Vinod Khosla has long been one of the most bullish investors in the robotics sector, predicting that the industry will enjoy its own "ChatGPT moment" within the next two years. But he also warns that today's hot robotics startups are attracting too much capital at inflated valuations, and a sweeping shakeout is coming.

In an interview, he said, "I think by 2030, more than half of these companies will be worth less than they are today." He added that the survivors at the top will command "incredibly impressive valuation multiples."

It is, of course, the norm in venture capital that most projects ultimately fail. But Khosla's remarks are notable because China has issued similar warnings about its own globally leading robotics industry. Last year, China's National Development and Reform Commission noted that more than 150 domestic companies were developing humanoid robots, a number it called excessive and a sign of bubble risk. Unitree, a leading humanoid robot maker, listed on the Shanghai Stock Exchange last month, and its shares swung sharply on the first day of trading, prompting Chinese regulators to tighten IPO approvals for humanoid robot startups.

The veteran tech executive and entrepreneur founded Khosla Ventures in 2004, and the fund has backed a number of robotics startups, including Rhoda AI, which trains foundation models for robots, and Aim, which enables autonomous operation of earthmoving equipment. Khosla remains optimistic that robotics will reach a ChatGPT-like inflection point and produce a group of humanoid robot winners. He predicts that by then, humanoid robots with fine manipulation skills will be able to handle entirely new industrial tasks after just one to two hours of data learning.

One core question remains: whose AI model will serve as the "brain" of these robots? Not long ago, models from physical-intelligence specialists such as Physical Intelligence and Skild were seen as frontrunners in this race. But recently, AI companies that train large language models have made major strides in robot control. In recent tests, for example, OpenAI's GPT-6 Astra outperformed some purpose-built robotics AI models. In one case, a robot equipped with Astra was even given a canvas and a brush, and the model completed a painting.

If large language model companies ultimately spark the robotics sector's "ChatGPT moment," Khosla is betting on OpenAI. Given that his fund was one of OpenAI's earliest investors, that preference is hardly surprising. "In my ranking, it's OpenAI, then possibly DeepMind, then maybe Anthropic. Meta is doing some specialized research on home robotics," he said.

Anthropic ranks lower on Khosla's list because, as he put it, "Anthropic is just getting started in robotics and entered too late. So I don't see it as a strong contender."

One factor that could slow robotics research at leading AI companies is their voluntary efforts to limit the pace of model iteration. Such proposals have recently gained popularity among top firms as a way to buy time to study AI safety, but the current restrictions focus only on AI's digital capabilities. Khosla said that once AI models can drive military robots, robotics technology will also be brought into the discussion about limiting development speed. At that point, "national security will be involved." Once the robotics industry reaches a ChatGPT-style inflection point, "I believe this will become a central focus of debate," he said.

Robotics is not the only sector with sky-high valuations. Khosla said, "There are plenty of companies I was interested in investing in but passed on purely because the valuations were too high. That's very common now. This frenzy of monthly funding rounds and ever-rising valuations is not normal."

Khosla keeps a special list of companies with solid business models that he passed on because of excessive valuations. On the other hand, some companies may look expensive at first glance, but their business quality is strong enough that "the high valuation is worth it." He cited Cognition, an AI coding startup he first invested in three years ago. At the time, its valuation looked high, but the company's subsequent growth proved the investment was "well worth it."