Sinica Podcast Summary

Episode: Paul Triolo on Nvidia H200s, Chinese EUV Breakthroughs, and the Collapse of the Sullivan Doctrine
Date: December 26, 2025
Host: Kaiser Kuo
Guest: Paul Triolo (SVP, China & Tech Policy Lead at DJA Albright Stonebridge Group; Non-resident Senior Fellow, Asia Society Policy Institute)

Episode Overview

This episode sees host Kaiser Kuo in conversation with Paul Triolo, a renowned expert on U.S.–China technology policy. They explore the recent U.S. decision to allow Nvidia H200 GPU chip sales to select Chinese customers, the reported Chinese breakthroughs in EUV lithography, and what these mean for U.S. export controls, industrial policy, and the so-called “Sullivan Doctrine” governing technology competition with China.

Key Themes & Discussion Points

1. The Shocking Green-light for Nvidia H200s

• [00:40] Trump’s announcement—post-Busan summit—unexpectedly permits Nvidia H200 sales to “vetted” Chinese buyers, imposing a 25% tax.
• Reactions: Bipartisan confusion and disagreement in the U.S.; Democratic China hawks oppose, while some Republicans defend the move.
• Underlying question: Does this signal the collapse of tech export controls or merely a tactical exception?

2. What Does the H200 Decision Actually Change?

• [06:44] Paul Triolo: “In the short term, [the decision] changes the ability of some number of Chinese AI developers to continue to use U.S. technology to train their models and run inference... What it doesn’t do is give China... some strategic advantage in AI writ large.” (06:44)
  • H200s are “not the most bleeding-edge technology”—18 months behind Nvidia’s Blackwell.
  • These GPUs will be valuable for training and inference in China for years, but don’t confer a decisive leap. (08:07)
  • The hyperbole about China gaining instant AI dominance is misplaced.

3. A Timeline of U.S. Chip Controls

• Background Context:
  • Starts with Trump’s entity list targeting ZTE and Huawei (2018+), originally due to non-AI, non-5G issues.
  • Biden administration: Shifts scope to AI, formalizes the “Sullivan Doctrine” (2022), which asserts the U.S. must retain an “absolute lead” in advanced compute, biotech, and green tech.
  • Export controls become cat-and-mouse: Companies like Nvidia repeatedly redesign to skirt thresholds, U.S. moves the goalposts, culminating in restrictions like those on the H20 GPU.
  • Trump’s move “essentially shot down” the degrade-and-hold China-in-place strategy. (11:32–13:50)

4. Inside the White House Decision-Making

• Influence of Industry:
  • David Sacks, now U.S. AI Czar, and Nvidia CEO Jensen Huang advocated maintaining U.S. firm competitiveness in China, arguing letting only Chinese companies (e.g., Huawei) serve the local market would “cede the AI stack.”
  • Financial context: Nvidia faced a $5bn inventory write-off when H20s were effectively banned.
  • Quote: “Jensen uses figures like the Chinese have more than half of the world’s AI developers... that was a market that [he] contended the U.S. shouldn’t concede.” (16:30)
  • Marco Rubio and others reportedly talked Trump out of relaxing Blackwell exports en route to Busan.

5. “Choke Point” Proponents vs. Market Advocates

• Arguments For Stricter Controls:
  • U.S. must “arrive at AGI/ASI first,” because whoever gets there first possesses strategic advantage (20:00–25:00).
  • Controls are meant to “buy time” for U.S. and allies to lead AI development and mitigate risks.
• Triolo’s Critique:
  • Key dates (“AGI by 2027… now 2028… maybe later”) are speculative.
  • Assumptions about immediate governmental use of AGI/ASI for offense are unverified.
  • China’s approach favors “diffusion” of AI (pragmatic, enterprise-led spread), not the AGI arms race imagined in D.C.

6. Technical Realities: What Can H200s Do?

• H200s significantly boost Chinese capabilities for model training (esp. over aging A100s), mainly via higher bandwidth.
• However, smuggling/diversion has only a minor impact; important loopholes remain where Chinese firms can use overseas data centers.
• Demand for H200s is real among China’s tech giants (Alibaba, Tencent, ByteDance), but limited by bottlenecks in supply (packaging at TSMC) and government policy.
• Quote: “The H200s will provide this sort of bridge to a potentially more viable domestic alternative in the future.” (37:47)

7. China’s AI Hardware Ecosystem

• Chinese firms balance between Nvidia and emerging domestic suppliers (Huawei, MetaX, Inflame).
• Huawei’s Ascend chips—not yet a 1-for-1 Nvidia replacement—use different architecture (NPUs vs. GPGPUs) and require new software stacks (e.g., Huawei’s CANN).
• Government stimulus for “indigenous innovation” exists, but practical needs mean Chinese firms keep using Nvidia gear when possible.

8. Does Foreign Tech Fuel Domestic Leapfrogging?

• Triolo is skeptical that China’s past approach to solar panels or EVs applies here.
  • “Semiconductors and AI are very different than solar panels... left to their own devices, Chinese companies would continue to buy the most advanced tools.”
  • Forced substitution (as with Huawei building HarmonyOS) is costly and inefficient, only prompted under duress.

9. EUV Lithography “Breakthroughs”

• [50:31] Reuters reports of Chinese advances in EUV lithography (crucial for <7nm chips): Triolo is cautious.
  • Many ASML EUV subsystems (Zeiss optics, advanced photoresists, computational lithography software) are extremely complex; a full indigenization is a massive challenge.
  • He suspects “cannibalized parts” are taken from DUV systems rather than EUV (few exist).
• Big Picture: U.S. choke-points light a fire under Beijing; “flywheel” of forced domestic innovation and feedback loops between toolmakers and fabs means China is catching up faster than anticipated (54:39–58:22).
  • “Choke-point technologies” can’t be locked in a box—China will find alternate routes.

10. Export Control Contradictions and the Future

• [46:06] U.S. now exports 5nm-class GPUs (H200s) with advanced memory to China, but still blocks semiconductor tools at far older thresholds (e.g., 14/16nm). Industry questions the logic—will tool bans be updated to match reality, or is the policy simply incoherent?
• Lack of a policy “center of gravity” post-Biden: No interagency lead, ad hoc influence (David Sacks), more lobbying, and possible further disorder ahead.
• Prediction: Early 2026 may bring attempts to rationalize export controls and clarify rules around new architectures. For now, “the doctrine is quietly collapsing... or at least visibly strained.” (61:41)

11. Special Focus: Taiwan’s Role

• Taiwan (TSMC) remains pivotal. Many Chinese companies still fab advanced silicon there—as long as end uses are “consumer focused.”
• Full technological autarky in China might reduce Taiwan’s strategic value, but in practice, China-Taiwan chip ties are sticky and likely to persist for years. (58:43–61:00)

12. What Would Prove the Critics Right (or Wrong)?

• [66:58] For critics of the H200 decision to be vindicated, Chinese firms would have to “develop more advanced models than OpenAI, Anthropic, Google, and Meta” as a result of access to these chips within 2–3 years—a scenario Triolo finds implausible.
• U.S. lead in data center investment, power infrastructure (subject to U.S. constraints), and open-source model development persists.
• Chinese innovation is impressive, but “H200 exports do not mean the U.S. loses its edge.”

Notable Quotes

• Paul Triolo [06:44]:
  “What it changes is... the ability of some number of Chinese AI developers to continue to use U.S. technology to train their models... What it doesn’t do is give China... some strategic advantage in AI writ large.”
• Triolo [16:30]:
  “Jensen [Huang] uses figures like the Chinese have more than half of the world’s AI developers... his argument is that we should allow sales... so that US AI companies should be allowed to compete in China so Huawei... didn’t come to dominate the AI stack.”
• Triolo [25:36]:
  “There’s this idea out there... the U.S. is AGI-pilled... and China is taking a more pragmatic approach.”
• Triolo [54:39]:
  “There’s like... ten different really advanced technologies that all have to work together perfectly [for EUV].”
• Triolo [55:04]:
  “What this [EUV news] has done is forced companies like Huawei... to figure out how to do a lot of stuff they never would have done.”
• Kaiser Kuo [70:24]:
  “Because the H200’s export does not mean that the US loses anything, any capability. It’s just giving China a little bit a step incremental ability to train better models.”

Timestamps for Key Segments

• US Export Controls Timeline: 09:23–13:50
• Policy Logic—Pro/Con on Export Controls: 19:06–25:36
• Technical Significance of H200: 27:12–33:08
• Chinese Demand and Hardware Stack: 34:44–40:15
• Evolution of Domestic Chinese Hardware/Software (Huawei): 37:06–38:48
• EUV Development and its Limits: 51:02–58:22
• Future of U.S. Export Control Policy: 61:41–66:58
• What Would Prove the Detractors Right?: 66:58–70:56

Book & Substack Recommendations

• Po Zhao’s Substack: “Hello China Tech” (hello-chinatech.com)
• Dean Ball’s Substack: “Hyperdimensional”
• Book: “The Life of Brzezinski: America's Grand Strategist” by Edward Luce
• Book: “Everything is Tuberculosis” by John Green
• Book: “The Anthropocene Reviewed” by John Green

Tone & Style

• Engaged, analytical, and measured.
• Both Triolo and Kuo combine wonky policy discussion with deep technical knowledge and a dash of wit (“December 8 may be a day that will live in infamy,” [06:01]).

Memorable Moments

• Triolo’s takedown of the “over-the-top” DOJ language about chip “building blocks for AI superiority”—called it “just wrong. Every sentence... is not accurate.” [31:43]
• The analogy comparing export controls to thinking you can “lock technology in a box” [55:04]—but “they’re gonna figure it out.”
• The persistent dichotomy: “Is the ‘Sullivan Doctrine’ quietly collapsing or being fiercely defended... even as its logic is strained?” [61:41]
• Kaiser: “Can you imagine having policy in the Biden era determined by a tweet?” [66:56]

Summary Takeaway

The U.S. decision to allow Nvidia H200 sales to China marks an inflection point in the semiconductor Cold War. While not ceding the technological high-ground, it signals both the practical limits of “choke-point” control strategies and the disorder introduced by a new, more industry-driven White House. Chinese innovation is accelerating, and supply chains are evolving fast. The future of export controls remains uncertain—caught between hawkish caution and industry realism as both sides race for AI leadership.