Humanoid Funding vs. Units Delivered: The Money Went In, Did Robots Come Out?

The four humanoid companies in the table below have announced roughly $2.7 billion in capital between September 2025 and July 2026. That much is easy to verify — funding rounds come with press releases, and every one of them was covered.

What came out the other side is much harder to establish. Across the six best-funded humanoid companies, exactly two have published a number you could audit: how many robots they delivered to customers. This piece is about that asymmetry, and about a pattern buyers can use — because the metric a humanoid vendor chooses to disclose tells you roughly where they are on the road from demo to product.

The capital is documented. The output mostly isn't.

Start with the money, all of it from company announcements or the outlets that broke them:

Now the output side, restricted to numbers the companies themselves put in writing:

• Figure — "delivering over 350 of our third generation humanoid robots," per Figure's own production update dated 29 April 2026. Same post: production rate went from one Figure 03 per day to one per hour, 9,000+ actuators built, 500+ battery packs shipped at 99.3% first-pass yield.
• Unitree — more than 5,500 humanoids shipped in 2025, defined in the company's own clarification as "the quantity actually sold and delivered to end customers, not order volume," with total mass-production output above 6,500 units. Both figures exclude its dual-arm wheeled robots.
• Apptronik — no unit count published. Apollo is in enterprise pilots at Mercedes-Benz, GXO and Jabil. Apptronik describes Apollo 2, unveiled June 2026, as a training and data platform, and calls Apollo 3 — targeted at 2027 — its first true commercial product.
• Agility Robotics — no unit count published. Instead, Agility says Digit has accumulated more than 65,000 operating hours across nine customer facilities, and passed 100,000 totes moved at GXO's Flowery Branch site in November 2025. Both figures reach us through trade coverage rather than a company filing, so treat the precision accordingly.
• 1X — 10,000 pre-orders in the first five days of taking them. Pre-orders are not deliveries.

Our view: the disclosure ladder is the real product roadmap

Look at what each company chose to report, and a ladder appears:

1. Nothing — Apptronik. The honest reading is that they agree: they have said in as many words that the commercial product is 2027.
2. Pre-orders — 1X. A measure of demand, and a real one, but it is a marketing number until the factory converts it.
3. Operating hours and totes — Agility. This is a genuinely useful disclosure and I would not dismiss it. 65,000 hours across nine sites means the machines are running unattended long enough to accumulate hours. But note what it *isn't*: hours can grow from a small fleet running longer, and the fleet size stays private.
4. Production rate — Figure. "One per hour" is a capability statement about the line, not a cumulative delivery figure. It is a forward-looking number dressed as an operational one.
5. Delivered units, audited — Unitree, in an IPO prospectus, with the revenue to match: RMB 1.699B in 2025, up 335%, adjusted net profit of RMB 590M, 60.13% core gross margin. Humanoids went from 27.6% of revenue in 2024 to 51.5% in the first nine months of 2025.

Companies disclose the highest rung they can reach. Reading the ladder backwards is a decent proxy for commercial maturity — better than reading the valuation, which as we are about to see runs almost exactly the other way.

The ratio nobody wants to compute

Take the two companies that have published a delivered-units number and divide:

A hundredfold difference. Let me be precise about what this does and does not show, because the number is easy to abuse.

It is not a valuation critique in the ordinary sense. Figure is priced on a bet about general-purpose autonomy — if Helix generalises, unit count today is irrelevant to what the company is worth. The dates are also mismatched: the valuation is September 2025, the delivery figure April 2026, and Figure has kept building since. And the two companies are selling different things at wildly different price points; a G1 and a Figure 03 are not substitutes.

What it does show is where the burden of proof sits. Figure's price implies the units are coming. Unitree's implies they already came. One of those is a claim about the future and one is an audited statement about the past, and a buyer should treat them differently even when both companies are equally sincere.

It is also worth noting the direction of travel is not what the Western narrative assumed. The company shipping the most humanoids, growing revenue 335%, and turning an adjusted profit is the one carrying the lowest valuation and heading for a public listing where the numbers get checked quarterly. This is the shipment side of the cost story we covered in why Chinese cobots undercut Western brands — structurally cheaper hardware turning into structurally higher volume.

What this means if you're actually buying

The gap between funding and delivery is not evidence of bad faith. Building a humanoid production line is genuinely hard, and Figure's actuator and battery yield numbers are the kind of unglamorous disclosure a company only makes when it is really doing the work. But it does change how you should read announcements:

• A funding round tells you about investor conviction, not availability. Apptronik raising $935M is a strong signal about the technology's prospects and no signal at all about whether you can buy an Apollo in 2026. The company has told you that directly.
• Ask which number the vendor is willing to put in a contract. Operating hours and totes moved are real, but they describe someone else's deployment. Delivered units and lead times describe yours.
• Treat pre-order counts as demand data. 1X's 10,000 pre-orders tell you a consumer market may exist. They tell you nothing about your delivery date.
• The proof-of-life question is boring and effective: how many of this exact model are in customer hands today, and can I speak to two of those customers? Every company above can answer that internally. The ones that answer externally are the ones further along.

This is the same discipline we applied to factory pilot announcements: the pilot press release and the deployment are separated by a gap that the press release is designed not to mention. The funding announcement and the delivery are separated by a bigger one.

What would change this analysis

Two things, and both are plausible inside twelve months. If Figure publishes a cumulative delivered-unit figure in the thousands, the ratio above collapses and the disclosure ladder stops being a useful proxy. And if Unitree's post-IPO filings show humanoid shipments flattening while revenue holds — meaning the growth came from price, not volume — then the one audited number in this space needs re-reading too.

Until then, the honest summary of the humanoid market in mid-2026 is this: about $3 billion went in, and the only companies that will tell you how many robots came out are the two at opposite ends of the valuation table.

Sources

• Figure — Ramping Figure 03 Production (29 April 2026): "over 350" third-generation robots delivered
• Figure — Series C: >$1B at $39B post-money valuation
• TechCrunch — Figure reaches $39B valuation
• Unitree — Clarification Regarding Unitree's 2025 Sales Data (22 Jan 2026): 5,500+ shipped, 6,500+ produced
• SCMP — Unitree ships more than 5,500 humanoid robots in 2025
• Caixin — Unitree wins approval for $618M STAR Market IPO (3 July 2026)
• Caixin — Unitree fast-tracks Shanghai IPO with $6.2B target valuation
• Apptronik — Closes over $935 million Series A
• CNBC — Apptronik raises $520M at $5B valuation
• The Robot Report — Apptronik and the Apollo product roadmap
• Forbes — Humanoid raises $152M at $1.35B valuation
• Technology.org — Humanoid Robots in 2026: What Is Actually Deployed (Agility operating hours, Apptronik pilot customers)