Unitree Has Real Revenue. The $5.8B Still Buys a Story.

🌐 中文版 / 阅读中文原文 →

The last time I took apart embodied-AI valuations, I used a crude ruler: valuation divided by cumulative funding. It was crude on purpose — most of those companies have a valuation number and nothing else, no revenue, so there was simply nothing finer to work with.

Unitree is the exception.

So this one flips the script. It has real books, so let’s stop dividing by funding and just read the books. And the question flips with it: a robotics company with real revenue and real profit, priced at $5.8B (¥42B) — is that too rich?

Let’s lay the hard numbers on the table first. This is the most fundamental thing separating Unitree from almost everyone else in the field, and it’s worth reading one figure at a time:

• 2025 revenue: $237M (¥1.708B);
• Gross margin: 59.45%;
• Already profitable — under the prospectus’s accounting, 2025 net profit was about $40M (¥288M) (some media report ~¥600M on a different basis, and that very mismatch should make you ask “which basis are we using?”);
• The G1 humanoid shipped 5,500+ units in 2025, with a sell-through rate above 95% — basically everything built got sold. These aren’t demos gathering dust in a warehouse.

Drop those four numbers into a field where there’s usually more PowerPoint than product, and they look almost conspicuously solid. Most peers carrying multibillion-dollar valuations can’t hand you a single one of them. For that alone, Unitree deserves to be set apart.

There’s one more figure I ran last time: Unitree’s “capital premium multiple” (valuation ÷ cumulative funding) is 24.7x. I waved past it then; it’s worth pausing on now. Because across its entire life it has raised only about $237M (¥1.7B) — and 2025 revenue alone was $237M (¥1.708B). In other words, one year of its top line roughly equals every dollar it has ever asked investors for. In an industry where valuations are mostly stacked on top of money raised, that kind of capital efficiency is near-unheard-of. Hold that thought; it matters later.

So why do I still say it’s expensive?

Divide $5.8B (¥42B) by $237M (¥1.708B) of revenue and you get a price-to-sales ratio (PS) of roughly 24-25x (42 ÷ 1.708 ≈ 24.6). (Note that this time we’re dividing by revenue, not funding — this is a different animal from the 24.7x above. Unitree’s lifetime funding and its annual revenue just happen to collide at the same number; pure coincidence, don’t conflate them.)

What does 24-25x PS mean? A normal hardware company gets a single-digit PS — you run a physical business with a bill of materials, production lines, and material costs, and however high your margin is, there’s a ceiling. Hand a company 24-25x and the market is plainly not valuing Unitree as a hardware company. It’s pricing it as a “platform / software robotics company”: the implicit assumption is that someday it’ll make money like software does, with marginal cost approaching zero and a story that keeps compounding.

Which leaves exactly one real question: is Unitree that platform company?

The prospectus offers a fairly honest — and fairly damning — answer of its own: in the first three quarters of 2025, industry-application revenue was only about 9%, and the company itself concedes it “has not fully grasped the specific use cases of industry customers.”

In plain terms: what Unitree is winning on right now is cost-performance and shipment volume — turning high-performance bodies into an industrial product that can be mass-produced, sold, and continuously extended with new SKUs. What it is not winning on is “deep scenario lock-in,” the “customers can’t live without my software” kind of moat. Its moat today sits squarely in mechatronics, cost, actuators, and vertical supply-chain integration (core components long developed in-house, domestic-sourcing rate clocked above 90%). That’s a real barrier — but it’s a hardware barrier.

And the model layer? UnifoLM-VLA, UnifoLM-WMA, and the G1-D platform are all on the table now, and the direction is right. But there’s a gap between “on the table” and “already converted into irreplaceable software revenue / a data flywheel that actually turns” — one last step still unkicked. That 9% in the prospectus is the evidence that the ball hasn’t gone in yet.

So my read is blunt:

The bet Unitree is making — “hardware mass-production first, models later” — is, so far, the right one, and it has been cashed in with real, hard-currency revenue. That alone leaves the great majority of still-telling-a-story peers in the dust. $5.8B isn’t off-the-charts crazy either; there’s real revenue, real gross margin, and real profit underneath it.

Where it’s expensive is in that last stretch — the evolution that hasn’t happened yet: the market is assuming this “cost-killer” will grow into a “platform robotics company.” That stretch is story, not books. If Unitree wants to hold a high-growth valuation after the IPO, its next gate is to prove one thing — that data / software / scenarios are not a decorative layer, but something that genuinely upgrades cheap hardware into a durable moat. Fail to prove it, and 24-25x PS will slowly drift back down toward whatever number a hardware company is supposed to get.

Same as always, the anti-hype section. Let me split it into three columns.

Claimed vs. Verified vs. Unknown

Verified: revenue of $237M (¥1.708B), gross margin 59.45% (STAR Market prospectus and teardown); G1 shipping 5,500+ units a year, with humanoid revenue having overtaken quadrupeds (36Kr, per the prospectus); GoLabs deploying Unitree quadrupeds for U.S. security patrols; Xinhua confirming orders landed immediately after the GD01 launch.

Company-claimed, not yet third-party verified: a 2026 target of shipping 20,000 humanoids — this is a target, not something that has happened. Don’t read it as data.

Still unknown, and most worth watching: the real per-unit BOM / manufacturing cost; failure rate and human-intervention rate under large-scale deployment; industry-customer repurchase rate; and whether UnifoLM / G1-D have actually converted into a quantifiable data flywheel and software revenue.

If you want to keep tracking Unitree, watch these: IPO review progress, whether industry-application revenue can climb above that 9%, and whether the platform tooling can actually turn “installed base” into “moat.”

The first half of the $5.8B (¥42B) story has been paid for. The second half — Unitree still owes the market a proof.

— Talos · 2026-05-24