The Humanoid Robot Race: XPeng vs Tesla, Who's Ahead and Who Could Profit

FRONTIER SIGNAL | DEEP DIVE | ROBOTICS | 7 MIN READ | TUESDAY, 10 NOVEMBER 2026

Both carmakers say they now build humanoids on production lines, but neither gives a number, while Chinese specialists ship thousands. What's real, who's ahead, and the listed and private companies to watch.

On 8 September, in a factory in Guangzhou, a humanoid robot called IRON had its last part fitted, stood up and walked off the production line. That is how its maker, the Chinese electric-car company XPeng, tells it, adding that more than 80% of the line's core processes are automated.

Two months on, the clip is everywhere again. US Google searches for "XPeng humanoid robot" rose more than 47,000% in the week to 9 November, and Tesla says its own robot, Optimus, is now in "early production" too. You could be forgiven for thinking the robots have arrived.

But neither company will answer the simplest question of all: how many have you actually built? Meanwhile, Chinese specialists you may never have heard of are already shipping humanoids by the thousand. Here is what XPeng's line really proves, who is genuinely ahead, and which companies could profit.

In late August, XPeng's robotics unit raised about $900m from investors including IDG Capital, Tencent and Alibaba, at a $6.3bn valuation, according to the company's Hong Kong filing. Two weeks later came the walk-off. XPeng says IRON has 76 "degrees of freedom", the separate movements a robot can make, with 21 in each hand, and that it runs on three of XPeng's own Turing AI chips.

Since then, the timetable has filled in. In Paris on 11 October, chief executive He Xiaopeng promised mass production in late 2026, first deliveries in the first half of 2027 and a European launch after that. At a Shanghai trade fair on 8 November, XPeng said IRON will start work in 2027 as a guide, a shopping assistant and a security patrol robot, according to Chinese media reports.

Dates are in 2026 unless stated. Future dates are XPeng's own targets.

Getting a brand-new product onto a production line is hard, and XPeng deserves credit for it. But the first unit off any new line is mostly a photo opportunity. Frontier Signal's founder spent 18 years building electric-vehicle, renewable energy and green hydrogen businesses in India, and the number that mattered was never the first unit. It was the hundredth unit a day, built to the same standard, at a cost the business could live with.

That is exactly what XPeng hasn't shared. As the trade site Humanoid Guide put it, the company gave "no rated capacity, cycle time, yield data or production cost", and "one completed robot also does not demonstrate volume output." Nobody outside XPeng has confirmed that the robot walked off unaided, either.

What XPeng has offered instead is ambition. In August, He Xiaopeng told investors that monthly capacity "can rapidly ramp up to several thousand units" next year. In Frontier Signal's illustrative scenario, 2,000 to 5,000 a month for a year would be 24,000 to 60,000 robots, or two to four and a half times the 13,318 humanoids Omdia counted shipping worldwide in 2025. That is a scenario, not a forecast, but it shows how big the claim is.

Judge by announcements, and XPeng and Tesla look neck and neck. Tesla's Fremont line is in the same position as XPeng's: production has started, but Tesla hasn't said how many robots it has made. Elon Musk says outside customers will get Optimus "around the end of next year", and admits the hands are still "the main issue", Electrek reported.

Judge by robots actually delivered, and the specialists lead. Unitree, which listed in Shanghai in August, says most of its 2025 humanoids went to universities and government-backed projects. Its private rival AgiBot says it built its 20,000th robot in September. And in the US, Agility's Digit robots already move goods for paying customers such as GXO, Schaeffler and Toyota.

Table 1. Company disclosures and named reports, as of 10 November 2026.

So why take the carmakers seriously? Because underneath, a humanoid is a box of car-like parts. The small motors, gearboxes and screws in its joints make up more than half of its hardware bill, according to NiMotion, a Chinese company that makes them.

That is why XPeng's 85% supply-chain overlap matters. A carmaker already buys millions of precise parts a year, so it can ask the same suppliers for robot parts at car-industry volumes and prices. XPeng also uses its own chips in both its cars and IRON. Most Western rivals buy their robot "brains" from Nvidia instead, whose Jetson Thor computer counts Figure, Agility and Boston Dynamics among its early users.

The race has a geography, too. China has a fast route to the stock market: Unitree's listing was approved just 104 days after it applied, and AgiBot and Galbot are preparing to list in Hong Kong. America has the deepest pockets. When Figure AI, a California start-up whose robots have been deployed at BMW, last raised money in September 2025, investors valued it at $39bn. That is about six times the value put on XPeng's whole robot business in August.

Politics could reshape both. In July, the US Federal Communications Commission banned imports of new foreign-made humanoid robots, the Associated Press reported, and so far Europe is the only overseas market XPeng has named. Beijing has its own worries: last November, its top economic planner warned of a possible bubble, with "remarkably similar robots from more than 150 companies."

Most humanoid makers are still private. Below, the ones you can buy shares in, and the ones you can't (yet). Figures are the latest public disclosures with dates; private valuations come from funding rounds, not markets.

Figures as of 10 November 2026. Private valuations are post-money from the latest announced round and are not market prices.

This is general information, not investment advice.

The case for caution is that none of this may matter for years. Neither carmaker has shown a robot doing paid work for an outside customer, and XPeng's own plan starts IRON in showrooms rather than on factory floors. Investors have already had one reality check: by 9 September, Unitree's shares were about 53% below their opening price, and one fund manager told Reuters that "commercial orders are not yet in sight."

Much of what you have just read comes from the companies themselves. XPeng's automation figure and the robot's unaided walk are unaudited claims, and even the industry's counts disagree: Unitree's prospectus gives it 32.4% of the 2025 market, while Omdia gives AgiBot about 39%. Some private valuations rest on talks, not deals. And timelines slip: Musk once said outside sales of Optimus would start in 2026.

How many IRONs will XPeng actually build in 2027, and what will they be doing?

XPeng's results on 16 November are the first chance to find out. Listen for a monthly output figure, a price and a customer outside XPeng's own stores. Until we hear those, the production line remains a promise.

Not yet. XPeng commissioned an IRON production line on 8 September 2026 and targets mass production by the end of 2026, with deliveries from the first half of 2027. It has not disclosed any output.

XPeng hasn't priced IRON, and Tesla hasn't priced Optimus. XPeng's chief executive said humanoid robots sell for about 2.5 to 3 times the cost of their parts.

Pure plays include Unitree (Shanghai: 688836), UBTech (Hong Kong: 9880) and Agility Robotics (Nasdaq: AGLT). Carmakers with humanoid programmes include XPeng, Tesla, Hyundai, which owns Boston Dynamics, and Xiaomi.

Yes. Unitree has traded on Shanghai's STAR Market under the code 688836 since 19 August 2026.

On disclosed facts, neither. Both say they build on production lines, and neither gives a unit count. XPeng targets deliveries in the first half of 2027; Tesla expects outside deliveries around the end of 2027. Chinese rivals Unitree and AgiBot ship far more.

Know someone tracking robots? Forward this, they can subscribe free at frontiersignal.ai/subscribe

Company figures are as disclosed by XPeng, Tesla, Unitree, UBTech, Agility and others and have not been independently verified. Private valuations are from announced funding rounds or, where stated, reported talks. Frontier Signal's scenario uses XPeng's August guidance, Unitree's prospectus and Omdia's 2025 shipment count. Search-interest data: Google Trends, 2–9 November 2026. Disclosure: Frontier Signal's founder owns shares in Nvidia and Alphabet, and has no position in the other companies named. This is general information, not investment advice.