This response follows DeepSeek's recent buzz after launching a popular chatbot and AI model at a much lower cost than what U.S. companies like DeepMind and OpenAI reportedly spent.
While speaking at the Artificial Intelligence Action Summit in Paris, Hassabis told Bloomberg Television that the $5.6 million figure from DeepSeek likely only reflects the expenses for the final training phase of their AI model, which is just a small part of the overall development costs.
“They seem to have only reported the cost of the final training round, which is a fraction of the total cost,” he noted.
DeepSeek's claim of using older Nvidia chips for training has raised questions among researchers and industry experts. Reports suggest that U.S. authorities are investigating whether the startup bypassed a chip ban by acquiring semiconductors through Singapore.
Moreover, Bloomberg News indicated that OpenAI and Microsoft are looking into whether a group associated with DeepSeek utilized a method called “distillation” to gather data from OpenAI’s models. Distillation is when one AI model pulls outputs from another for training. Hassabis hinted that DeepSeek might have depended on Western models for this process but didn’t go into specifics.
On the topic of AI development, Hassabis dismissed the notion that DeepSeek's rise has changed the game economically.
“We don’t see any new silver bullet technologies. DeepSeek is not an outlier on the efficiency curve,” he stated.
He also pointed out that Google’s AI model, Gemini, is more efficient than DeepSeek when it comes to training-to-performance and cost-to-performance ratios.
“We just don’t talk about it very much,” he added.
Even with all the excitement around DeepSeek’s budget-friendly methods, big names in Silicon Valley, like Alphabet Inc., are still all in on pouring money into AI.
Just last week, Alphabet revealed it’s planning to shell out $75 billion on capital expenses in 2025, mainly targeting its cloud-computing sector and AI services like Gemini, which is being rolled out into Google’s search and other offerings.
Earlier this January, Microsoft also shared its intention to invest about $80 billion in AI-driven data centers to train AI models worldwide.
Brad Smith, Microsoft’s Vice Chairman and President, mentioned this in a blog post that emphasized the company’s dedication to AI as part of a larger strategy to tackle China’s bold moves in the tech arena.