The robot economy is getting real: NVIDIA, China, LG, and robots building houses

The humanoid race is shifting from bodies to brains

One of the biggest changes in humanoid robotics right now might be happening almost invisibly: the hardware is starting to look increasingly similar. And getting cheaper.

At the World Humanoid Robot Games in Beijing, industry executives noticed that sensors, motors, reducers, and other core components are improving, becoming more standardized, and getting produced at larger volumes.

And that is crushing costs.

EngineAI founder Zhao Tongyang says a humanoid capable of running that might once have cost several million yuan can now be built for around 100,000 yuan — roughly $14,000 U.S.

That changes the competitive landscape: if everybody can increasingly buy decent motors, actuators, sensors, batteries, and compute, the differentiator moves upstairs ... into the robot’s brain.

That means foundation models, vision-language-action models, world models, training data, and the ability to generalize from one task and environment to another.

We’ve spent years asking who can build the best humanoid body. Increasingly, the question is becoming: who can make it useful? And that, of course, is a sign of increasing maturing in the humanoid robot space.

China’s robotics boom hits the stock market

Investors are betting heavily that somebody will make all this work.

Hong Kong and Shanghai have raised more than $54 billion through IPOs and secondary listings so far this year, according to LSEG, already exceeding last year’s total. AI, chips, and robotics are a major part of that enthusiasm.

Of course, Unitree is probably the poster child. The Chinese humanoid and quadruped robot maker closed its first day of trading 460 percent above its IPO price in August.

That does not necessarily mean the fundamentals are 460 percent better (in fact, Unitree shares subsequently fell sharply from their first-day peak).

But it does tell us something important: humanoid robotics has moved from a venture-capital story into a public-markets story. And with AGIBOT, Deep Robotics, and other Chinese robotics companies also looking toward listings, investors are going to start demanding something beyond impressive demos.

Revenue, margins, deployments, and repeat customers are about to matter a lot more.

(Again, another maturity signal!)

LG wants a bigger piece of physical AI

LG Electronics is also treating robotics as something much bigger than a side project. The Korean electronics giant is recruiting researchers and engineers in North America for robotics, AI, and mobility, and says physical AI will be one of its future growth businesses.

That builds on an increasingly serious robotics strategy.

LG has established a dedicated Robotics Business Center, is developing home and commercial robots, and has already announced a major collaboration with Nvidia around robotics training, data infrastructure, and commercialization. That combination is interesting because LG already knows how to manufacture things at enormous scale ... it has motors, actuators, batteries and sensors across the broader LG ecosystem, plus decades of factory and logistics data.

For the next generation of robotics companies, manufacturing is something they need to learn. For LG, the challenge may be the opposite: taking all that manufacturing expertise and successfully turning it into intelligent machines.

Robots are moving into homebuilding

Physical AI is also moving into an industry that has historically resisted automation: construction as Reframe Systems has raised another $40 million to expand its network of robotics-enabled microfactories for building homes.

The company was founded by former Amazon Robotics leaders, and the concept is essentially to take lessons from highly automated fulfillment centers and apply them to housing. Instead of one enormous centralized factory, Reframe wants smaller automated factories closer to where homes are needed.

Its new Massachusetts facility is designed to eventually produce up to 500 multifamily units or 250 single-family homes per year.

Reframe says its approach can deliver homes significantly faster and cheaper than conventional construction. Those are company claims, and they still need to be proven at scale, of course.

But housing is exactly the kind of enormous, labor-constrained industry where even partial automation could have a very big impact. It's beyond-ripe for disruption.

Osaka gets a humanoid station attendant

Meanwhile, in Japan, humanoids are getting customer-service jobs as Osaka Metro has started testing an AI-powered humanoid developed with KDDI and AVITA at its stations.

It uses Google Gemini to answer questions about train transfers, station information, and nearby attractions, it can support around 60 languages, which could make it particularly useful for tourists, and apparently it has already passed one extremely important Osaka benchmark: knowing where to get good food.

(A passenger asked it where to get takoyaki.)

The bigger experiment is whether robots like this can handle routine passenger questions while human station staff spend more time on safety and higher-value tasks.

That is a much less dramatic humanoid use case than backflips or boxing. It might also be a lot more commercially relevant.

It's a valid question whether you need a full embodied humanoid robot for this ... but apparently at least some people think so.

Number of the day: $10 billion

And finally, our number of the day is $10 billion.

That is approximately how much annual revenue NVIDIA is already generating from what it calls physical AI — chips and software for robots, autonomous vehicles, drones, and other machines operating in the real world.

Ten billion dollars.

Today.

And NVIDIA CEO Jensen Huang believes that business can grow tenfold over the coming decade.

China is especially important here. Despite all the competition between the United States and China in AI and semiconductors, Chinese robotics companies remain significant users of American chips and software.

So NVIDIA is building something extremely powerful.

It doesn’t need to know which humanoid company ultimately wins. If a large percentage of them train on NVIDIA GPUs, develop on NVIDIA robotics software, and deploy NVIDIA compute inside their machines, NVIDIA gets to sell the picks and shovels for the robot gold rush.

And apparently, it already is.

Sources

1. Yicai Global — https://www.yicaiglobal.com/news/falling-costs-converging-hardware-push-humanoid-robot-race-toward-ai
2. Fortune — https://fortune.com/2026/08/31/chinas-ai-ipo-boom-shein/
3. The Chosun Daily — https://www.chosun.com/english/industry-en/2026/08/31/VOYXFMWQW5DCRHAD5IRSO2I6NE/
4. The Robot Report — https://www.therobotreport.com/reframe-systems-raises-40m-scale-robotic-microfactories-home-building/
5. News On Japan — https://newsonjapan.com/article/150576.php
6. The Wall Street Journal — https://www.wsj.com/tech/ai/nvidia-wants-to-run-the-worlds-robots-china-is-an-eager-customer-bdf46169?st=nLQELH&reflink=desktopwebshare_permalink
7. Humanoid Mega Hub — https://www.humanoidmegahub.com/news/nvidias-robot-business-is-already-worth-10-billionyear-and-china-is-its-best-cus