Artificial Intelligence (AI) today is dominated by heavyweights like OpenAI and Google, but a Chinese startup DeepSeek has shaken things up with its AI model named R1. The new model has risen to the global top 10 for AI performance and demonstrated that innovation thrives under scarcity.

The decision of America to restrict Chinese access to Nvidia’s most advanced AI chips inadvertently sparked creativity among Chinese AI developers. DeepSeek used less advanced Nvidia H800 chips to develop R1.

The appeal of R1 lies in its efficiency as it matches o1 performance of OpenAI at just 3%–5% of the cost. Companies no longer feel compelled to pay for the most expensive models when affordable alternatives deliver comparable results. Startups like Anthony Poo’s generative AI company have already shifted to DeepSeek.

The ability of DeepSeek to do more with less is remarkable. The company trained R1 for just $5.6 million. Their use of 2,000 Nvidia chips highlights the efficiency of their process.

The man behind DeepSeek is Liang Wenfeng who previously ran an $8 billion hedge fund. Its team re-engineered their processes to maximize capabilities of the available H800 chips when U.S. export restrictions limited chip availability.

The success of DeepSeek is resonating globally and venture capitalist Marc Andreessen called it as one of the most impressive breakthroughs that he has seen. R1 has also become a trending model on platforms like HuggingFace.

The rise of DeepSeek signals potential challenges for Nvidia. If startups follow the path of DeepSeekthe demand for Nvidia’s premium products might slow.