A small AI lab in China is shaking things up in Silicon Valley. DeepSeek pulled off what many thought was impossible — releasing cutting edge, open source AI models that outperformed the industry’s top players, CNBC reports. Even more surprising? DeepSeek did it on a budget under USD 5.6 mn — compared to OpenAI’s USD 6 bn — using less powerful Nvidia H800 chips instead of the high-end hardware its rivals rely on — all while evading US strict export controls.

(Tap or click the headline above to read this story with all of the links to external sources.)

This may be a wake up call for Silicon Valley to do more with less. DeepSeek honed the art of efficiency by using techniques like model distillation — where small models learn from big models — and advanced methods like Multi-head Latent Attention (MLA) and Mixture-of-Experts. They’re winning over competitors’ users, and instead of guarding their breakthroughs, DeepSeek went open source — a move OpenAI’s Sam Altman revoked for his own company in November of last year.

The US may need to rethink its strategies. As existing sanctions to stifle international competition appear less effective, and with the rise of open source models, DeepSeek’s success can encourage a new wave of innovation and level the playing field for smaller players worldwide.

Where are the competitors? OpenAI has launched Operator, an AI agent that can book hotels, planes and can even online shop for you, Business Insider reports. Operator is available to ChatGPT Pro subscribers in the US, with plans to expand globally and to lower tier subscription plans in the coming months.

But there are challenges ahead for DeepSeek: The salmon-backed paper mentions that US tech companies are rapidly expanding their AI computing capabilities, deploying massive arrays of Nvidia’s newest Blackwell GPU chips, which could potentially outpace the processing power available to Chinese competitors.

Tags: