DeepSeek’s latest disruptive AI training method could become the industry standard

DeepSeek, the great Chinese Artificial Intelligence (AI) disruptor, has announced a new idea for training AI more efficiently than ever before and it’s believed that it might end changing the way that all foundational AI models are trained in the future, which analysts say would have a massive impact on the industry.

The Chinese AI startup published a research paper on Wednesday, describing a method to train Large Language Models (LLMs) that could shape "the evolution of foundational models," it said.

The paper, co-authored by its founder Liang Wenfeng, introduces what DeepSeek calls "Manifold-Constrained Hyper-Connections," or mHC, a training approach designed to scale models without them becoming unstable or breaking altogether.

As language models grow, researchers often try to improve performance by allowing different parts of a model to share more information internally. However, this increases the risk of the information becoming unstable, the paper said. DeepSeek's latest research enables models to share richer internal communication in a constrained manner, preserving training stability and computational efficiency even as models scale, it added.

Wei Sun, the principal analyst for AI at Counterpoint Research said on Friday that the approach is a "striking breakthrough."

DeepSeek combined various techniques to minimize the extra cost of training a model, Sun said. She added that even with a slight increase in cost, the new training method could yield much higher performance.

Sun said the paper reads as a statement of DeepSeek's internal capabilities. By redesigning the training stack end-to-end, the company is signaling that it can pair "rapid experimentation with highly unconventional research ideas."

Deepseek can "once again, bypass compute bottlenecks and unlock leaps in intelligence," she said, referring to its "Sputnik moment" in January 2025, when the company unveiled its R1 reasoning model. The launch shook the tech industry and the US stock market, showing that the R1 model could match top competitors, such as ChatGPT's o1, at a fraction of the cost.

Lian Jye Su, the chief analyst at Omdia, a technology research and consulting firm said that the published research could have a ripple effect across the industry, with rival AI labs developing their own versions of the approach.

"The willingness to share important findings with the industry while continuing to deliver unique value through new models showcases a newfound confidence in the Chinese AI industry," Su said of DeepSeek's paper. Openness is embraced as "a strategic advantage and key differentiator," he added.

The paper comes as DeepSeek is reportedly working toward the release of its next flagship model R2, following an earlier postponement.

R2, which had been expected in mid-2025, was delayed after Liang expressed dissatisfaction with the model's performance, according to a June report by The Information. The report said the launch was also complicated by shortages of advanced AI chips, a constraint that has increasingly shaped how Chinese labs train and deploy frontier models.

While the paper does not mention R2, its timing has raised eyebrows. DeepSeek previously published foundational training research ahead of its R1 model launch.

Su said DeepSeek's track record suggests the new architecture will "definitely be implemented in their new model."

Sun, on the other hand, is more cautious. "There is most likely no standalone R2 coming," Sun said. Since DeepSeek has already integrated earlier R1 updates in its V3 model, the technique could form the backbone of DeepSeek's V4 model, she added.