Jensen Huang keeps saying $40 trillion. He said it again in Taipei in June. He said it at CES in January. He will probably say it again next month. At some point, the repetition stops being noise and starts being a signal worth taking seriously.

The number is $40 trillion in total addressable market for labor automation through humanoid robots. To put that in perspective: it dwarfs every consumer technology category that has ever existed. EVs. Smartphones. The internet. All of it.

Now here is where it gets interesting.

The crowd is in Tesla. At roughly 208x forward earnings, the market has already priced in a version of Optimus that does not fully exist yet. On the Q4 2025 earnings call, Musk acknowledged the units running inside Tesla factories were operating “primarily for learning, not productive tasks.” Then on April 22, confirming production would start in late July or August 2026 in Fremont, he added that initial output would be “quite slow” and called the production rate “literally impossible to predict” given Optimus has 10,000 unique parts across an entirely new line. On July 1, Musk posted a photo walking the new Optimus production floor. The infrastructure is real. The volume is not there yet. And 208x forward earnings assumes it arrives on time, at scale, without the timeline slippage that has defined this program since 2021.

That is a lot of things that have to go right simultaneously.

Here is where I keep landing. The infrastructure layer is already generating revenue. Nvidia’s Isaac GR00T platform, which now includes a fully open Reference Humanoid Robot unveiled at GTC Taipei on June 1, 2026, is the compute and simulation backbone that robot makers actually use to train their products. ABB Robotics, Agility, Figure, KUKA, Universal Robots, all of them building on Nvidia’s stack. The platform model is simple: Nvidia collects rent on every humanoid trained, simulated, and deployed, regardless of which robot brand wins the market. Q1 FY2027 revenue hit $81.6 billion, up 85% year over year, beating the Wall Street consensus of roughly $78.8 billion. Q2 guidance came in at $91 billion, above the $86.84 billion consensus, even after stripping out all China data center compute revenue. The robotics story is a second, longer-duration growth arc layered on top of an already-dominant data center business. That combination is unusual.

Slight tangent, but it matters: the first deployment wave is already happening in factories, not living rooms. Boston Dynamics’ Atlas production version launched at CES 2026, with all 2026 deployments committed to Hyundai’s Robotics Metaplant Application Center and Google DeepMind. Full deployment at Hyundai Motor Group Metaplant America in Georgia is targeted for 2028, with a planned production capacity of 30,000 robots per year from that facility alone. The history of technology suggests the first commercial deployment is almost always underestimated, because the procurement pipeline behind it is already in place by the time anyone writes the headline.

The part people skip is the component layer. No publicly traded company appears to be generating significant revenue from selling humanoid robots directly today. But the actuators, vision systems, edge AI processors, and sensing hardware that go inside each unit are shipping right now. These businesses have existing revenue, existing margins, and growing exposure to a market that has barely started to scale. Morgan Stanley puts the full opportunity at roughly $5 trillion by 2050, with global adoption reaching around one billion units. The analysis from that research points specifically to the supplier layer, not the robot brands, as the more durable entry point.

Those projections could be wrong. In both directions.

What is harder to argue with is the structural driver underneath all of it. Labor shortages in developed economies are not cyclical. They are demographic. And the industries with the most acute shortages, manufacturing, logistics, elder care, construction, are exactly the sectors where the first two waves of humanoid deployment are being targeted. Companies that solve a structural problem, even partially, tend to get paid for a long time.

The risk is real and worth naming: many of the most technically advanced humanoid companies remain private. The public options are either expensive, speculative, or early. The concentration risk is significant at this stage of the cycle. But the picks-and-shovels layer, the companies selling into every robot OEM regardless of who wins, offers a cleaner entry point than chasing the headline names at peak enthusiasm.

The question was never whether humanoid robots would matter. It is who gets paid while everyone else is still debating the timeline.