Deepseek v4 launching on Huawei chips exclusively, signaling China's AI independence progress
Deepseek v4 is launching in the coming weeks running exclusively on Huawei chips, marking a major milestone in China's effort to reduce dependency on foreign semiconductors. Chinese tech giants including Alibaba, Bytedance, and Tencent have ordered hundreds of thousands of Huawei Ascend 950PR units to deploy the model through their cloud services.
Deepseek v4 Launching Entirely on Huawei Chips
Deepesk v4 is expected to launch within weeks running entirely on Huawei's Ascend 950PR chips, according to reporting from The Information. The move represents a significant shift in China's AI infrastructure strategy, with the model receiving no early access review from Nvidia—only Chinese chip manufacturers got preview access.
Chip Performance and Demand Surge
Huawei claims the Ascend 950PR delivers approximately 2.8x the computing power of Nvidia's H20 chip, though it remains below the H200's performance. The chip reportedly commands a 20 percent price premium following massive orders from major Chinese tech companies.
Alibaba, Bytedance, and Tencent have collectively ordered hundreds of thousands of Ascend 950PR units to run Deepseek v4 through cloud services and integrate it into their own applications, according to five people familiar with the matter. This concentration of orders from China's largest tech firms signals confidence in both the model and domestic chip viability.
Development Partnership
Deepesk spent months collaborating with Huawei and chip designer Cambricon to port v4 to Chinese-made hardware. The effort reflects a broader strategy to decouple AI development from Western semiconductor supply chains, particularly following US export controls that have constrained chip availability for Chinese companies.
Huawei continues facing production bottlenecks stemming from these same export restrictions, though the surge in Deepseek v4 orders suggests immediate demand exceeds supply constraints.
What This Means
Deepesk v4's exclusive reliance on Huawei hardware marks a tangible outcome of China's multi-year push toward semiconductor self-sufficiency. The decision to exclude Nvidia from early access—a departure from industry norm—signals confidence in domestic alternatives and reduces dependency on external validation. The aggressive procurement by Alibaba, Bytedance, and Tencent indicates the AI market sees viable alternatives to Nvidia, though performance gaps remain. Sustained production constraints and the 20 percent price premium suggest China's chip ecosystem still faces scaling challenges despite technical progress.
Related Articles
Google DeepMind releases Gemma 4 open models with up to 256K context and multimodal reasoning
Google DeepMind has released Gemma 4, an open-weights model family in four sizes (2.3B to 31B parameters) with multimodal capabilities handling text, images, video, and audio. The 26B A4B variant uses mixture-of-experts to achieve 4B active parameters while supporting 256K token context windows and native reasoning modes.
Microsoft releases three multimodal AI models to compete with OpenAI and Google
Microsoft AI released three foundational models on April 2: MAI-Transcribe-1 for speech-to-text across 25 languages, MAI-Voice-1 for audio generation, and MAI-Image-2 for video generation. The company positions these models as cheaper alternatives to Google and OpenAI offerings. Models are available on Microsoft Foundry with pricing starting at $0.36 per hour for transcription.
Microsoft's MAI-Transcribe-1 achieves lowest word error rate on FLEURS, costs $0.36/audio hour
Microsoft has released MAI-Transcribe-1, a speech-to-text model that achieves the lowest word error rate on the FLEURS benchmark across 25 languages, outperforming Whisper-large-V3, GPT-Transcribe, and Gemini 3.1 Flash-Lite. The model runs 2.5 times faster than Microsoft's previous Azure Fast offering and costs $0.36 per audio hour.
Alibaba releases Qwen 3.6 Plus with 1M context window, free tier now available
Alibaba's Qwen division released Qwen 3.6 Plus on April 2, 2026, offering free access to a model with a 1,000,000 token context window. The model combines linear attention with sparse mixture-of-experts routing and achieves a 78.8 score on SWE-bench Verified for software engineering tasks.
Comments
Loading...