DeepSeek V4 Will Run Entirely on Huawei Ascend 950PR Chips — Alibaba, ByteDance, Tencent Order Hundreds of Thousands of Units, Prices Up 20%

On April 3, 2026, multiple sources confirmed that DeepSeek V4 — the next major version of Chinas leading open AI model — will run entirely on Huawei-made chips, marking a major milestone in Chinas effort to break free from dependence on foreign semiconductor technology.

KEY FACTS:

1. HARDWARE INDEPENDENCE: DeepSeek spent months working with Huawei and chip designer Cambricon to port DeepSeek V4 to Chinese-made Ascend 950PR chips. This is the first time a frontier-class AI model from a leading Chinese lab will run exclusively on domestic hardware.

2. NVIDIA EXCLUDED: In a notable break from industry norms, NVIDIA did not receive early access to DeepSeek V4 — only Chinese chip companies did. This signals a deliberate strategic shift away from Western semiconductor dependency.

3. MASSIVE ORDERS: Chinese tech giants including Alibaba, ByteDance, and Tencent have ordered hundreds of thousands of Huawei Ascend 950PR units to run DeepSeek V4 through their cloud services and integrate it into their AI applications. The surge in demand has pushed chip prices up by approximately 20%.

4. PERFORMANCE SPECS: Huawei claims the Ascend 950PR delivers roughly 2.8x the computing power of NVIDIAs H20 (the chip currently allowed for export to China), though it still falls short of the H200 (which is restricted). Production bottlenecks persist due to US export controls on advanced chip manufacturing equipment.

5. TIMELINE: DeepSeek V4 is expected to launch within the coming weeks. It reportedly features a 1-trillion parameter Mixture of Experts architecture with multimodal capabilities and has faced two prior delays (late January and late February).

STRATEGIC CONTEXT:

This development represents a significant inflection point in the global AI chip race. US export controls were designed to slow Chinas AI development by restricting access to advanced NVIDIA GPUs. Instead, they appear to have accelerated Chinas domestic chip ecosystem:

• Huaweis Ascend 950PR is a competitive inference chip, even if it trails NVIDIAs best
• The massive volume orders indicate real production capacity, not just lab samples
• Chinese cloud providers are building entire inference stacks on domestic silicon
• The 20% price increase suggests genuine supply-demand dynamics, not subsidized adoption

The DeepSeek + Huawei combination creates a fully Chinese AI stack from model to silicon, which has implications for global AI infrastructure competition and supply chain diversity.

IMPLICATIONS FOR AI INFRASTRUCTURE MARKET:

The emergence of a competitive Chinese AI chip ecosystem creates a genuine alternative to the NVIDIA-dominated global market. Cloud providers outside China may eventually have access to Huawei chips as an alternative supplier, potentially reducing NVIDIAs pricing power. For the AI agent industry specifically, this means more diverse hardware options for running inference workloads and reduced risk of single-vendor dependency.