3 min read
Here is the state of the robot uprising in 2026, and it looks nothing like the keynote. The company shipping the most humanoid robots on Earth is not Tesla, not some Silicon Valley darling, but Unitree, a Chinese firm selling machines at roughly a tenth of the price everyone else charges. The robots actually doing paid work in factories are made by Figure and Agility. And Tesla’s Optimus, the one with the glossy reveals and the dancing videos? As of the middle of this month, it had not started production at all.
The one flooding the market: Unitree
Unitree is the story nobody in California wants to tell. The Hangzhou company shipped around 5,500 humanoid units across its range in 2025 and is targeting somewhere between 10,000 and 20,000 this year, at prices that undercut Western rivals by an order of magnitude. It is doing to robots exactly what Chinese manufacturers did to solar panels, drones and electric cars: not necessarily inventing the flashiest version, but building it cheaper, faster and at volume until everyone else is priced out. Its Q1 profit halved, which tells you it is buying market share with margin, the classic land-grab. Underestimate that at your peril.
The ones that actually work: Figure and Agility
If you want to see humanoids doing real jobs today, look at Figure and Agility, not Tesla. Figure’s 03 robots are in expanding paid deployments, including sequencing tasks at BMW’s Spartanburg plant, and between them Figure and Agility hold the strongest verified deployment records in the industry. Boston Dynamics, meanwhile, is putting its electric Atlas through pilots with Hyundai and Google DeepMind. This is the unglamorous truth of the field: the robots earning their keep are doing boring, repetitive, specific factory work, not folding your laundry or minding your kids.
The one that keeps slipping: Optimus
Which brings us to Tesla. Optimus V3 is a genuinely impressive spec sheet: 22 degrees of freedom in each hand, Tesla’s AI5 chip, Grok bolted in for voice. It is also, as of mid-July, not in production. The guidance points to late July or August for low-volume manufacturing at Fremont, focused on factory tasks. Which would be fine, except this is the robot Elon Musk has spent years telling investors will be bigger than the car business, will be worth trillions, and is always, somehow, about a year out. There is a pattern with Musk timelines, and it is not that they arrive early. While Optimus does its umpteenth “coming soon”, Unitree is shipping thousands and Figure is billing clients.
What to take from it
The robot revolution is real. It is just less cinematic, and more Chinese, than the hype promised. The winners so far are the firm making them cheap enough to flood the market and the firms boring enough to actually deploy them for money. The company with the best show and the most famous CEO is, for now, still rehearsing. If you want to know who wins the humanoid decade, watch shipment numbers and signed contracts, not reveal-event applause. The machines that matter are the ones already clocking in.
The AI tool stack actually worth paying for
One email a week. The tools, models and moves that matter, minus the hype and with the horseshit filter set to maximum. Free.
Did you know: “degrees of freedom”, the spec robotics firms love to quote, simply means the number of independent ways a joint or limb can move. A human hand has around 27. Getting a robotic hand up into the twenties is genuinely hard, which is why hand dexterity, not walking, is the real battleground in humanoid robotics right now.