The heyday when venture capitalists were content to shell out billions to the latest AI startup while researchers burned their cash with nothing to show for it, may be all but over. There will soon be a “reckoning” for AI companies that fail to make profits as the new technology matures, Kai-Fu Lee, chairman and CEO of Sinovation Ventures, said at the Fortune Innovation forum in Hong Kong on Wednesday.
Lee said too many large language model (LLM) startups focus on pursuing breakthrough advancements and too little on commercializing their work. “Many of the LLM companies out there are run by researchers who are only interested in creating a great model,” he said in a conversation with Fortune editor-in-chief Alyson Shontell. “That phase of the science fair has to end.”
If there’s one thing the three leading U.S. mega-cap tech stocks all have in common, it’s that they’ve successfully monetized an emerging technology: Microsoft with the personal computer, Apple and Google with the smartphone.
Lee, former president of Google China and a researcher in the field himself, founded his own AI startup in March 2023. The company, called 01.AI, was valued at more than $1 billion in less than eight months.
Lee said his own former employer Google is a cautionary tale. Even with the densest network of AI talent in existence in the world to this day, he argued that Google lost its edge over OpenAI because it wasted time and resources by indulging the competing schemes of its employees.
“If you have too many researchers and a culture where everyone can try out their ideas, you will quickly run out of money as a startup,” he said.
Huawei’s focus versus Google’s ‘let a hundred flowers bloom’
Lee argued that if his company is ever to be among the world’s leaders in this field, it must be brutally efficient with every dollar it spends.
On Wednesday, the AI expert pointed to Huawei as an example of how such a focus could work in practice. China’s largest telecom equipment maker took advantage of an obscure advance by Turkish IT researcher Erdal Arıkan and invested its efforts almost exclusively in commercializing his polar code breakthrough. This ultimately allowed them to surpass larger Western competitors such as Ericsson and control the majority of the 5G mobile network market.
“That made all the difference,” Lee said. “We take the same approach of being very, very diligent about saving GPU [costs].”
Thanks to its focus on efficient execution, he believes 01.AI – which publishes all its research on open sites like Hugging Face – has narrowed the gap with US companies like OpenAI from eight years to less than 12 months in just a year.
AI rivals that instead embrace Google’s strategy of “letting a hundred flowers bloom,” as Lee put it, would struggle to become profitable by comparison.
“There comes a point of reckoning when investors start saying: what do you have to show for yourself?” Lee said. “What is your profit and loss account? What is your income? What is your growth? When will you break even?”
If an AI startup doesn’t have a compelling answer, the days of the “science fair” are over.