Lead
Chinese artificial intelligence startup DeepSeek has launched preview versions of its latest models, V4-Pro and V4-Flash, a year after its previous model R1 sent shockwaves through the global tech sector. The release comes amid a broader push by Chinese firms to adapt the new model to domestic hardware, including chips from Huawei Technologies.
Coverage Comparison
Al Jazeera reported that DeepSeek unveiled the preview versions on Friday, touting the ability to compete with US rivals such as OpenAI and Google. According to the outlet, the V4-Pro model beats all rival open models for maths and coding, while the V4-Flash offers similar reasoning abilities but faster response times and more cost-effective pricing. The release follows DeepSeek-R1, which stunned the tech sector in January last year with capabilities comparable to ChatGPT and Gemini.
The South China Morning Post (SCMP) covered the model’s adoption by domestic semiconductor manufacturers and AI chipmakers. Huawei was among the first to act, with the V4 fully adapted to its Ascend 950PR chip platform, according to SCMP. The company said its entire Ascend SuperNode product line was “fully adapted” to DeepSeek V4 for model inference, with the adaptation announced in a livestream on Friday.
The SCMP also reported Nvidia CEO Jensen Huang’s comments on a podcast, where he said it would be a “horrible outcome” for the US if DeepSeek optimised its models on Huawei chips. Huang expressed concerns about China becoming superior to the US in AI development if “future AI models are optimised in a very different way than the American tech stack.”
Key Claims
According to SCMP, the optimised model delivers up to 2.87 times the single-card inference performance of Nvidia’s China-specific H20 processor, while improving multimodal generation efficiency by 60 per cent. Deployment costs are reported to be about one-tenth of comparable GPT-based services.
SCMP also reported that China’s three largest internet groups – Alibaba Group Holding, ByteDance and Tencent Holdings – have collectively placed orders for hundreds of thousands of Ascend 950 processors following the V4 launch, citing the Financial Times.
Huawei’s next-generation Ascend 950PR and 950DT chips enjoyed “day zero” adaptation to DeepSeek’s V4 model, with the company stating in a livestream that its entire Ascend SuperNode product line was “fully adapted” for model inference. The collaboration underscores progress in China’s tech self-reliance amid US efforts to block access to advanced semiconductor chips and chipmaking equipment.
SCMP reported that inference demand is expected to surpass training demand globally by 2030, according to McKinsey & Company. Huawei said its Ascend 950PR supernodes would “ship at scale” in the second half of the year.
Perspectives
Nvidia (US viewpoint): Nvidia CEO Jensen Huang said on the Dwarkesh Podcast that if DeepSeek optimised its models on Huawei chips, it would be “a horrible outcome” for the US. He warned that if China’s AI technology becomes the standard, China “will become superior to” the US.
US lawmakers: US lawmakers and experts have accused China of buying “what they can” and stealing “what they cannot” in the AI industry, calling for the government to evaluate putting DeepSeek, Moonshot AI and MiniMax on the entity list for export control, as reported by SCMP.
China’s tech sector: Chinese companies, including Huawei and Cambricon, have moved quickly to adapt DeepSeek’s V4 to domestic hardware, with Huawei touting “day zero” adaptation and significant performance improvements, reflecting a national priority for technological self-reliance.