DeepSeek’s emergence sent ripples of concern through Silicon Valley, raising questions about the competitive landscape of the rapidly evolving artificial intelligence (AI) sector. However, Yann LeCun, Meta‘s chief AI scientist, has cautioned against overreacting to the Chinese company’s emergence.
What made DeepSeek’s R1 model popular
DeepSeek recently unveiled a new AI model that, according to third-party benchmarks, outperforms existing models from industry giants like OpenAI and Meta, all while reportedly utilising less powerful hardware and a significantly smaller budget.
This news triggered a tech sell-off this week, resulting in a staggering $1 trillion loss in market capitalisation, with Nvidia, a leading supplier of high-end AI chips, experiencing a near $600 billion decline – essentially questioning the need of pouring billions into AI development.
In a Threads post, LeCun argued that the massive investments pouring into US AI companies are primarily driven by the need for inference capabilities, not just training. Inference, the process by which AI models apply their learned knowledge to new data, is crucial for powering generative AI chatbots like ChatGPT. As user demand increases, so does the need for inference processing, leading to higher costs.
Major misunderstanding about AI infrastructure investments:
Much of those billions are going into infrastructure for inference, not training.
Running AI assistant services for billions of people requires a lot of compute.
‘Cost will increase’
LeCun predicts that inference costs will continue to rise as AI systems become more sophisticated, incorporating features like video understanding, reasoning, and large-scale memory. He believes that the market’s reaction to DeepSeek’s advancements is “woefully unjustified.”
Once you put video understanding, reasoning, large-scale memory, and other capabilities in AI systems, inference costs are going to increase.
The only real question is whether users will be willing to pay enough (directly or not) to justify the capex and opex.
So, the market reactions to DeepSeek are woefully unjustified.
Meta CEO Mark Zuckerberg recently announced plans for over $60 billion in capital expenditures for 2025, earmarking funds for expanding its AI infrastructure and teams.
