EngineAI T800 Robot

How much does the EngineAI T800 cost?

EngineAI publishes a four-tier price list on its Chinese domestic marketplace storefront. As of 30 August 2026 the tiers are:

**Basic Edition — CNY 180,000 (about $26,800).** No dexterous hands. This is the machine behind every "$25,000 humanoid" headline.

**Eco Edition, open source — CNY 240,000 (about $35,700).** Adds dexterous hands and, per EngineAI, supports secondary development — meaning the buyer can build on top of the control stack rather than just running canned behaviours.

**Pro Edition — CNY 280,000 (about $41,700).** Adds the NVIDIA Jetson Thor compute module.

**Max Edition — CNY 360,000 (about $53,600).** Thor plus industrial-grade components.

A build slot is held with a CNY 5,000 deposit, roughly $740.

The dollar figures are conversions at 6.7209 CNY to the dollar, the rate on 28 August 2026, and they are ours rather than EngineAI's. The company does not quote in dollars because it does not sell in dollars. That is the single most important fact on this page and it survives every price comparison you will read: there is no US list price for a T800, because there is no US channel for a T800.

What the tiers tell you about the product is worth reading too. The gap between Basic and Eco — CNY 60,000, about $8,900 — is what dexterous hands and an open development path cost. The gap between Eco and Pro is the price of onboard AI compute. And the top tier's justification is "industrial-grade components", which is EngineAI conceding, quietly, that the cheaper tiers are not built to industrial duty. A CNY 180,000 humanoid is an astonishing thing to exist at all. It is not a CNY 360,000 humanoid with a discount.

Can anyone outside China actually buy a T800?

Not through an authorised route, and as of 30 August 2026 we could not find one.

EngineAI's English-language material has a buy path, and that path resolves to a Chinese domestic marketplace listing. There is no authorised international distributor, no published export price, no confirmed North American shipping date and no stated warranty terms for a unit outside mainland China. Several grey-market resellers will happily quote you a number. What they are selling is an import they have to arrange themselves, with no manufacturer support behind it, and usually at a markup that erases the entire reason you wanted a T800.

For a United States buyer there is a second, harder wall. On 28 July 2026 the FCC added foreign-produced advanced robotic devices to its Covered List, acting on a national security determination issued the day before. The rule is country-neutral in its text — a device counts as domestic only if it is built in the US with more than 65% domestic component cost, which catches German and Japanese machines as surely as Chinese ones — but the effect on the T800 is unambiguous. A covered device cannot obtain new FCC equipment authorization, and without authorization it cannot lawfully be imported, marketed or sold. The T800 has no US authorization on record and, having launched after the cutoff, cannot now obtain one absent a conditional approval from the Department of War.

What the rule does not do is confiscate anything. Models authorized before 28 July 2026 keep selling, and the FCC waived its own rules so those keep receiving firmware and security updates until at least 1 January 2029. If you already own a Chinese humanoid, nobody is coming for it. But a T800 ordered into the US today has, in the plainest terms, nowhere to land.

The Shenzhen police patrol — what really happened

In 2025 the T800 appeared on a Shenzhen street alongside uniformed officers, and the images went everywhere: a black, heavy-shouldered humanoid walking a beat, with a name borrowed from a killer robot in a 1984 film. "China deploys Terminator robocop" wrote itself.

What can be verified is that the appearance happened. What cannot be verified — because no police authority and no EngineAI release says it — is that the T800 has any operational policing role, any patrol schedule, any assigned duties or any autonomy in that context. It was a public demonstration in a country that runs a great many public robot demonstrations, and reading it as a procurement decision is exactly the mistake the imagery was designed to encourage.

The same discount applies to the combat footage. In late July 2026 the T800 took part in what organisers billed as the world's first freestyle humanoid robot fighting event, and in August a clip of one landing a head kick did the rounds again. Those events are real, they are genuinely hard on the hardware — coverage of the July event noted a machine continuing to fight after losing head-mounted sensors, which is a legitimate robustness signal — and they are also a marketing format. A humanoid that can throw a scripted spinning kick under a referee's supervision has told you something about its actuators and its balance controller. It has told you nothing about whether it can pick a box off a shelf for eight hours.

Is the T800 really in mass production?

More than most of its competitors, yes — and the claim is unusually specific, which is what makes it checkable in principle.

On 29 May 2026 EngineAI announced the launch of its intelligent manufacturing base at Honghualing, Shenzhen, spanning roughly 12,000 m², and said the first batch of T800s had come off the line with shipments beginning in early June. The company describes a closed-loop process running from incoming material inspection through component assembly test, final assembly, end-of-line test and after-sales maintenance; a stated cadence of one robot every fifteen minutes; 79 dimensional quality inspections and 46 working-condition simulation tests per unit before delivery; and a capability the company frames as ten-thousand-unit scalable delivery.

Here is the part to keep in mind. A line that can produce a robot every fifteen minutes is not the same as a line that does. "10,000-unit capability" is a capacity statement, not a shipment number, and EngineAI has published no delivered-unit count. The company filed confidentially for a Hong Kong IPO in June 2026, and a confidential filing keeps the revenue hidden — so as of 30 August 2026 there is no audited number that would tell you whether the factory is running at one an hour or one a week.

What is verifiable is that the building exists, the first batch left it, and paying Chinese customers have taken delivery. In a sector where most announced factories are renderings, that already puts EngineAI in a small group.

Who is EngineAI and can the company be trusted to still exist in 2029?

EngineAI Robotics was founded in Shenzhen in 2023 and is led by chief executive Zhao Tongyang, who according to reporting on the company's IPO filing previously founded the quadruped firm Dogotix and led XPENG's humanoid programme after the EV maker acquired it. That is an unusually operational background for a humanoid founder — most of this field comes out of research labs.

The funding record: a $200 million Series B in April 2026 valuing the company at $1.5 billion, led by Henan CICC Huirong Fund Management with Luxshare-ICT participating, taking total disclosed funding to at least $350 million. Luxshare is the interesting name there. It is a contract electronics manufacturer at genuine scale, and a manufacturer on the cap table of a company that just opened an assembly base is not a passive financial bet.

On 12 June 2026 Bloomberg reported that EngineAI had filed confidentially for a Hong Kong IPO, joining a queue of Chinese robotics firms heading for that market — Unitree's own listing in Shanghai on 19 August 2026 opened up more than 600% before sliding, and the "robot bubble" coverage that followed is now part of the context any of these listings trade into.

So: real money, real factory, real product, and a company barely three years old operating in the most crowded hardware market on earth at prices that leave very little margin. For a Chinese research buyer that risk profile is normal. For anyone outside China it compounds with the distribution problem — you would be buying an unsupported grey import from a young company you cannot reach, in a currency you do not hold.

T800 versus the other cheap full-size humanoids

The comparison everyone wants is against Unitree, and it is genuinely close.

Against the **Unitree G1**, the T800 is a different class of machine: the G1 is 1.32 m and about 35 kg, a desk-and-lab robot, while the T800 is adult-sized at 1.73 m and 75 kg. The G1's advantage is overwhelming on the things that are not spec sheet entries — a large public code ecosystem, thousands of units in the field, English documentation, and an actual international sales operation. Its disadvantage as of 29 August 2026 is a pair of disclosed vulnerabilities in the G1 EDU that allow root-level remote code execution, one of them over Bluetooth. That is a real finding and it deserves to be weighed, though it is also a reminder that the G1 gets security research done on it precisely because so many people own one. Nobody has published a comparable audit of a T800.

Against the **Unitree H1 and H2**, the T800 is roughly size-comparable and considerably cheaper.

Against Western industrial humanoids — Digit, Apollo, Figure — the comparison is not really a comparison. Those machines come with a fleet-management cloud, a safety certification argument, a service organisation and a customer support contract, and they are priced at five to ten times the T800 accordingly. Agility's Digit is modelled at about $8,500 per robot per month. You could buy three Max-edition T800s outright for one year of that. What you would not get is anyone to call.

The honest framing: the T800 is the cheapest credible route to a full-size, high-torque, walking humanoid in 2026 — if you are in China, have engineers who read Chinese documentation, and are buying a platform rather than a solution.

What can the T800 actually do today?

Verified by video and by manufacturer demonstration: it walks and runs, it performs martial-arts sequences with balance recovery, it takes impacts and stays upright, and it has been shown handling objects with the optional dexterous hands. The solid-state battery is a genuine engineering distinction rather than a marketing one — very few humanoids ship with one, and if the four-to-five hour claim holds up in the field it addresses the single most limiting property of this whole product category.

Not verified, and important: any autonomous task performance. EngineAI's demonstrations are choreographed or teleoperated, or the company has not said which, and there is no published deployment where a T800 executes useful work unsupervised. There is no published payload rating. There is no published mean-time-between-failures, no uptime figure, no fleet count, and no third-party integration case study. The teleoperation support field in our own database reads "unknown" because EngineAI has not documented it.

The T800 is at the stage where the hardware is impressively ahead of the demonstrated autonomy. That is not a criticism unique to EngineAI — it describes almost every humanoid in 2026 — but at this price point it is easy to read the number as "a worker for $27,000" when what is actually on offer is a very capable body waiting for someone to write the software.

What would make us upgrade this entry?

Four things, and none of them have happened as of 30 August 2026.

An **audited shipment number**. The Hong Kong listing, if it goes effective, would force disclosure of revenue and unit volumes. That would turn the "10,000-unit capability" statement into something a buyer could act on.

An **authorised international distributor** with a published price, warranty and spares path. Right now the entire non-Chinese market is served by grey imports.

A **rated payload and a duty cycle**. Two numbers, both trivially publishable, both absent. Their absence is the clearest signal of where the T800 sits on the research-versus-industrial line.

An **independent security review**. The Unitree G1 vulnerabilities disclosed on 29 August 2026 are a warning about the whole category, not about one vendor. A networked 75 kg machine with LiDAR, cameras and microphones is a serious device to put on a corporate network, and no public audit of the T800's firmware or update channel exists.