DeepSeek rolled out a bigger AI model and cut the price — then Huawei showed up almost immediately to run it. The Chinese AI startup’s new V4 model is designed to compete with top systems from OpenAI and Google DeepMind while dramatically lowering costs.
Huawei also pledged full support through its Ascend chips, signaling closer coordination between the model and the hardware it runs on.
A larger model built for scale and lower cost
The South China Morning Post reported that DeepSeek released two versions of its V4 model: a 1.6-trillion-parameter V4-Pro and a 284-billion-parameter V4-Flash. Both models support a context window of up to one million tokens, a major increase over earlier versions.
The company said the models deliver strong cost efficiency while remaining competitive with top closed-source systems. CGTN noted that V4-Pro matches leading models in several areas and improves agent capabilities for multi-step tasks.
Pricing is a key differentiator. V4-Pro costs about $3.48 per million output tokens, according to Fortune — compared with roughly $25 to $30 charged by rivals like Anthropic and OpenAI — while V4-Flash drops to as low as $0.28.The pricing strategy could put pressure on competitors, who are already raising prices and limiting usage to manage demand.
Huawei aligns chips and software from launch
Huawei said its Ascend chips were ready to support the model immediately. According to SCMP, its latest processors achieved “day zero” adaptation with DeepSeek V4, reflecting close coordination between the two companies. The company added that its Ascend SuperNode lineup was fully adapted for V4 inference workloads.
“The entire Ascend SuperNode product line was fully adapted to DeepSeek V4 for model inference, which had significantly improved due to the two companies’ close collaboration before the model’s release,” the Huawei engineers explained during the livestream.
CGTN also reported compatibility across multiple chip families, including Ascend A2, A3, and 950 series processors. This tight integration extends to Huawei’s Compute Architecture for Neural Networks platform, which has been optimized alongside the model.
Analysts from Huatai Securities also emphasized that “the release of V4 explicitly mentions compatibility with domestic chips,” adding that broader adoption of local GPUs could follow this year.
Short-term limits, bigger stakes ahead
SCMP said that according to DeepSeek, V4 may face throughput challenges until the second half of the year, when Huawei’s Ascend 950PR supernodes are expected to ship at scale. Even so, the trend is hard to miss. As inference demand grows, how efficiently models run is becoming just as important as how they are trained.
DeepSeek’s lower pricing and hardware alignment could put pressure on rivals, especially as the gap with US models continues to narrow.
Read more: Huawei is pushing ahead on multiple fronts, including its Pura X Max foldable that beats Apple and Samsung to a new format in China.
