On January 20, DeepSeek, led by Chinese quant hedge fund manager Liang Wenfeng, released R1. It caught Silicon Valley and the AI world by surprise because the LLM could match or surpass some of the capabilities of rival models built by OpenAI, Google, and Meta.

Those US companies poured billions of dollars into acquiring highly advanced chips and data to build models that can solve complex problems. DeepSeek, however, appears to be building models that can perform at similar benchmarks — at a fraction of the cost.

In a paper released late December, DeepSeek researchers estimated that they built and trained their V3 model for under $6 million using about2,000Nvidia H800 chips.

US companies, like Meta, are racing to secure billions of dollars worth of Nvidia H100 chips to build their chatbots. H100 chips are Nvidia's flagship GPUs. Due to US sanctions, China cannot import H100s and instead imports H800s, which have lower data transfer rates, Reuters reported.

Spokespeople for Deepseek, Meta, and OpenAI did not respond to a request for comment.

That a little-known Chinese startup is closing the gap with some of the largest tech companies in the world with significantly fewer resources could undercut US efforts to build an AI moat against global competitors.

Following President Donald Trump's inauguration, OpenAI announced a joint venture with the federal government to spend $500 billion on AI infrastructure over the next four years.

OpenAI's ChatGPT currently sits at number two on the same Apple chart.