Share with your CTO
Hana Asset Management is betting that the AI chip supply chain is investable as a discrete asset class, launching the 1Q Nvidia-Fixed U.S. AI Semiconductor ETF on the Korea Exchange with a fixed 25% Nvidia weighting at each rebalancing. The fund holds just 10 stocks spanning GPUs, CPUs, memory, and networking, deliberately tighter than the 30-stock Philadelphia Semiconductor Index. Hana’s rationale is explicitly agentic AI-driven: as inference workloads (computation during live service delivery, not just model training) scale, demand should broaden from Nvidia GPUs to Arm, Micron, Broadcom, and Marvell.
What this means for your business
The story isn’t the ETF. It’s the thesis embedded in the ETF’s construction. A Korean asset manager designing a product around inference-era demand signals that the “GPU monoculture” framing of AI infrastructure is losing credibility in capital markets, not just in architecture discussions. If you’re a CTO still treating your chip procurement strategy as Nvidia-first and everything-else-optional, you’re increasingly out of step with where sophisticated money thinks the load is shifting.
Hana’s 10-stock basket is essentially a public declaration about which components matter in the inference era. CPUs from Arm and Qualcomm, networking silicon from Broadcom and Marvell, memory from Micron, all sitting alongside Nvidia in a fund explicitly designed around agentic AI workloads. That’s not a diversification hedge. That’s a structural claim about what inference at scale actually consumes. The recurring failure mode in enterprise AI infrastructure planning is overbuilding GPU capacity for training while underprovisioning the networking and memory layers that throttle inference throughput. This ETF’s composition is a direct counter to that pattern.
The falsification condition here is simple: if Nvidia’s actual revenue share of AI data center spend holds above 70% through 2027, the inference-broadening thesis is premature and the ETF’s deliberate de-concentration will look like a bet made too early. Watch Broadcom’s and Marvell’s custom ASIC (application-specific chip) pipeline growth over the next two earnings cycles. If hyperscaler custom silicon is genuinely displacing general-purpose GPU spend, those numbers will move before any ETF rebalancing tells you anything useful.
Based on reporting from Hana Asset Management Launches U.S. AI Chip ETF With 25% Nvidia Weighting, originally published 2026-10-18 20:54:00.
