China's EUV Manhattan Project and Export Control Mythbusting with Chris McGuire

US-China Tech Competition & Export Controls
📌 Semiconductor equipment export controls are the core lever to constrain China's technological progress in AI and other advanced technologies.
⚙️ China is significantly lagging in producing advanced chips because they cannot acquire the necessary sophisticated equipment, such as Extreme Ultraviolet (EUV) machines.
📰 Reports of China assembling an EUV prototype using salvaged components are more a story of export control evasion than true indigenous innovation, as it lacks the ability to fully develop the complex domestic supply chain.
📉 Huawei’s next chip iteration (Ascend 950) shows lower processing power and memory bandwidth than its current best chip, indicating ongoing technical struggles despite government funding.

Huawei's AI Compute Capacity
📊 Even using Huawei's most optimistic projections for chip quantity, their aggregate AI computing power next year is estimated to be only 4% of what Nvidia is making.
📉 The quality gap between US and Chinese chips is widening substantially, projected to go from a 5x gap now to a 17x gap in two years, even before accounting for technical advantages like precision levels (e.g., FP8 vs. FP16).
🛑 The massive boost in Chinese AI compute capacity in 2022-2024 was largely due to the TSMC violation, which likely delayed the full impact of export controls by two years.

Policy Effectiveness and Gaps
❌ The claim that export controls accelerate indigenization is considered fiction by one expert, as China's dedication to self-sufficiency started in 2014-2015, long before comprehensive controls were implemented.
🚫 Chip smuggling is "embarrassingly fiction" to deny; the DOJ busted a ring the same day H200 sales to China were announced, and Malaysia's status as the #2 AI chip importer suggests diversion.
☁️ The biggest missing policy piece is regulating access to rented compute (cloud services), allowing China nearly unlimited remote access to US-based cloud capacity without physical chip ownership restrictions.

Key Points & Insights
➡️ The biggest mistake policymakers could make is reverting to a sliding scale approach for export controls; an absolute line in the sand is necessary because large quantities of second-best chips can aggregate to achieve similar effects as advanced ones.
➡️ Selling advanced chips to China in the short term may destroy Nvidia's long-term addressable market by helping Chinese cloud hyperscalers (like Alibaba, Tencent) grow large enough to compete globally, unlike US companies that remain reliant on US suppliers.
➡️ US export control policy must focus on controlling the make, buy, and rent components of compute; specifically, closing loopholes on compute smuggling/remote access is critical to maximizing the strategic advantage.

📸 Video summarized with SummaryTube.com on Dec 28, 2025, 19:01 UTC