Nvidia H20 export debate: Aaron Ginn argues sales to China increase US soft power, not military risk

Summary

Aaron Ginn makes the case that selling Nvidia H20 chips to China increases American soft power rather than military risk — and that the export ban crowd is, at best, overestimating what the US can actually control.

The core argument is straightforward: Nvidia's architecture is not just hardware. It is a software platform, and Jensen Huang is pushing it further in that direction with Grace and Rubin. Every Chinese company that builds on H20s becomes more dependent on Nvidia's stack, which Ginn argues makes conflict less likely, not more. The analogy he reaches for is Apple — if you want Chinese companies running on your platform, you have to sell to them first.

The soft power vs. containment divide

Ginn's sharpest point is that the Beijing-vs.-Alibaba dynamic is being flattened. Chinese tech companies — Alibaba, ByteDance — want Nvidia chips. The Financial Times reported that Beijing is already pressing them to justify orders for US-made semiconductors, which Ginn reads as evidence of internal tension the US could exploit rather than resolve. Selling chips, he argues, widens that gap. Banning them closes it.

On the military-capabilities risk, Ginn is less dismissive but still skeptical. He acknowledges there is a real difference between possessing model weights and being able to inference them a trillion times a day across a population. But his counter is that China is going to develop regardless. Over half of OpenAI's superintelligence team is Chinese nationals; add people of Chinese descent and it reaches roughly 75%. Math and engineering are universal, and the knowledge cannot be contained the way an F-35 can.

Where the IP risk actually sits

Ginn draws a clear line on IP leakage. The risk worth worrying about is foundry equipment — ASML, EDA software, the mirrors inside lithography machines — not Nvidia chips. The Trump administration's approach, he argues, reflects this distinction: targeting the ability to make chips rather than restricting the ability to buy them, which inverts the Biden-era logic. Nvidia has an R&D center in Shanghai, but it does not manufacture anything, so the chip-level IP exposure is low.

The Huawei substitution problem

A ban does not leave China GPU-poor for long — it accelerates Huawei Ascend adoption. Ginn's argument is that a China running on H20 clusters stays dependent on Nvidia's ecosystem; a China that is forced onto Huawei builds indigenous capability at scale. Chinese customers want Nvidia, and pushing them toward Huawei out of principle hands Huawei the volume it needs to close the gap.

On the tariff revenue angle

Dylan Patel of SemiAnalysis has argued that the 15% tariff on H20 exports could compress Nvidia's margins on that chip to AMD-like levels. Ginn treats this as marginal — the H20 was already less profitable than the H100, and China is a small share of Nvidia's overall business at this point given the write-down already taken. For smaller chip companies — Etched, Groq, Cerebras — the export licensing requirement could be a proportionally bigger deal.

The bottom line of Ginn's position: containment creates the conditions for conflict by isolating China and accelerating domestic chip development, while dependency creates leverage. The US is best positioned as the dominant platform exporter, not as a gatekeeper whose restrictions China is already routing around through Malaysia and other diversion channels.