Demis Hassabis, CEO of Google’s DeepMind, has expressed skepticism regarding Chinese startup DeepSeek’s reported development costs for its latest AI model, Bloomberg reports.
The company claims to have developed its AI system for under $6 million, a figure that Hassabis describes as “exaggerated and a little bit misleading.”
DeepSeek’s announcement of its AI model, developed at a fraction of the cost incurred by U.S. counterparts like DeepMind and OpenAI, has garnered significant attention. However, Hassabis contends that DeepSeek “seems to have only reported the cost of the final training round, which is a fraction of the total cost.”
The emergence of DeepSeek has led to discussions about the economics of AI development. Hassabis addressed these concerns, stating, “We don’t see any new silver bullet technologies. DeepSeek is not an outlier on the efficiency curve.”
DeepSeek reported spending $5.6 million on computing costs to train its model using older Nvidia chips. This claim has been met with skepticism from several researchers. Additionally, U.S. authorities have initiated an investigation to determine if DeepSeek circumvented a chip ban by making purchases through Singapore.
Further complicating the narrative, reports have surfaced suggesting that OpenAI and Microsoft are investigating whether a group associated with DeepSeek obtained data from OpenAI using a method known as distillation. This process involves one AI model leveraging outputs from another for training purposes.
Hassabis also noted that DeepSeek appears “to have relied on some Western models to distill from,” though he did not provide specific details.
In response to DeepSeek’s claims, Alphabet, Google’s parent company, has maintained its substantial investment in AI infrastructure. The company recently announced plans for $75 billion in capital expenditures in 2025, focusing on its cloud-computing division and services like Gemini.
Hassabis emphasized the efficiency of Alphabet’s AI initiatives, stating, “Gemini is more efficient than DeepSeek in terms of its training to performance or cost to performance. We just don’t talk about it very much.”
Want to see more of our stories on Google?
P.S. Want to keep this site truly independent? Support us by buying us a beer, treating us to a coffee, or shopping through Amazon here. Links in this post are affiliate links, so we earn a tiny commission at no charge to you. Thanks for supporting independent Canadian media!
