Can robots keep the AI boom alive?

Hello friends, and welcome to The Unsophisticated Investor! Brought to you by Scott & Rob, the founders of Shuttle!

If you want to invest alongside the VC funds who've backed breakout companies like Revolut, Asana, JustEat, Bolt, Lets Get Checked, Loom, Runna, Charlotte Tilbury, Deel, Aircall, AngelList, Carta, TransferWise and many more, regardless of your knowledge, network or net worth, then sign up using the link below.

Now, let’s get into it 👇

Flying between Belfast and London Gatwick recently, I saw BellaBot, the cat-faced robot that ferries trays between tables at Belfast International Airport. It is the sort of thing you would expect in Tokyo rather than in an airport in Northern Ireland, and it got me wondering how quickly machines like it might become an ordinary sight.

As the AI boom is now moving towards sustaining itself on new application layers for AI to justify vast investment requirements, one of the most advanced fields is that of humanoid robotics, touted to do the work people no longer want to. One of those is being built back in London. A company called Humanoid, founded in 2024, is developing a human-sized machine designed to walk into a warehouse and move boxes.

Its founder, Artem Sokolov, spent the previous decade turning a family jewellery business into a billion-dollar company, watched what repetitive manual work did to the people doing it, and has since put roughly $30 million of his own money into the venture. He is reportedly in talks to raise around $200 million more in a Series A. The robot runs on an Nvidia-based software stack, with German manufacturing conglomerate Bosch agreeing to manufacture it. Sokolov has told reporters the order book already points to billions in future recurring revenue.

Humanoid is one of a small number of well-backed European firms vying to battle it out in a corner of the AI-race against the US and Chinese incumbents. It is also one more argument for ever-greater compute, the kind that justifies the multi-billion-dollar data centres going up around the world. The signals of genuine demand look real enough, but until they show up as sustained industrial orders rather than research-lab pilots, a degree of scepticism is the fair position. 

How VC money is being placed

Robotics has become one of the largest single destinations for venture capital, and humanoids are the segment of it where investors have decided lives will be transformed. This spans from factories and warehouses to fixing social care for an ageing population. The level of conviction by investors can be seen among the valuations. Figure, the American frontrunner, raised more than a billion dollars last September at a $39 billion valuation. That figure is larger than Goldman Sachs's estimate for the entire global humanoid market in 2035, and Figure has shipped robots in the low hundreds. Physical Intelligence, which builds the software that runs inside robots rather than the robots themselves, was valued at $5.6 billion in November and is reportedly raising again at more than $11 billion, without a commercial product to its name. Skild AI, doing much the same thing, is worth over $14 billion on around $30 million of revenue.

The money is also concentrated. A handful of companies have taken most of the capital in the category, with everyone else sharing a thin remainder. None of this is unusual for a young industry. New sectors are priced on a story long before there are real revenues to price them on. It does mean the numbers are based on assumptions of the future, and it’s worth looking at what those assumptions are.

Why anyone is building these at all

The demand story is mostly about demographics, and most of it holds up. Ageing populations and shrinking workforces are a genuine pull. One widely cited estimate puts the global labour shortfall at around 85 million workers by 2030. A Capgemini survey found roughly three-quarters of manufacturing executives naming labour shortages as their main reason for looking at humanoids at all. Japan, furthest into its demographic decline, where the population is expected to halve by 2100, already has 600,000 unfilled industrial jobs.

The problem is that the robots are not yet as good as a trained, experienced person. They perform well in controlled demonstrations and badly once they are let loose in a real, unstructured setting. The clearest indicator is who is actually buying them. Unitree, which ships more humanoids than almost anyone, has disclosed that only about 9% of its humanoid revenue comes from industrial customers; nearly three-quarters comes from universities and research labs. The flagship workforce robots are, for now, mostly bought by people who want to study their application. Battery life runs to a few hours, and per-unit prices stretch from the tens of thousands into six figures. By the middle of this year forecast humanoid supply was running ahead of humanoid demand.

The whole case rests on a shortage of workers, which makes the timing of one recent story seem at odds with this. As I was researching this week's newsletter, Volkswagen was reported to be preparing to cut up to 100,000 jobs, close to one in six of its global workforce, and to close four German plants. VW's problem is not too few workers but too much capacity and lost ground to cheaper Chinese carmakers, whereas the robot-makers are chasing the basic, unfilled roles that demographics are quietly draining away. The two can be true at once. But it is worth sitting with the fact that in the same industry, and the same country, where humanoids are being trialled, the near-term reality is a workforce being cut, and the cause is Chinese competition rather than any robot. 

Where are humanoids being deployed

Where humanoids are earning their keep, it is in structured, repetitive industrial settings. Figure ran an eleven-month pilot on a BMW line in South Carolina, though BMW has since picked a different machine for its European plants. Apptronik's Apollo has gone into Mercedes-Benz. Agility's Digit moves totes for Amazon and, more recently, inside Toyota. The clearest service example is Japan Airlines, which began a three-year pilot at Tokyo's Haneda airport in May, using low-cost Chinese robots for ground handling. Almost everything beyond that, the robot folding laundry or minding an elderly parent, is still at the earliest stage. Barclays, which is bullish, sets out the order plainly: manufacturing, logistics and warehousing this decade, then consumer and household work, elder care included, some time in the 2030s.

The test for whether a humanoid is the right tool is narrower than the marketing suggests. It only makes sense where a job means moving around a mixed, human-shaped space and handling objects that keep changing. For fixed, high-volume work, a bolted-down arm or a wheeled cobot is cheaper and more reliable, and will probably stay that way.

How it could play out in China, US and Europe

The three main players are building for different ends. China's game is cost and volume: it accounted for about 85% of humanoid installations last year, builds at roughly half the Western price, and is backed by state money on a scale no one else can match. The logic is industrial self-sufficiency. An ageing country facing a projected 37-million-worker shortfall, with a $5 trillion manufacturing base to defend, has chosen to automate rather than import labour, and Barclays reckons it could have as many as 24 million humanoids at work by 2035. As the binding constraint shifts from clever software towards raw inputs, energy, actuators, critical minerals, China's grip on those is what will keep it ahead of its competitors.

America's advantage sits elsewhere, in the frontier models and the corporate pilots. Europe's position is more awkward. The parts inside these robots are often European, the actuators, sensors and industrial know-how of Bosch, Schaeffler and Siemens, and the continent has two credible integrators in Germany's NEURA Robotics and Britain's Humanoid. NEURA closed a Series C of up to $1.4 billion in June at a reported $7 billion valuation, led, by the stablecoin issuer Tether, with Nvidia, Amazon, Qualcomm, Bosch, Schaeffler and the European Investment Bank alongside.

The awkwardness is that making the parts is not the same as keeping the profit. Both of Europe's flag-carriers run on Nvidia's chips, and both face the same Chinese cost pressure now reshaping European carmaking. The plausible failure mode is a rerun of the car industry: Europe keeps the engineering while the demand and the margin migrate to American software and Chinese assembly. The one card Europe holds is not technical but political. Security concerns about Chinese-made humanoids on Western factory floors are likely to bring procurement rules and guardrails, and the EIB has shown it will fund the alternative. If Europe takes a real share here, it is more likely to be because it protected and bought its way into a market than because it out-built anyone.

Where is the money flowing to?

In every earlier computing wave the hardware commoditised and the money settled somewhere less visible: in the platform, and in whoever sold the hardest component everybody needed. Physical AI is lining up the same way. The body is the capital-heavy layer, hard to scale and, as China keeps demonstrating, cheap enough to copy that its price is already sliding towards commodity levels. The brain, the general model that lets a robot follow an instruction and improvise around it, is where firms like Physical Intelligence and Skild believe the durable margin sits, which is why investors are writing billion-dollar cheques to companies that build no robot at all. And then there is Nvidia, positioned to take a cut whoever wins. Its chips sit inside nearly every Western humanoid, and its GR00T models and Isaac and Cosmos software increasingly sit on top. At CES, Jensen Huang more or less said it himself: the humanoid business is running on the AI factories Nvidia is building anyway.

What’s the size of the prize?

The striking thing is how little agreement there is about the size of the market. The credible forecasts do not differ at the margins but differ by orders of magnitude. Yole Group sees a humanoid market of perhaps $51 billion by 2035. Goldman Sachs puts 2035 it at around $38 billion; Barclays' base case is $40 billion, its bull case $200 billion. Morgan Stanley, looking further out, has floated something near $5 trillion by 2050. These estimates describe materially different futures, and the gap between them runs from a decent mid-sized industry to one of the largest on earth.

That spread is the most honest thing on offer. Remember that McKinsey, as respected as it is in the research world, valued the metaverse at up to $5 trillion by 2030 as recently as 2022. Meta has all but effectively shelved the metaverse. When serious analysts scatter from tens of billions to several trillion, the indicators are that this is still an early-stage sector based on assumptions of uptake, that if they materialise, could see major winners and losers. If the demographic pull is real and the machines become good enough, then there’s a high chance we’ll see much improved versions of BellaBot and more life-like versions in the decades to come. Either way, if the AI-boom and bubble continues or bursts, the use case and arguments for humanoid robots will likely remain as a reason for selling and applying AI long into the future.   

What we’ve been working on at Shuttle

• Invited onto the FCA's AI Lab Supercharged Academy 📚 

• Updated and deployed a new compliance tool for advertising 🛠️

• Putting the final touches on the next iteration of our product before QA 🧑‍💻

• Finalising the latest opportunity for our next Drop 🤝

The Unsophisticated Investor is brought to you by Scott & Rob, the founders of Shuttle. We’re both sick of private markets being a playground exclusive to the ultra-wealthy so we started a company to challenge the status-quo. Shuttle’s singular focus is to unlock private markets for Millennial and Gen Z tech professionals and help them build wealth through the highest performing private market opportunities.

Scott & Rob
Shuttle Co-Founders