So, you're telling me that China's DeepSeek built an advanced AI model for just US$6 million (RM26.8 million)? That's like saying you built a spaceship on a bicycle budget. Naturally, Google DeepMind's CEO, Demis Hassabis, isn't buying it. According to Hassabis, DeepSeek's cost claims are "exaggerated and a little bit misleading." (Translation: "Nice try, but AI doesn't run on pocket change.") Speaking to Bloomberg TV, he explained that DeepSeek "seems to have only reported the cost of the final training round, which is a fraction of the total cost."

No AI Cost-Cutting Hack?

DeepSeek made waves last month when it launched a chatbot and AI model for a fraction of what major U.S. players like OpenAI and Google have been spending. Some thought this might be a game-changer for AI development costs, but Hassabis quickly shut that down.

"We don't see any new silver bullet technologies," he said at the Artificial Intelligence Action Summit in Paris. "DeepSeek is not an outlier on the efficiency curve."

Translation? There's no secret sauce here—just some creative accounting.

Wait… Where Did the Chips Come From?

DeepSeek reported spending US$5.6 million (RM25 million) on computing costs, using older Nvidia chips to train its model. But some experts are giving those numbers the side-eye. Meanwhile, U.S. authorities have launched an investigation to check whether DeepSeek bypassed chip bans by purchasing through Singapore.

And speaking of questionable tactics, Bloomberg News reported that OpenAI and Microsoft are looking into whether a DeepSeek-linked group borrowed (read: possibly stole) AI model data from OpenAI using a process called distillation. That's when one AI learns by copying another AI's outputs—basically the machine-learning version of peeking at your classmate's test.

Google's Not Worried (Or So They Say)

Despite all the buzz around DeepSeek's supposed cost efficiency, Alphabet (Google's parent company) isn't exactly panicking. In fact, it's planning to spend US$75 billion (RM335 billion) in 2025 on cloud computing and AI projects like Gemini. That's a whole lot of zeroes, but according to Hassabis, it's justified.

"Gemini is more efficient than DeepSeek in terms of training to performance or cost to performance," he said. "We just don't talk about it very much."

Google DeepMind's CEO, Demis Hassabis

According to Hassabis, DeepSeek's cost claims are "exaggerated and a little bit misleading." (Translation: "Nice try, but AI doesn't run on pocket change.") Speaking to Bloomberg TV, he explained that DeepSeek "seems to have only reported the cost of the final training round, which is a fraction of the total cost."

No AI Cost-Cutting Hack?

DeepSeek made waves last month when it launched a chatbot and AI model for a fraction of what major U.S. players like OpenAI and Google have been spending. Some thought this might be a game-changer for AI development costs, but Hassabis quickly shut that down.

"We don't see any new silver bullet technologies," he said at the Artificial Intelligence Action Summit in Paris. "DeepSeek is not an outlier on the efficiency curve."

Translation? There's no secret sauce here—just some creative accounting.

Wait… Where Did the Chips Come From?

DeepSeek reported spending US$5.6 million (RM25 million) on computing costs, using older Nvidia chips to train its model. But some experts are giving those numbers the side-eye. Meanwhile, U.S. authorities have launched an investigation to check whether DeepSeek bypassed chip bans by purchasing through Singapore.

And speaking of questionable tactics, Bloomberg News reported that OpenAI and Microsoft are looking into whether a DeepSeek-linked group borrowed (read: possibly stole) AI model data from OpenAI using a process called distillation. That's when one AI learns by copying another AI's outputs—basically the machine-learning version of peeking at your classmate's test.

Google's Not Worried (Or So They Say)

Despite all the buzz around DeepSeek's supposed cost efficiency, Alphabet (Google's parent company) isn't exactly panicking. In fact, it's planning to spend US$75 billion (RM335 billion) in 2025 on cloud computing and AI projects like Gemini. That's a whole lot of zeroes, but according to Hassabis, it's justified.

"Gemini is more efficient than DeepSeek in terms of training to performance or cost to performance," he said. "We just don't talk about it very much."

Translation? Google is basically saying: "We could flex on DeepSeek, but we have better things to do."

Meanwhile, DeepSeek? No comment. Maybe they're too busy enjoying all the free PR.

Final Thoughts

DeepSeek's claim of developing an AI model on a US$6 million (RM26.8 million) budget might not be the full story. Google's AI chief argues the real costs are much higher, and investigations are underway into whether DeepSeek sourced banned Nvidia chips or borrowed OpenAI's data. Meanwhile, Google is staying on its US$75 billion (RM335 billion) spending spree—because, apparently, AI doesn't come cheap.