DeepSeek’s next flagship model has become a fresh example of how much advanced AI still depends on high-end computing hardware. According to insiders cited by the Wall Street Journal, the Chinese AI startup ran into development problems after trying chips from Huawei and other Chinese manufacturers last year, then turned back to Nvidia chips for some training tasks.
The reported problem behind the new model
DeepSeek initially attempted to build its new flagship model using hardware from Huawei and other Chinese chipmakers. The effort did not deliver results that were good enough, according to the source article.
The company then switched to allegedly smuggled Nvidia chips for some training tasks. That move reportedly helped the work progress, and the new model is expected to ship in the coming weeks.
The key point is not only that DeepSeek used Nvidia hardware. It is that the reported fallback came after an attempt to rely on Chinese-made alternatives. For a company working on a flagship AI model, that difference matters because training tasks depend heavily on the quality and availability of the chips used to run them.
Why AI chips are central to the story
Large AI models are not shaped by software alone. The source article places the pressure on hardware: DeepSeek’s model development ran into trouble when the company tried to use chips from Huawei and other Chinese manufacturers, and progress improved only after the company used Nvidia chips for part of the training work.
That makes the episode a practical signal about the current state of AI competition. If a leading Chinese AI startup still needs better hardware to move a flagship model forward, then chip access remains one of the main constraints on model development.
The issue is also bigger than DeepSeek. At a recent conference in Beijing, leading Chinese AI researchers acknowledged that Chinese AI models will struggle to keep pace with US companies without access to better hardware.
Researchers see a difficult race with US labs
The source article names Justin Lin from Alibaba's Qwen team as one of the voices weighing the challenge. He put the odds of overtaking OpenAI or Anthropic within three to five years at 20 percent at best.
That estimate captures the gap as researchers described it: not merely a matter of ambition, but a matter of hardware capability. If Chinese AI teams cannot obtain the level of chips they need, their ability to train competitive models may be limited, even when the talent and model-building work are in place.
For readers following AI development, the important detail is the link between hardware access and model progress. The source does not present this as a side issue. It frames better hardware as a condition for Chinese AI models to keep pace with US companies.
China’s policy push adds another layer
The reported DeepSeek fallback also sits against a broader government priority. The Chinese government is pushing to cut US chip imports and boost domestic production, according to the source article.
That creates a difficult balance. On one side, there is a push to reduce reliance on US chips. On the other, DeepSeek reportedly had to use Nvidia chips after Chinese alternatives did not perform well enough for its new flagship model.
This tension helps explain why the DeepSeek report is significant. It is not only about one company’s model launch. It points to a broader question for China’s AI sector: how quickly domestic hardware can become strong enough to support the most demanding model training work.
What to watch next
The next immediate marker is DeepSeek’s new model, which is expected to ship in the coming weeks. Its release will arrive after a development process that, according to insiders, involved a failed first attempt with chips from Huawei and other Chinese manufacturers and a later move to Nvidia chips for some training tasks.
Several implications follow from the facts in the source:
- DeepSeek remains an important company to watch because its flagship model work reflects the pressure facing Chinese AI startups.
- Nvidia chips remain central because DeepSeek reportedly returned to them when other hardware did not produce good enough results.
- Chinese chipmakers face a high bar because researchers say better hardware is needed for Chinese AI models to keep pace with US companies.
- Domestic production is a strategic priority because the Chinese government is pushing to reduce US chip imports.
The core lesson is straightforward: AI competition is not only a race to build better models. It is also a race to secure the hardware needed to train them. DeepSeek’s reported experience shows how quickly that hardware question can move from policy debate to product reality.