AMD CEO: China Remains a Critical Market

Understand the strategic reliance of major AI chip providers on the Chinese market amidst geopolitical tensions.
30-Second TL;DR
What Changed
China contributes roughly 20% of AMD's total revenue.
Why It Matters
This highlights the dependency of major US chipmakers on the Chinese market despite ongoing export restrictions. It suggests that AMD will continue to balance compliance with expansion efforts in the region.
What To Do Next
Monitor AMD's quarterly earnings reports for shifts in regional revenue distribution to assess the impact of potential future trade policy changes.
Key Points
- •China contributes roughly 20% of AMD's total revenue.
- •Lisa Su reaffirms the strategic importance of the Chinese market.
- •AMD continues to navigate global trade complexities while maintaining its presence in China.
Deep Insight
Background and context from public sources — not the original article. 9 sources cited.
Enhanced Key Takeaways
- •AMD's revenue from China has seen a decline, dropping from approximately 24% of total revenue in 2024 (equating to about $6.2 billion) to the current 20%, reflecting the tangible impact of evolving US export restrictions.
- •Starting January 2026, the US implemented new licensing reviews for select advanced AMD chips, such as the MI325X, requiring individual approval for sales to Chinese buyers, while sales of the MI308 now include conditions like revenue sharing with the US government.
- •AMD has strategically developed modified, less powerful versions of its advanced chips, like the MI308 accelerator, specifically for the Chinese market to navigate export controls and maintain market access without requiring special approval.
- •A significant portion of AMD's current revenue from China is derived from less restricted segments, including PC processors and gaming chips, which face fewer stringent export regulations compared to its advanced AI products.
- •AMD is actively pursuing a goal of capturing 25% of the discrete GPU market in China, a strategy supported by an alleged internal management reshuffle that has integrated its GPU businesses under the control of the company's CPU sales team in the region.
Competitor Analysis
- Key Products/Focus in China
- CPUs (Ryzen, EPYC), GPUs (Radeon, Instinct MI308/MI325X)
- Market Share (China AI Accelerators)
- ~4% (units, AI accelerators)
- Technical Differentiators/Challenges
- Offers x86 CPUs compatible with existing data centers; MI350X matches NVIDIA B200 on FP8 compute and exceeds memory (288GB HBM3E vs 192GB HBM3E); open-source ROCm software stack; faces US export restrictions on advanced AI chips.
- Key Products/Focus in China
- GPUs (H100, H200, B200, H20)
- Market Share (China AI Accelerators)
- ~55% (units, AI accelerators), ~80% (revenue, global AI accelerators)
- Technical Differentiators/Challenges
- Dominant in AI accelerators with established CUDA software ecosystem; strong performance and market adoption; faces US export restrictions, leading to China-specific downgraded chips (e.g., H20).
- Key Products/Focus in China
- CPUs (Xeon), AI Accelerators
- Market Share (China AI Accelerators)
- Not specified for AI accelerators, but strong in server/PC CPUs
- Technical Differentiators/Challenges
- Advancements in 18A semiconductor technology; competitive in server and PC sectors; also subject to US trade policies.
- Key Products/Focus in China
- Ascend AI Accelerators
- Market Share (China AI Accelerators)
- ~41% (units, AI accelerators)
- Technical Differentiators/Challenges
- Domestic Chinese competitor rapidly closing performance gap with homegrown solutions; not subject to US export controls, benefiting from national push for self-sufficiency.
| Competitor | Key Products/Focus in China | Market Share (China AI Accelerators) | Technical Differentiators/Challenges |
|---|---|---|---|
| AMD | CPUs (Ryzen, EPYC), GPUs (Radeon, Instinct MI308/MI325X) | ~4% (units, AI accelerators) | Offers x86 CPUs compatible with existing data centers; MI350X matches NVIDIA B200 on FP8 compute and exceeds memory (288GB HBM3E vs 192GB HBM3E); open-source ROCm software stack; faces US export restrictions on advanced AI chips. |
| NVIDIA | GPUs (H100, H200, B200, H20) | ~55% (units, AI accelerators), ~80% (revenue, global AI accelerators) | Dominant in AI accelerators with established CUDA software ecosystem; strong performance and market adoption; faces US export restrictions, leading to China-specific downgraded chips (e.g., H20). |
| Intel | CPUs (Xeon), AI Accelerators | Not specified for AI accelerators, but strong in server/PC CPUs | Advancements in 18A semiconductor technology; competitive in server and PC sectors; also subject to US trade policies. |
| Huawei | Ascend AI Accelerators | ~41% (units, AI accelerators) | Domestic Chinese competitor rapidly closing performance gap with homegrown solutions; not subject to US export controls, benefiting from national push for self-sufficiency. |
Technical Deep Dive
- US export controls specifically target advanced AI chips, including AMD's MI325X, which now requires individual licensing for sales to China, and the MI308, which is subject to conditional sales with revenue sharing.
- AMD has developed the MI308 as a modified, less powerful AI accelerator for the Chinese market to comply with these export restrictions.
- The MI308 is a pure GPU accelerator and a downgraded version of the more advanced MI300X; the integrated CPU+GPU product, MI300A, currently does not have a China-compliant version.
- AMD's MI350X, a high-end AI chip, offers 4,600 TFLOPS of FP8 compute and 288GB of HBM3E memory, which surpasses the memory capacity of NVIDIA's H100 (192GB HBM3).
- AMD leverages its open-source ROCm software stack, which contrasts with NVIDIA's proprietary CUDA, aiming to provide an alternative ecosystem for developers.
- AMD's integrated strategy includes its x86 CPUs, AI GPUs, and the MI300A (which combines CPU and GPU on a single chip, sharing HBM memory to reduce data transfer latency), offering a full-stack solution that is particularly advantageous in Chinese data centers with existing x86 software infrastructure.
Future ImplicationsAI analysis grounded in cited sources
Timeline
- 2005-10AMD establishes Greater China headquarters and licenses x86 Geode technology to Chinese entities.
- 2016AMD announces a joint venture (THATIC) with Chinese partners to license and build x86-compatible CPUs.
- 2019-06The US government places AMD's Chinese joint venture and partners on its export control Entity List.
- 2024China accounts for approximately 24% of AMD's total revenue, or about $6.2 billion, before significant impact from new restrictions.
- 2025-05AMD forecasts a $1.5 billion revenue decline for 2025 due to US export restrictions on advanced semiconductors to China.
- 2026-01US implements new licensing reviews for advanced AMD chips like MI325X, and conditional sales for MI308 to China begin.
Sources (9)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
Weekly AI Recap
Read this week's curated digest of top AI events →
AI-curated news aggregator. All content rights belong to original publishers.
Original source: cnBeta (Full RSS) ↗
This is a summary, not the original. Read the source, or get the weekly briefing.
The weekly digest
One email a week. Unsubscribe anytime.
