DeepSeek – lessons from AI’s 'Sputnik moment'

Cost/performance improvements to DeepSeek R1, a new AI reasoning model from Chinese company High-Flyer, have been making waves in the tech world and investment community. The big deal was the perception that early reports that the new approach was as much as 45 times as cost-effective to train and run as the current version from OpenAI and Anthropic. And it's open source – well, sort of.

This was shortly followed by the Chinese company Alibaba’s new Qwen2.5-Max model, which claimed to have an even better ranking than DeepSeek on AI benchmarks. However, it's unclear how it faired regarding training or inferencing cost-effectiveness.

The stock prices of AI companies collectively dropped over a trillion dollars over a day before they partially recovered. This initial response seems to have been a reaction to the naïve reasoning that cheaper AI will mean fewer chips. But then those dumping stocks ...