NVIDIA H200 Returns to China, Huawei's Tech War View, and Google TPU Bottlenecks

NVIDIA H200's Conditional Return

• 💡 The US has conditionally allowed NVIDIA's H200 AI chip to re-enter the Chinese market, potentially restoring NVIDIA's revenue from the region.
• 🎯 Although the H200 is a previous-generation chip, it represents the highest-spec AI chip permitted after years of US restrictions.
• 📌 Chinese chip manufacturers are expected to purchase H200 for official hardware and software technical support, despite China's ongoing push for chip self-sufficiency.
• 📈 Local Chinese AI chip companies are not directly competing with NVIDIA but are focusing on cost-effective markets like inference, with diversified supply becoming a key trend.

Huawei's Perspective on US-China Tech

• 💬 Huawei founder Ren Zhengfei discussed the US-China tech competition, framing it not as a zero-sum game but as a path for China to innovate under pressure.
• 🔑 He stated that Huawei's self-reliance was a forced choice, not a proactive one, due to US sanctions.
• ✅ Ren Zhengfei acknowledged the benefits of US technological civilization and emphasized that most Chinese companies still utilize US technology, which aids China's overall industry.

Ren Zhengfei's Vision for Tech Cooperation

• 🧠 He highlighted that global industry progress relies on technology sharing and commercialization, advocating for an open approach.
• 🚀 Ren Zhengfei suggested the US focuses on philosophical, future-oriented visions, while China excels in practical AI applications like smart mines and healthcare.
• 💡 He believes the US attracting global talent is positive and that China must foster a more open and integrated talent environment to thrive globally.

Google TPU Production Challenges

• 📊 Google's TPU demand is strong, projected to last until 2027, indicating significant market interest in its ASIC solutions.
• ⚠️ Initial concerns about CoWoS and HBM shortages for TPUs are being re-evaluated, with traditional packaging firms capable of supporting basic CoWoS-S production.

Key Supply Chain Bottlenecks

• 🛠️ The primary bottleneck for Google TPU production is the TSMC 3nm capacity, with discussions about converting 5nm facilities to meet demand.
• 🔬 Another critical shortage is T-Glass fiberglass, a raw material essential for high-end substrates and PCB boards used in AI servers.
• 📈 DDR5 DRAM is experiencing significant price increases and profitability, making it competitive with HBM, which has seen some price adjustments from manufacturers like Samsung.