Chamath Palihapitiya Warns Of AI Shift: 'The Battle Is Now More About AI Inference Vs Training' As DeepSeek Sparks Nvidia Selloff

Billionaire investor Chamath Palihapitiya has issued a stark warning about the shifting AI landscape, following Chinese startup DeepSeek’s demonstration of a cost-efficient AI model that triggered a $600 billion selloff in Nvidia Corp. NVDA shares.

What Happened: In a detailed analysis shared on X, the former Facebook executive emphasized that the AI industry’s focus is pivoting from training to inference capabilities.

“The battle of usage is now more about AI inference vs Training,” Palihapitiya noted, suggesting that while the U.S. should maintain export controls on AI training chips, inference chips should be treated differently.

AI training is the process of teaching an AI model to recognize patterns in data, while AI inference is the process of using that trained model to make predictions.

Drawing a parallel to nuclear technology, Palihapitiya argued that while the U.S. shouldn’t export advanced AI training capabilities, it should promote its inference solutions globally. “We should never export our knowledge of enriching uranium… but we should export our ability to build nuclear energy if it can help advance American priorities,” he explained.

DeepSeek, a Chinese AI startup, has released DeepSeek-V3, an open-source LLM that matches the performance of leading U.S. models while costing far less to train.

The large language model uses a mixture-of-experts architecture with 671B parameters, of which only 37B are activated…
— Chamath Palihapitiya (@chamath) December 29, 2024

The venture capitalist, who disclosed his stake in AI chip company Groq, called for urgent collaboration with Middle Eastern allies to establish global inference infrastructure. He warned of potential market volatility affecting major tech stocks, particularly noting that Nvidia faces the highest risk while Tesla Inc. TSLA remains least exposed.

Why It Matters: DeepSeek’s achievement – developing a competitive AI model for just $5.57 million compared to traditional costs in the hundreds of millions – prompted Palihapitiya to criticize U.S. innovation strategy. “We’ve been running towards the big money/shiny object spending programs… vs thinking through the problem more cleverly,” he stated.

The commentary comes as Nvidia acknowledged DeepSeek’s compliance with export controls, while major tech companies like Microsoft Corp. MSFT and Meta Platforms Inc. META maintain substantial AI infrastructure spending plans exceeding $60 billion each for 2025.

Read Next: