What Happened
According to the latest Financial Times report, Huawei’s AI chip business revenue is projected to grow 60% in 2026, reaching approximately $12 billion. The primary driver is Chinese tech giants collectively shifting from Nvidia chips to Huawei Ascend series AI chips.
Key Data
| Metric | Value | Notes |
|---|---|---|
| 2026 Projected Revenue | ~$12B | 60% year-over-year growth |
| Major Customers | Alibaba, Tencent, ByteDance, Baidu | China’s top tech companies |
| Core Product | Ascend 910C Series | Competing with Nvidia A100/H100 |
| Growth Driver | US export controls + domestic substitution | Policy and market dual push |
Background: Why the Collective Shift
Long-term Impact of US Export Controls
Since the US tightened AI chip export controls to China in 2023, Nvidia’s AI chip sales to China have been strictly restricted. Although Nvidia launched special versions (like H20), both performance and supply are unstable.
This has forced Chinese tech giants to seek alternatives, with Huawei Ascend series becoming the most mature option:
| Dimension | Nvidia H20 (Special) | Huawei Ascend 910C |
|---|---|---|
| Compute Power | Restricted | Approaching A100 level |
| Supply Stability | Policy-dependent | Domestic supply chain |
| Software Ecosystem | CUDA mature | CANN rapidly catching up |
| Price | Premium | Cost-effective |
| Technical Support | Restricted | Localized service |
Explosive Demand for Domestic Large Model Training
2026 is the white-hot phase of China’s large model competition:
- Dense releases of Qwen3.6, Kimi K2.6, DeepSeek V4 and more
- Training compute demand growing exponentially
- Every company needs stable large-scale compute supply
In this context, Huawei Ascend’s domestic supply chain advantage becomes highly attractive.
Landscape Assessment
1. China’s AI Compute Ecosystem is Taking Shape
The substantial growth in Huawei’s AI chip revenue is not just a commercial number, but an ecosystem signal:
- Hardware: Ascend chip performance continues to improve, gradually approaching international levels
- Software: CANN (Compute Architecture for Neural Networks) ecosystem is maturing
- Applications: Mainstream frameworks (PyTorch, MindSpore) have improving Ascend support
- Customers: Top tech companies moving from “trial use” to “scaled deployment”
2. Impact on Global AI Chip Landscape
| Region | Trend | Impact |
|---|---|---|
| China | Domestic chip substitution accelerating | Nvidia’s market share in China continues declining |
| US | Export controls tightening | Short-term benefits for US chip companies’ long-term competitiveness |
| Global | Supply chain bifurcation | AI compute may develop “two ecosystems” |
3. Impact on Hyperscaler Capital Expenditure
Reports indicate global hyperscaler AI infrastructure capital expenditure is projected to reach $725 billion in 2026, up 77% year-over-year. Of this:
- $520K per $1M → GPUs and accelerators (Nvidia, AMD, Broadcom custom chips)
- $150K per $1M → Networking and optical
But in the Chinese market, this allocation is changing — Huawei Ascend’s share is eating into Nvidia’s cake.
Impact on Developers and Enterprises
If you are doing AI development in China:
- The maturing Ascend ecosystem means more compute options
- Learning cost for MindSpore framework and CANN toolchain is decreasing
- Huawei Cloud’s Ascend instances offer compelling cost-performance
If you are doing AI development overseas:
- Need to monitor how US-China AI chip ecosystem bifurcation affects model development
- The same model may need adaptation for different hardware backends
- Cross-platform adaptation work in the open-source community will become increasingly important
If you are watching investment opportunities:
- Huawei’s AI chip supply chain (including upstream and downstream suppliers) is a sector worth attention
- Domestic AI software ecosystem (frameworks, toolchains, optimization libraries) also has huge space
- But pay attention to matching valuations with actual performance
Risks and Uncertainties
- Technology gap: Ascend 910C still has a generation gap compared to Nvidia’s latest flagship (e.g., B200)
- Software ecosystem: CUDA’s ecosystem barrier cannot be broken overnight
- Geopolitics: Export control policies may change further
- Competition: Other domestic chip manufacturers like Hygon and Cambricon are also catching up
Huawei AI chip’s 60% growth is an important milestone, but “domestic substitution” is a marathon, not a sprint.