About a year ago, DeepSeek shook up the AI world.
The prospect of a new model from a relatively unprecedented Chinese AI research institute that challenges assumptions of American dominance in the field sent markets into a panic and the stock prices of some major Western tech companies plummeted.
Nvidia has plummeted 17%, wiping nearly $600 billion from its market capitalization. us chip manufacturer broadcom also fell by 17%, ASML It fell 7% in one day.
Eleven months later, these companies have not only recovered, but are continuing to grow. Nvidia became the first company to reach a $5 trillion valuation in October, Broadcom stock is up 49% through 2025, and ASML stock is up 36%.
“January (Deep Seek) (R1) changed the global perception of the frontier model cost curve and China’s competitiveness, causing widespread and visible repricing. And it did so in a way that hit the semiconductor and hyperscaler narratives squarely,” Gartner senior director analyst Haritha Khandabatu told CNBC.
Since then, DeepSeek has released seven new model updates. Nothing has caused the kind of waves seen in January. So why isn’t the market responding?
shock factor
Founded in 2023, DeepSeek released a free, open-source large-scale language model (LLM) called V3 in late 2024, which it said was trained using less powerful chips and delivered at a fraction of the cost of models built by the likes of OpenAI and Google.
A few weeks later, in January 2025, the company released its inference model R1, which achieved similar benchmarks or outperformed many of the world’s leading LLMs.
Alex Pratt, senior analyst at investment firm DA Davidson, told CNBC that China AI Lab’s January release “really surprised the market.” “The narrative[at the time]was that China was nine to 12 months behind the United States.”
Morningstar senior equity analyst Brian Colello told CNBC that the promise of models that achieve similar results to state-of-the-art systems using less compute has heightened market concerns that demand for AI infrastructure will be affected, hurting the bottom line of companies like Nvidia.
“In fact, we don’t see a slowdown in spending in 2025, and looking forward, we expect spending to accelerate from 2026 onwards.”
There are also types of releases that DeepSeek has made since January. They were all updates to the V3 and R1 models rather than completely new models.
While DeepSeek’s recent model releases are “credible incremental changes” in efficiency and functionality, the market views them as “continuation and consolidation rather than new shockwaves,” Khandabatu said.
limited computing
One reason DeepSeek hasn’t released new models is likely due to computing limitations, analysts told CNBC.
“Computing is a big bottleneck,” Pratt said. “There’s only so much algorithmic research you can do to discover so much architectural ingenuity.”
The AI company postponed the release of its R2 model, originally scheduled for May, citing challenges with training it on its own Huawei chips, the Financial Times reported in August.
The magazine said Chinese authorities had encouraged DeepSeek to use the processor in the face of Nvidia’s most aggressive chip export restrictions, aiming to reduce dependence on U.S. alternatives. DeepSeek has been contacted for comment on this report.
“China has been limited in the amount of computing power it has access to in recent years, largely because of U.S. chip sales restrictions,” Chris Miller, author of “The Chip Wars,” told CNBC.
“If you want to build advanced models, you need access to advanced computing.”
In a research paper published earlier this month, DeepSeek said it recognized “certain limitations when compared to frontier closed-source models” that include computing resources such as Gemini 3.
The market is also reassured by the continued US leadership in AI with the release of new advanced models from Western frontier labs.
In August, OpenAI announced GPT-5, Anthropic released Claude Opus 4.5, and Google announced Gemini 3 in November.
“Competition among these providers is intense with rapid model releases and incremental feature improvements,” Gartner analyst Arun Chandrasekaran told CNBC. “As a result, concerns about a sudden commoditization shock have allayed.”
However, there are signs that DeepSeek is gearing up to release more significant models in the coming months. On New Year’s Eve, the company published a paper detailing a more efficient way to develop AI models.
Wedbush Securities’ Dan Ives thinks there are more market shocks to come. “Some of these moments that we’ve seen, we’ll continue to see next year,” he told CNBC.
“There will be another deep seek.”
