Humanoid Robots: AI Gets Hands. Buy the Supply ChainRockwell Automation, Inc.BATS:ROKEXCAVO There is a post making the rounds right now. About the SOX index. The semiconductor index. It draws a comparison between the dot-com crash and what is happening now. Visually, I will be honest, it looks convincing. The pattern is similar. The angle of the decline is similar. Everything lines up. But there is one very important detail that changes the entire picture - and that is where I am starting this post. First, though - what is happening in the physical world. AI completed its first season: text, images, code. All of it lived inside the data center - clean, elegant, no calluses. The second season is starting right now, and it is messy. AI is crawling out of the server room into the physical world: hauling boxes, tightening bolts, working on warehouse floors and assembly lines. The brain learned to think - now it is getting hands. This is where it gets interesting for anyone who does not buy fairy tales. First, SOX - and why the dot-com comparison is a trap The SOX-versus-dot-com post is spreading across every channel. I have seen it. I understand why it grabs people - the pattern is clean, the comparison is visual, and fear sells better than boredom. Here is the detail that post leaves out. The semiconductor boom of 2000 was built on demand that did not exist. Companies were buying servers and networking gear for traffic that had not arrived yet. Corporate IT budgets inflated on the expectation of an internet revolution that would come - but ten years out, not tomorrow. SOX ran on air because the orders were real but the end-user revenue was not. Today is different. NVDA is trading at $197.58. That level is not floating on air. Underneath it sit real contracts with Microsoft, Google, Amazon, and Meta for GPU deliveries. Real data centers being built physically. Real revenue you can touch in quarterly reports. BlackRock holds $5 billion in tokenized Treasuries on Ethereum - that is AUM, not a slide deck. The spot Bitcoin ETF gathered $60 billion in its first year - that is money, not a concept. The SOX pattern may look similar. But the foundation beneath it is different. In 2000 the market was paying for a dream. Now it is paying for infrastructure that is already running and already generating cash. That does not mean a correction cannot happen. A correction can come - and the humanoid theme will ride that wave along with everything else. But a direct dot-com analogy is a framing error. Be careful with clean-looking patterns that sell fear without digging into the details. The idea, without the rose-tinted version The fairy tale is free: "tomorrow every home will have a robot butler that folds your socks." Skip that version. By the time a metal humanoid actually folds your socks, the obvious tickers will be priced like a piece of a Boeing. At the stage where everything is obvious, the money is already gone. The real trade is simpler and harder. The market already pays for AI's brain - generously, look at NVDA. Next it will start paying for AI that does physical work. The play is the supply chain underneath the robot, before the market fully reprices the theme - not the finished robot on stage. McKinsey is direct about the diagnosis: the bottleneck is the maturity and scale of the component base, not AI capability. One humanoid costs between $30,000 and $150,000 today. Until that number compresses, robot OEMs earn almost nothing. The ones earning are the companies selling them reducers, servo drives, and sensors. Why now One. AI has finally given the robot a brain. Before, it was a metal device with pre-programmed movements. Now it sees, understands, and adapts. Two. The cost curve has broken downward. Global humanoid production in 2025 crossed 20,000 units versus fewer than 2,000 in 2024 - ten times growth in a year. Venture capital into robotics grew more than threefold from 2023 to 2025, reaching $40.7 billion per year. China launched a state fund of $138 billion targeting AI, robotics, and adjacent high-tech. Three. Industrial and auto names trade like cyclical junk. They get valued on auto-cycle and industrial-downturn multiples, as if nothing is happening underneath. Underneath, a new layer of demand may be forming - sensors, power semiconductors, microcontrollers, actuators, reducers, edge computing, factory automation. Why the human form actually makes sense A specialized robot is always better at one specific task. True. But the entire world is already built for humans - door handles, staircases, shelves, boxes, vehicles, conveyors, warehouse aisles. A humanoid-shaped robot means you do not have to rebuild the world. You place a metal trainee where a human stood yesterday. Business hates revolutions that require tearing down a facility and rebuilding it. Business loves anything that slots into the existing process while headcount stays the same and EBITDA stays happy. The most underrated bridge - teleoperators Robots do not need to be fully autonomous on day one. If one operator somewhere overseas supervises four or five robots in the US and intervenes only when the AI gets stuck - the economics already work. Think of the call centers of the 2000s, except the headset now controls a spine instead of a phone call. Dirty, practical, corporate automation. The market tends to respect that kind of thing earlier than it becomes pretty. Morgan Stanley has already mapped the space: 73% of confirmed humanoid companies and 77% of integrators are in Asia; 56% and 45% respectively are in China. That is about industrial depth, not the elegance of the model. Where the money is - by name Play a basket, not a single "winner." Most pure-play beneficiaries are still private - Figure, Unitree, Apptronik, Agility. On the public market, you get exposure through the names selling the brain, nervous system, muscles, joints, and factory infrastructure. NVDA - $197.58. Brain and simulation: Jetson Thor, Isaac, GR00T, edge computing. Downside: obvious and fully priced in, this has been public for a long time. And -1.3% on the day as I write - the crowd is already here. That is exactly why I opened with SOX: when people compare NVDA to the dot-com bubble, the key point is that real contracts sit underneath it, not air. Buying into the hype is a separate conversation. TXN, MCHP, ON. Analog chips, power electronics, microcontrollers, motor drivers, sensors - the boring silicon without which the terminator falls flat on its face. Downside: some of them still trade as cyclical semiconductors, not as robotics beneficiaries. That is simultaneously the risk and the entry point. RRX (Regal Rexnord). Motion: motors, reducers, actuators, industrial components. The robot needs joints and gear reduction, not just smart talk in its head. Downside: tied to the industrial cycle - if it rolls over again, it pulls the stock with it. TER, ROK. Teradyne (Universal Robots, automation) and Rockwell (factory automation). A more mature way to play the theme. Downside: these are cyclical industrials, entry matters - buying into a rally is not the move. GXO, AMZN - $241.70. The warehouse is the first real proving ground: boxes do not complain, aisles are standardized, tasks repeat, and turnover motivates management. GXO is already testing humanoids; Amazon is essentially an entire country of warehouse automation. Downside: for AMZN, robotics is a drop in the revenue ocean - it does not qualify as a pure-play. TSLA - $425.30. Optimus optionality. The biggest call option on the theme and the most religion-driven asset in it. Downside: this is not a clean robot trade - you get EV, autonomy, Musk, China, and margin all in one. Believers will buy at any level, and believers also blow up spectacularly. SYM (Symbotic). High-beta warehouse robotics. Downside: swings hard in both directions, thin book - not for the faint-hearted. ISRG (Intuitive Surgical). Not a humanoid, but proof that robotics can be a massive, profitable, and defensible business. Its connection to the humanoid theme is indirect - a reference point rather than a direct bet. What could break all of this One. Robots stay demo videos, not economics. Figure runs clips of a robot sorting parcels for 200 hours - impressive, but a clip is not a P&L. Two. The auto and industrial cycle rolls over again and drowns the cyclical half of the basket. Three. China crushes prices faster than public suppliers can earn anything. 45% of integrators in China is not an abstraction. Four. TSLA and private robots inflate the hype while the listed supply chain receives zero orders. Five. The market prices in the theme too early and marinates investors for years. The forecast range is enormous: some estimates put the market at $38 billion by 2035, others at $205 billion; Goldman sees revenue above $150 billion by 2035; Morgan Stanley draws a TAM of nearly $5 trillion by 2050. When the spread is that wide, nobody actually knows. Drop the mantra "robots are inevitable, buy everything." Inevitable - maybe. But your account can die long before the inevitable reaches a P&L. And that is exactly where the SOX story closes the loop. That is how people lost money in 2000 - the theme was right. The internet was the right theme. They lost because they bought the right theme at the wrong moment, at the wrong price, with no understanding of how many years stood between them and real revenue. The humanoid theme is right. The question is at what price and at what moment you board this train. My plan Playing a basket, not a single stock. Steady core: NVDA, TXN, MCHP, ON, ROK, TER. More asymmetry: RRX, GXO, SYM. Optionality and conviction: TSLA. Proven robotics economics: ISRG. The ideal setup is to avoid buying everything at once after a compelling post. Wait for the cyclical industrials and analog semiconductors to offer an entry while the market still prices them as old economy. As I have written about oil - I trade the boundaries of the range, not the crowd's emotions. The same logic applies here: the market front-runs capex announcements, pilots, and partnerships, then overprices suppliers before any real revenue appears. The job is to get into the car while the ticket is still selling at scrap-metal prices. The market is a redistribution mechanism. It takes money from whoever bought the theme at peak hype - whether they were scared by a pretty SOX chart or not scared enough. It gives that money to whoever bought the robot's spine and nervous system while they were still boring. Bottom line A scary SOX chart is making the rounds - and it has a detail that changes everything. The AI trade is moving out of the data center into the physical world. First the market paid for the brain. Next it pays for the hands, joints, sensors, motors, batteries, warehouses, and factories. And as always, the biggest money goes to those selling the robot its spine, its nervous system, and its factory badge - not to those selling the pretty robot on stage. Regards, EXCAVO