Did DeepSeek’s AI model really cost just under $6 million to develop?
Demis Hassabis, Google’s artificial intelligence (AI) chief, says he isn’t so sure, Bloomberg News reported Monday (Feb. 10).
In an interview with Bloomberg Television, Hassabis, who leads Google’s DeepMind, called the idea that the Chinese startup spent so little to develop an AI system that rivals American tech giants “exaggerated and a little bit misleading.”
DeepSeek rocked the tech world last month when it unveiled a chatbot and AI model that it said it developed spending far less than companies like OpenAI. The launch caused many tech stocks to tumble, wiping out nearly $1 trillion in market value.
Gokul Naidu, a consultant for SAP, told PYMNTS last month that DeepSeek “challenges the narrative that innovation must come at an unsustainable cost.”
“For businesses, this means AI could soon be accessible to small and medium enterprises, not just tech giants with deep pockets,” Naidu added.
But Hassabis, interviewed at the Artificial Intelligence Action Summit in Paris, argued that DeepSeek “seems to have only reported the cost of the final training round, which is a fraction of the total cost.”
He also dismissed the notion that the rise of DeepSeek has shaken up the economics of artificial intelligence development.
“We don’t see any new silver bullet technologies,” Hassabis said. “DeepSeek is not an outlier on the efficiency curve.”
His comments come at a time when, as PYMNTS wrote last week, the “AI arms race is getting pricey,” with Google, Meta Microsoft and Amazon planning to collectively spend at least $320 billion on capital expenditures in 2025, the bulk of it for AI.
Meta’s budget for capital expenditures could reach as high as $65 billion, while Google has set aside $75 billion, primarily for data centers, servers and networking infrastructure. Amazon projects it will spend $100 billion, while Microsoft is booking $80 billion to construct data centers, train AI models and launch AI and cloud-based applications.
Microsoft President Brad Smith wrote on the company’s blog last month that “artificial intelligence is the electricity of our age,” setting up the next industrial revolution.
“However, AI requires hefty investments,” PYMNTS wrote.
“Training large language models uses thousands of GPUs (each Nvidia GPU costs about $10,000 or more) or specialized AI chips for a total of tens or hundreds of millions of dollars. Running these AI models at scale also requires high-performance data centers, which need more servers and require more cooling and maintenance.”