When DeepSeek unexpectedly outperformed existing AI models, the world took notice. It wasn’t just an incremental improvement – it was a fundamental shift. The most shocking part? DeepSeek achieved this with significantly fewer resources than anyone thought possible.
For a moment, the implications seemed clear: if AI could be trained more efficiently, the insatiable demand for NVIDIA’s chips might finally slow down. Investors panicked. NVIDIA’s stock took a nosedive. After all, if AI models could achieve better performance with fewer GPUs, wouldn’t cause problems for the dominant AI hardware supplier?
But then, something fascinating happened. Instead of pulling back, the biggest players – Meta, Microsoft, Google, Open AI – doubled down on spending.
In theory, a breakthrough like DeepSeek’s should have reduced the need for NVIDIA’s expensive chips. In practice, it did the opposite. Mark Zuckerberg himself framed it in game-theoretic terms:
“The incentive to not lose means that you are willing to spend. You will spend anything not to lose this.”
This is the essence of the game where each player drives full speed toward AGI (Artificial General Intelligence), knowing that slowing down could mean losing the entire game. Even if efficiency increases, the sheer magnitude of AI ambitions means companies won’t invest less – they’ll simply do more with the same resources.
The result? NVIDIA still wins.
The short-term market reaction to DeepSeek’s breakthrough made sense on paper: if AI models become more efficient, fewer GPUs should be needed. But in reality, no company can afford to use those efficiency gains to slow down. Instead, they use them to scale up. Meta announced a $65 billion AI investment – after DeepSeek’s performance was revealed. Also none of the Stargate investors pulled out and scaled back their commitments.
In a world where everyone needs to stay in the game, efficiency doesn’t lead to reduced spending – it enables even more spending at an even greater scale. As paradoxical as that might sound.