Over the past few weeks, Chinese AI startup DeepSeek has been making waves in the tech industry. It has outperformed OpenAI’s o1 reasoning model in areas like math, science, and coding, all while keeping development costs significantly lower. To put it into perspective, DeepSeek trained its R1 V3-powered AI model for a mere $6 million using reinforcement learning techniques, which is only about 3% of what it cost to develop ChatGPT.
This impressive achievement has led to a surge in interest for DeepSeek, knocking ChatGPT off the top spot as the most downloaded free AI app on Apple’s App Store in the U.S. Despite this, former OpenAI board member Helen Toner believes that DeepSeek has the potential to grow even more and dominate the AI landscape, but it faces hurdles due to the U.S. government’s decision in 2023 to restrict the export of advanced AI chips to China.
Toner has suggested that removing these restrictions could be a major win for China and DeepSeek. Talking to Fortune, she expressed hope that Nvidia might convince even someone like former President Trump that these controls are stifling U.S. industry and should be lifted.
Amid DeepSeek’s rise, there are allegations that it utilized data from Microsoft and OpenAI without permission, potentially infringing on copyrights. Microsoft and OpenAI themselves have been at the center of multiple copyright infringement legal battles. OpenAI CEO Sam Altman has defended their practices, stating that copyright laws don’t necessarily forbid using copyrighted material for training AI models. He acknowledges that privileged data is essential for creating powerful tools like ChatGPT.
Toner is uncertain about whether Trump’s administration will continue enforcing the export rules set by Biden. “It’s a big question whether Trump will continue where the previous administration left off,” she commented.
Interestingly, during his final days in office, President Biden introduced measures to tighten these export rules, aiming to stop China from circumventing restrictions. These included blocking China from working around NVIDIA by using Huawei and TSMC through shell companies. Toner noted the uncertainty of Trump’s potential reaction to these fast-moving developments in AI.
As the R1 development by DeepSeek remains somewhat cryptic, anticipation builds to see how it stacks up against leading AI models. This curiosity is especially keen after OpenAI announced a staggering $500 billion investment in its Stargate project, aimed at constructing data centers across the U.S. to power its AI advancements.
“So far, DeepSeek is playing catch-up rather than leading the charge,” noted Toner. “China is pulling out all the stops to keep up with the U.S. in AI and is keeping pace well, but saying they’re ahead is certainly an overstatement.”
