“If AI is going to eat the world, hardware will be its teeth.”
Haomiao Huang, a founding partner at Matter Venture Partners, said that when the firm closed its second fund at $450 million in September. The fund backs early-stage HardTech companies across semiconductors, robotics and physical AI, AI infrastructure and more.
I spend my days in AI infrastructure, and most of that is software. But I started in electrical engineering, and I’ve been reading a lot about robots this month. The quote stuck with me because it describes something I keep seeing: the AI buildout is turning into a hardware story, and robots are in the middle of it.
The leases say it’s real
Real estate is a good lie detector. Companies don’t sign leases on hype.
According to JLL, as reported by The San Francisco Standard , the amount of commercial space leased by robotics companies in San Francisco this year alone passed the cumulative total from 2020 to 2025. Most of that was driven by autonomous vehicle companies. Most of these companies needed R&D and industrial space, and a third of the leases included offices.
The same story has a useful caveat. Of all AI robotics deals JLL has tracked since 2017, San Francisco accounts for only about 11% of the square footage leased in the Bay Area. The biggest deals go to the East Bay and Silicon Valley, where there’s room for industrial space. Robots need floor space. Startups that start in a loft eventually need a building.
Robots building the AI buildout
The company at the center of that Standard story is the part I find most interesting.
Bright Machines uses AI-enabled robots and software to manufacture data center infrastructure: servers, racks and storage systems. Its robots use vision and sensing for precise assembly and inspection. Its chief growth officer, Fiaz Mohamed, told the Standard, “Everyone wants to deploy new servers [into data centers] faster.” The company says its robots are deployed at more than 150 sites worldwide, and it’s moving to a space five times larger.
Think about the loop there. Robots assemble servers. Servers go into data centers. Data centers run the models. The models make the robots better at handling the real world.
That’s what Huang means by teeth. The software gets the headlines, but somebody has to build the racks.
Still a small slice
I want to keep this grounded, because hype is easy here.
Anthropic’s report What work can robots do? notes that spending on robots is still about 1% of total US equipment investment. Business surveys it cites show 13% of US firms, weighted by employment, use robotics now, and 22% expect to within three years. That’s a big jump, but from a small base.
And from the earlier posts in this series : robots can do most physical tasks in some setting, but they’re cost-competitive on just 0.3% of tasks today. The money flowing in is a bet that cost curve bends faster. It hasn’t yet.
Washington doesn’t see many robots
One more data point that stuck with me, from Kai Williams’ writeup of the SCSP AI+ Robotics summit in Washington on October 14.
The summit’s motivating chart: in 2025, about 354,000 industrial robots were installed in China, according to the International Federation of Robotics. That’s over nine times as many as in the US, and more than the rest of the world combined.
But the detail I’d point to is smaller. Williams wrote that all day he saw only two robots, both versions of Boston Dynamics’ Spot. During a fairly basic demo, a lot of the audience pulled out their phones to record it. He got the sense it was the first time many of them had seen a legged robot. His point was that it’s much easier to see robots working in San Francisco, or in major Chinese cities, than in DC, and that may make it harder for policymakers to understand where the technology really is.
I don’t read that as a knock on anyone. It’s the same thing I’ve seen in software for years. People who’ve never watched a system run make decisions about it from slides. The fix is the same too: put the real thing in the room.
What I take from it
Three things.
AI infrastructure is physical. Chips, servers, racks, power and buildings. Robots are becoming part of how those get built, not just a customer for the models.
Follow the boring signals. Leases, fund closes, equipment spending and battery specs tell you more than demo videos. Right now they say growth is real and still early.
See it running. Whether you’re an investor, a policymaker or an engineer, watch a robot do a real shift before you form an opinion. The demo is the easy part.
Hardware is the teeth. It’s also the part that has to be maintained, powered, cooled and repaired.
Bring floss.