🐯虎嗅•Stalecollected in 14m
Jensen Huang's Beijing visit and chip industry signals
💡Jensen Huang's movements often signal major shifts in the global AI chip supply chain and market access.
⚡ 30-Second TL;DR
What Changed
Huang's Beijing visit follows high-level diplomatic engagements, signaling potential shifts in China-US semiconductor relations.
Why It Matters
The visit suggests a potential thawing or recalibration of Nvidia's engagement with the Chinese market, which is critical for global AI hardware distribution.
What To Do Next
Monitor Nvidia's official supply chain announcements regarding China-specific product variants to adjust your hardware procurement strategy.
Who should care:Founders & Product Leaders
Key Points
- •Huang's Beijing visit follows high-level diplomatic engagements, signaling potential shifts in China-US semiconductor relations.
- •Historical correlation between Huang's public 'food tours' and major Nvidia milestones like Blackwell production.
- •Nvidia's strategic reliance on global supply chains, including HBM from Samsung and foundry services from TSMC.
- •Huang's previous public support for the importance of the Chinese market despite export restrictions.
🧠 Deep Insight
Web-grounded analysis with 27 cited sources.
🔑 Enhanced Key Takeaways
- •The US Commerce Department has approved the sale of Nvidia's H200 AI accelerators to approximately ten Chinese firms, including Alibaba, Tencent, ByteDance, and JD.com, with each approved customer allowed to purchase up to 75,000 chips.
- •Despite US approval for H200 sales, no deliveries have been made to Chinese companies due to Beijing's guidance encouraging domestic firms to prioritize homegrown AI semiconductor development, leading to a stalled major technology deal.
- •Nvidia's market share in China's advanced AI chip market has significantly declined, with CEO Jensen Huang stating the company is "100% out of China" and its share falling from approximately 95% before export restrictions to effectively zero, while Huawei is projected to capture a 60% share of the Chinese AI market by the end of 2026.
- •The US policy on AI chip exports to China has shifted from blanket restrictions to more selective controls, allowing H200 exports with conditions like a cap on shipments (not exceeding 50% of US customer volume) and a revenue-sharing arrangement where Nvidia remits 25% of sales to the US government.
- •Nvidia has strategically halted H200 chip production specifically for China, reallocating that capacity to its next-generation Vera Rubin platform, while also booking entire server plant capacities through 2026 for Blackwell and Rubin AI servers to meet overwhelming demand from Western hyperscalers.
📊 Competitor Analysis▸ Show
| Company | Key AI Chips/Platforms | China Market Share (approx.) | Performance/Notes (vs. Nvidia) |
|---|---|---|---|
| Nvidia | Blackwell (B200/GB200), Hopper (H100/H200), Rubin (upcoming) | ~0% (after restrictions, previously ~95%) | Global leader, CUDA ecosystem, H200 approved for China but stalled. |
| Huawei | Ascend 910C, Ascend 950, 950DT | Projected 60% by end of 2026 | Ascend 910C achieves ~60% of H100 inference performance; Chinese cloud providers are securing Ascend chips. |
| AMD | Instinct MI455X, Helios rack-scale platform, MI325X | N/A (global ~7% AI accelerator market share Q3 2025) | Offers 20-30% lower cost alternative on comparable workloads to Nvidia; MI325X also subject to US export rules. |
| Alibaba (T-Head) | Zhenwu 810E inference accelerator | N/A | Positioned to compete with Nvidia's H20 in price-performance for inference tasks; deep co-design with Alibaba Cloud. |
| Cambricon | Siyuan 590, Siyuan 690 (upcoming) | N/A | Siyuan 590 competitive with Nvidia A100; Siyuan 690 targeting H100-class performance expected H2 2026. |
🛠️ Technical Deep Dive
- Nvidia's Blackwell-architecture GPUs pack 208 billion transistors and are manufactured using a custom-built TSMC 4NP process.
- The architecture features two reticle-limited dies connected by a 10 terabytes per second (TB/s) chip-to-chip interconnect, forming a unified single GPU.
- It incorporates a second-generation Transformer Engine with custom Tensor Core technology, supporting new microscaling formats like MXFP4 and MXFP6, enabling high-accuracy 4-bit floating point (FP4) AI for generative AI inference and training.
- The NVIDIA GB200 Grace Blackwell Superchip connects two high-performance Blackwell GPUs and an NVIDIA Grace CPU using the NVLink-C2C interconnect.
- The NVIDIA GB200 NVL72 system connects 36 Grace CPUs and 72 Blackwell GPUs in a liquid-cooled, rack-scale design, capable of delivering 30X faster real-time trillion-parameter large language model (LLM) inference.
- Blackwell GPUs feature 186GB HBM3e memory with 8 TB/s bandwidth and a total NVLink bandwidth of 130TB/s.
- The architecture supports PCIe 6.0 for datacenter products and GDDR7 memory for consumer products.
- It includes NVIDIA Confidential Computing for hardware-based security and a new decompression engine combined with Spark RAPIDS™ libraries for enhanced data analytics performance.
🔮 Future ImplicationsAI analysis grounded in cited sources
Nvidia will likely continue to pursue a dual strategy of complying with US export controls while attempting to maintain a presence in the Chinese market through compliant or custom chips.
Despite current stalls, the potential $50 billion annual Chinese AI market is too significant for Nvidia to abandon entirely, necessitating adaptation to evolving regulations.
China's push for AI chip self-sufficiency will accelerate, leading to increased market share for domestic players like Huawei and a more fragmented AI chip ecosystem in China.
US export restrictions have spurred Chinese tech giants to expand testing and adoption of domestic alternatives, reducing their dependence on foreign suppliers.
Geopolitical tensions and trade negotiations between the US and China will continue to be a primary determinant of Nvidia's market access and revenue potential in China.
Recent approvals and subsequent stalls in H200 shipments demonstrate the direct impact of high-level diplomatic engagements and policy shifts on semiconductor trade.
⏳ Timeline
2022-10
US Biden Administration imposed sweeping restrictions on AI chip exports to China, impacting Nvidia's market.
2024-03
Nvidia officially announced the Blackwell architecture at GTC 2024.
2025-04
New US restrictions on Nvidia's H20 chip led to a $4.5 billion charge for excess inventory.
2025-12
US President Trump announced a policy shift allowing Nvidia to sell H200 chips to China.
2026-01
The US Commerce Department formalized regulations permitting H200 sales to China, with conditions.
2026-05
Jensen Huang joined President Trump's delegation to Beijing for a summit with Xi Jinping, amidst reports of stalled H200 deliveries to China despite US approval.
📎 Sources (27)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
📰
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: 虎嗅 ↗