Leading AI Researcher at Meta Responds to DeepSeek Criticism as ‘Unjustified’
The Impact of DeepSeek on AI Competition
Overview of DeepSeek’s Emergence
DeepSeek, a rising player in the AI sector from China, has recently provoked strong reactions within Silicon Valley. The company unveiled a new model that reportedly outperformed well-established competitors like OpenAI and Meta in various evaluations. This performance has reignited concerns among U.S. companies regarding the intensifying competition from Chinese tech firms.
Pricing Disparities and Market Reactions
DeepSeek’s models are notably more affordable compared to those offered by leading developers. For instance, the latest reasoning model from DeepSeek, referred to as R1, costs approximately $0.55 for one million tokens processed. In contrast, OpenAI’s competing model, o1, charges $15 for the same amount. This significant price difference has raised eyebrows and sparked frenzy within the markets.
On a recent Monday, news of DeepSeek’s success led to a substantial sell-off in tech stocks, resulting in a staggering loss of around $1 trillion in market value across various companies. Notably, Nvidia, renowned for producing high-end chips, saw its valuation plummet by nearly $600 billion, drawing attention to the market’s concern over the implications of DeepSeek’s advancements.
Insights from Industry Experts
Yann LeCun, Meta’s chief AI scientist, has been vocal in his belief that the market’s panic is unfounded. In a post on Threads, he emphasized that significant investments in AI primarily focus on inference rather than the foundational training of AI models. Inference refers to how AI systems utilize their training to process new data, which is an essential function for applications like generative chatbots, including ChatGPT.
As AI tools evolve and become increasingly complex, the costs associated with inference are expected to rise. LeCun stated, "Once you introduce features like video understanding and large-scale memory, the costs for inference will increase." He believes the financial market’s reactions to DeepSeek are “woefully unjustified.”
The Growing Demand for Inference
Thomas Sohmers, co-founder of Positron, a hardware startup geared towards AI model inference, echoed LeCun’s sentiments. He argued that while DeepSeek has made training its models cheaper, the demand for inference will grow rapidly as its user base expands. Sohmers noted the tendency of investors to focus too narrowly on training expenses, neglecting the broader implications for inference costs and requirements.
As DeepSeek garners more users, it will inevitably face increased demands on its infrastructure for inference. This trend is already prompting a wave of new startups entering the AI inference market, which seek to streamline output generation. Industry experts anticipate that as competition grows, the costs associated with inference might eventually decline for smaller scale operations.
Inference Costs in Different Contexts
However, experts like Wharton professor Ethan Mollick have pointed out that inference costs for high-demand services, such as those provided by models like DeepSeek V3, could be much higher when serving a large audience without charging for access. He noted that while using AI models internally for specific tasks can be inexpensive, large-scale consumer-focused services would likely face significant costs due to high usage demand.
Corporate Responses to the AI Landscape
Amidst these developments, major tech firms are responding by ramping up their investments in AI infrastructure. For instance, Meta’s CEO Mark Zuckerberg announced plans for over $60 billion in capital expenditures for 2025, highlighting the company’s commitment to expanding its AI capabilities further. While specific allocations for inference costs were not detailed, the intent to bolster AI teams and resources was clear.
Additionally, President Donald Trump announced a joint venture, Stargate, aimed at channeling up to $500 billion into AI infrastructure in the U.S., further underscoring the escalating importance of AI in global tech competition.
The developments surrounding DeepSeek spotlight ongoing changes in the AI landscape and underline the importance of understanding inference costs as companies strive to compete in this rapidly evolving market.
