – and all without further bloating its record-breaking R&D investment of 6.73 trillion South Korean won ($4.7 billion USD).
The buyback will return cash to shareholders, while the VC will inject excess capital into startups, helping SK hynix maintain its pledge to return more than 50% of cumulative FCF (free cash flow) to shareholders.
By co-creating innovations with start-ups already working on data center power efficiency and optical networking, SK hynix hopes to embed itself into next-gen hardware designs, while also future-proofing its existing AI memory leadership through its NVIDIA alliance (which is said to have secured about two-thirds of all HBM4 allocations for NVIDIA’s newest Vera Rubin AI supercomputer platform).
Whether SK hynix will succeed in expanding its data center market share through this VC arm will depend on the success of its early-stage optical and power-efficiency proof-of-concepts. After a blockbuster debut in July raising $26.5 billion, the United States now generates 65% of SK hynix’s total revenue.
The company’s upcoming $3.87 billion Purdue Research Park advanced packaging plant in Indiana might be just the first of other potential plants on U.S. soil, showing it’s attempting to become part of the “local” U.S. supply chain. That would help insulate its clients – Nvidia, Microsoft, Dell, HP, Apple, and others – from global logistics shocks.
Susana Schwartz
Technology Editor
RCRTech
AI Infrastructure Top Stories
1st NPO computing cluster: Huawei’s Near-Packaged Optics -based computing cluster supports AI models approaching 10 trillion parameters, delivering 8 EFLOPS of FP8 compute and 16 EFLOPS at FP4 precision alongside 1 PB of HBM capacity.
So. Korea grows occupier demand: Cushman & Wakefield’s 2026 Asia Pacific Data Center Investment Landscape forecasts $4.1 billion in annual colocation rental revenue by 2030, with 255MW of pre-leased capacity as of the first half of 2026.
AI Today: What You Need to Know
ECOC news – Marvell optical interconnect: Marvell’s 2nm optical interconnect demos at the European Conference on Optical Communication feature optical PAM4, coherent-lite PAM4/IMDD and ZR/ZR+ coherent solutions in the 2nm node.
ECOC news-data-tapping prevention: CUbIQ Technologies and Coherent partnered in a POC to address data tapping of fiber-optic lines running between separate AI data center campuses using CV-QKD transceiver in a QSFP-28 pluggable form factor.
Microsoft India AI hub: At the Microsoft AI Infrastructure Summit, Microsoft designated its Hyderabad region as a strategic hub for Asia and the Global South, confirming its Indian regions now boast high-efficiency cooling and hardware.
Alpha Compute in No. PA: Alpha Compute moved to acquire land and natural gas rights for a 200 MW campus in PA, amending the binding term sheet for its planned natural gas-powered AI data center, and securing $47 million in seller financing.
Crusoe shifts to modular “Spark” units: Crusoe raised US$3.9B in a Series F round, which will help it pivot toward factory-built, mobile “Spark” modules that can be shipped to areas with power access to handle AI inference workloads.
New York tightens DC restrictions: Gov. Kathy Hochul announced a regulatory oversight framework for frontier AI safety, which through the Responsible AI Safety and Education (RAISE) Act, requires reporting of incidents within 72 hours.
RCR Events
Industry Resources
Report: NTN in motion — evolving standards, expanding services
