Humanoid robots may be captivating investors, but Unitree’s chief executive is urging a reality check: a machine that can truly cope with an ordinary home remains a distant prospect.

The debate has unfolded against a rush of money and expectation around “physical AI.” Last year, venture capitalist Vinod Khosla forecast a robotics “ChatGPT moment” within two to three years, imagining that most households in the 2030s could have a humanoid assistant. He pictured early models taking on contained kitchen work — chopping, cooking and cleaning — for $300 to $400 a month.

Nvidia CEO Jensen Huang struck an even more bullish note last month, saying the breakthrough “happened a couple of years ago already.” His argument treats the technological inflection point as underway, with the remaining challenge largely one of scaling a vast commercial market.

Then came Unitree’s market validation. Earlier this week, the Chinese robotics maker raised about $900 million in an oversubscribed Shanghai debut; its shares jumped 460% on the first day, putting its valuation near $50 billion. Yet Wang Xingxing’s assessment at the World Robot Conference in Beijing on Thursday was notably less exuberant.

For Wang, the benchmark is not a polished demo or a task performed after extensive training. It is a humanoid placed in a home it has never encountered and able to complete “about 80% of the tasks” through voice or language commands.¹

That standard exposes the gap between robots that work in controlled factory tests and those that can generalize in messy, unpredictable homes. Wang said the milestone could arrive in two or three years under favorable conditions, but might take five to 10 years if progress slows. Unitree’s G1 already sells for $13,500, but affordability alone will not produce a household revolution; reliable physical intelligence is the harder prize.

Khosla and Wang broadly share a near-term possibility for progress. Their split is over confidence: Khosla sees a consumer service poised to spread, while Wang says the industry has yet to prove that its robots can handle the real world.¹