Last month, Chinese AI startup DeepSeek made headlines after unveiling its R1 V3-powered AI model. At a fraction of the development cost, this model surpasses OpenAI’s proprietary o1 reasoning model across several benchmarks, including coding, math, and science.
Consequently, the news caused a dramatic shift in the market, causing NVIDIA to lose up to $600 billion in market valuation in a single day. As a result, NVIDIA fell behind Microsoft and Apple as the third-most valuable company in the world.
However, Meta’s lead AI scientist, Yann Lecun, says there’s a “major misunderstanding” about the billion dollars invested in AI and how the resources will be used. The executive claims that the money invested in US-based AI firms was dedicated to inference, not to train AI models.
For context, inference is where AI models leverage their trained knowledge to new data. According to the scientist, the cost of inference will likely rise as AI models become more advanced:
“Once you put video understanding, reasoning, large-scale memory, and other capabilities in AI systems, inference costs are going to increase. So, the market reactions to DeepSeek are woefully unjustified.”
While speaking to Business Insider, Positron founder Thomas Sohmers shared Lecun’s sentiments about the market’s reaction to DeepSeek, indicating:
“Inference demand and the infrastructure spend for it is going to rise rapidly. Everyone looking at DeepSeek’s training cost improvements and not seeing that is going to insanely drive inference demand, cost, and spend is missing the forest for the trees.”
R1 is an open-source AI model, trained using $6 billion and the reinforcement learning technique. On Apple’s App Store, the model dethroned ChatGPT as the most downloaded free AI app in the US. As such, DeepSeek will need to invest in inferencing as the model rapidly gains popularity and adoption to handle user requests.
Related: Meta AI lead scientist claims “open-source” is the secret ingredient to DeepSeek’s triumph
Following DeepSeek’s R1 launch, Microsoft CEO Satya Nadella touted the model as super impressive, saying that AI developments from China should be taken very seriously. He further indicated that DeepSeek is good for business amid rising concern among investors over the company’s spending on AI despite little profit returns. “DeepSeek has some real innovations,” Nadella added. “When token prices fall, inference computing prices fall, that means people can consume more, and there’ll be more apps written.”
