Brian Reed: Hey. Good to be back.
Eliza Ward: Yeah, likewise. You see the Huang statement this week?
Brian Reed: The 'failure' thing. Yeah. Jensen Huang, NVIDIA's CEO, publicly calling U.S. export controls a failure — while NVIDIA is simultaneously lobbying to relax them. That collision is what's strange to me.
Eliza Ward: And the timing. April 2025 — the U.S. bans all AI GPU exports to China. NVIDIA takes a projected $5.5 billion revenue hit. Stock drops nearly seven percent.
Brian Reed: Seven percent in — what, a day?
Eliza Ward: Pretty much. And then three months later, July 2025, Washington reverses course — reinstates licenses for the H20. Which NVIDIA had specifically engineered to stay just under the export-control thresholds. It's almost like a car manufacturer building a vehicle right at the emissions limit every time the limit tightens.
Brian Reed: But here's what the reinstatement actually revealed — because the July reversal is when Beijing made its own move. The Cyberspace Administration issued informal guidance, ByteDance and Alibaba specifically named, telling them to stop testing, stop purchasing NVIDIA chips. Washington opens the door, Beijing blocks it from the other side.
Eliza Ward: And separately — state-funded data centers, newly built ones, directed to use only domestically produced chips. That's not a tariff, that's not a sanction. That's structural exclusion.
Brian Reed: So when Huang says controls are the failure — I mean, which controls? Because the constraint that's actually shrinking his market isn't coming from BIS.
Eliza Ward: Right, and the market-share number makes this concrete. NVIDIA reportedly goes from around 95% to roughly 55% in China by 2025. That sounds catastrophic. But wait — AMD is at 4%. Four percent. So where did the other forty points go? It went to Huawei's Ascend line, which is now at roughly 20% of China's domestic AI chip market as of early 2026. American competitors didn't gain. A Chinese company did. That's — that's actually a different story than the headline.
Brian Reed: Although — and I want to flag this — the sourcing on that 55% figure is thin. We should hedge that.
Eliza Ward: No, fair — the directional story holds, but yeah, I wouldn't hang a percentage point on it. What I'm more confident about is the BIS January 2026 move: H200 licenses, AMD MI325X, shifted to case-by-case review, aggregate China and Macau shipments capped at 50% of U.S.-destined volume, customer-verification requirements on top. Washington kept tightening even while Huang was making his 'failure' argument publicly.
Brian Reed: And the part we haven't gotten to yet is going to make the 'principled critic' framing fall apart pretty fast — because Huang's public argument looks very different next to what he's declined to answer.
Eliza Ward: Right — and that's the thing that makes the 'principled critic' frame hard to hold. Warren and Banks — bipartisan, that matters — sent a letter to Huang directly. About a reported Shanghai R&D facility. The specific concern was military-civil fusion: that an active NVIDIA presence there could expose U.S. chip designs to China's military development pipeline.
Brian Reed: And NVIDIA has not confirmed or denied the facility's current status. That's — I mean, that's not a minor gap. That's the central operational question sitting unanswered.
Eliza Ward: Unconfirmed, yeah. We should say that plainly — the Shanghai facility allegation is unresolved in the sources we have.
Brian Reed: But here's what is confirmed: Huang declined the congressional invitation to testify publicly. Warren's exact quote was 'the American people deserve answers in a public forum.' And separately she raised whether NVIDIA's board is actually exercising oversight on export-control compliance — directly challenging Huang's claim that there's 'no evidence of any AI chip diversion.'
Eliza Ward: Wait — the diversion claim specifically. Huang says no evidence. Warren says allegations exist that contradict that. Neither is adjudicated. Those are two stated positions, not a resolved question.
Brian Reed: Right. And that's actually — that's the weak point in the 'principled critic' framing. It's not that Huang is necessarily wrong about policy. It's that he's making a public compliance argument while simultaneously refusing the public forum where the compliance questions would get tested. Those two things don't sit comfortably together.
Eliza Ward: And he's doing that while personally lobbying Trump administration officials against the controls. The lobbying and the 'failure' critique are running at the same time. That's not principled critic territory — or at least, the evidence doesn't get us there cleanly.
Brian Reed: And there's actually a paper — May 2026, Liu and Lee on arXiv — that makes Huang's counterproductive-controls argument on the merits. Independently. Like, WTO obligations under GATT Article XXI(b), strategically counterproductive, the whole thing. Policy kept tightening anyway. So he might be right about the policy and still be impossible to separate from the lobbying position. Those aren't mutually exclusive.
Eliza Ward: That's — yeah, that's the part that doesn't resolve. Even if Liu and Lee are correct, Huang still hasn't answered the accountability question. The Shanghai facility: planned, paused, abandoned — publicly unresolved. The diversion question: his assurance on one side, congressional suspicion on the other, nothing adjudicated. The merits of the policy argument don't touch either of those.
Brian Reed: So what would actually settle it? I mean — the Shanghai facility status, that's confirmable. NVIDIA knows whether it exists. And the diversion question either gets tested in a public forum or it doesn't. Huang declining to testify is the thing that keeps both of those open.
Eliza Ward: Until those two things get answered publicly — facility status, diversion evidence — the 'failure' argument and the lobbying strategy are just... sitting on top of each other. Indistinguishable.
