DeepSeek V4 Officially Announces Support for Both NVIDIA and Huawei
NVIDIA CEO Jensen Huang recently stated that if DeepSeek's new model were to launch first on Huawei's platform, it would be a concerning day for the US. He explained this would mean AI models are optimized for Chinese AI hardware, potentially establishing Chinese technology as the global standard.
Previously, reports suggested DeepSeek V4 would support Huawei's Ascend platform, with some even claiming it would be exclusive to Huawei. However, this was unlikely. Today's release of DeepSeek V4 confirms simultaneous support for both Huawei and NVIDIA platforms.
DeepSeek stated in its announcement that DeepSeek V4 and its fine-grained expert parallelism (EP) scheme will be validated on both NVIDIA GPUs and Huawei Ascend NPUs.
This means Huang's biggest concern hasn't materialized – DeepSeek V4 won't exclusively support Huawei or other domestic AI platforms. It will continue to support NVIDIA's CUDA ecosystem, and is actually the first to do so.
Of course, tonight's focus will be on whether DeepSeek V4's immediate inclusion of Huawei's Ascend platform will, like last year, impact NVIDIA's stock price. If the stock remains stable, Huang can rest easy.
Furthermore, there's another aspect of DeepSeek V4 worth noting: the price of V4 Pro has increased 6-8 times compared to the previous V3.2. Outputting one million tokens costs 24 yuan, with a cost of 12 yuan for uncached outputs and 1 yuan for cached outputs. This is a mixed bag compared to other domestic large models like Kimi 2.6 or GLM-5.1, losing the previous super cost-effectiveness.
Many factors influence the cost. After all, V4 Pro is a massive 1.6 trillion parameter model, and the computational cost is inevitably high. The official API page also mentions that, due to high-end computing limitations, the service throughput of the V4 Pro model is currently limited.
However, DeepSeek has also clearly stated that the Pro price is expected to drop significantly after the mass launch of the Ascend 950 supernodes in the second half of the year, so more good news is yet to come.