Figure vs Tesla Optimus: Who Is Actually Shipping Humanoid Robots?

Two companies keep making headlines in the humanoid robot race, and they keep telling very different stories. One has finished an 11-month factory pilot with a documented production count. The other has a CEO who told investors, on the record, that his robots aren’t doing “useful work” yet. Here’s what the dated evidence actually shows.

Who is actually shipping right now?

As of mid-2026, Figure AI is the company with a completed, measured industrial deployment. Its Figure 02 humanoid ran for 11 months on the line at BMW Group’s Spartanburg, South Carolina plant, contributing to the production of more than 30,000 BMW X3 vehicles before BMW and Figure wound down the pilot and announced it publicly on November 19, 2025. Tesla, by contrast, has Optimus units inside its own Fremont factory, but on the Q4 2025 earnings call on January 28, 2026, Elon Musk told investors directly that no Optimus was doing “useful work” — the “few hundred” units in the building existed purely for training and R&D. That statement came roughly a year after Musk had told investors, in January 2025, that Tesla would build around 10,000 Optimus units in 2025 and have several thousand performing useful work by year’s end.

So on the narrow question of who has verifiable, dated proof of a humanoid working a real production task at volume: it’s Figure, not Tesla, at least through the first half of 2026.

That doesn’t settle the long-run race — Tesla has manufacturing scale, capital, and a stated roadmap most humanoid startups can’t match. But “who is shipping” and “who will eventually ship the most” are different questions, and conflating them is where a lot of hype gets generated.

What did Figure 02 actually do at BMW Spartanburg?

The Figure 02 pilot wasn’t a demo reel — it was a bounded assignment inside a working automotive plant, with numbers BMW and Figure both put on the record. Over the course of the deployment, the robot logged more than 1,250 hours of operational runtime and loaded more than 90,000 parts, working a measured cycle time of 84 seconds per part — including roughly 37 seconds dedicated to the loading action itself. BMW reported per-shift accuracy above 99%, which is the figure that matters most to a plant manager deciding whether to keep a robot on the line versus a person or a fixed-automation cell.

Those specifics come from Figure AI’s own account of the project and were corroborated independently by BMW Group’s press office and by trade coverage in Manufacturing Dive. That triangulation matters, because humanoid robotics announcements are frequently press-release-only claims that nobody else can check. This one has a named factory, a named line, and hour/part/accuracy counts from both sides of the partnership.

What happens after Figure 02 retires?

Figure 02’s Spartanburg assignment has ended — the unit is retiring from that role — and BMW is now bringing in its successor, Figure 03, for a more demanding logistics task: sequencing, a job that requires picking and staging parts in the correct order for downstream assembly rather than the more repetitive loading work Figure 02 handled. BMW Group’s press office confirmed this next phase itself, and it signals something notable about how Figure is positioning itself: rather than treating the Spartanburg pilot as a one-off proof of concept, BMW is expanding the robot’s scope of responsibility as each generation matures.

Figure 03 itself was unveiled on October 9, 2025, several weeks before the Spartanburg pilot results were announced. According to Figure’s own materials, the new generation carries 9% less mass than Figure 02, adds fingertip tactile sensors sensitive enough to register roughly 3 grams of pressure, supports 2 kW wireless inductive charging, and uses cameras with double the frame rate, a quarter of the latency, and 60% wider field of view compared to its predecessor. Note that these are the only Figure 03 specifications confirmed on Figure’s own site — absolute figures like height, weight, or degrees of freedom circulating in secondary blogs vary too widely (some list 168 cm and others 173 cm for height, for instance) to treat as verified, so we’re leaving them out rather than picking one arbitrarily. For a broader primer on what “degrees of freedom” and other core humanoid design terms mean, the glossary is a faster reference than any single spec sheet.

How is Figure funded, and how fast can it actually build robots?

Deployment numbers only matter if a company can also manufacture at scale, and Figure has been raising capital and scaling manufacturing at the same time. In September 2025 the company closed a Series C round that exceeded $1 billion, at a post-money valuation of $39 billion — a massive jump from the $2.6 billion valuation it held in February 2024. Backers named in the round include Parkway Venture Capital, Brookfield, NVIDIA, Intel Capital, Macquarie Capital, Salesforce, and Qualcomm Ventures, a list that mixes traditional venture money with strategic industrial and cloud-infrastructure capital.

On the manufacturing side, Figure has opened a dedicated facility called BotQ, designed with initial capacity for up to 12,000 humanoids per year and a stated four-year target of 100,000 units. Figure has also claimed it accelerated its build rate to one robot per hour, describing that as a 24-fold improvement achieved in roughly 120 days. Those are Figure’s own production claims, echoed in trade press coverage, and — like most manufacturing-ramp announcements from any young hardware company — worth treating as a target being chased rather than a steady-state fact until independently audited over a longer period.

Figure AI itself was founded in May 2022 by Brett Adcock, who serves as CEO and previously co-founded Archer Aviation (which went public in 2021) and the recruiting platform Vettery. That founder history matters as context: Adcock has already taken one hardware company through the IPO gauntlet, which is part of why investors have been comfortable backing an unproven humanoid category with billion-dollar checks.

What has Tesla Optimus actually delivered so far?

Tesla’s public position has shifted considerably over about twelve months. In January 2025, Musk projected Tesla would build roughly 10,000 Optimus units during the year and have several thousand doing useful work by the end of it. By the January 28, 2026 Q4 earnings call, the actual count Musk described was “a few hundred” units, all confined to training and R&D roles inside Tesla’s own operations — explicitly not performing productive factory work. That’s a direct admission from the company’s own leadership, not a third-party estimate, which is why it carries more weight than the various unverified claims of “1,000+ units running in Fremont” that circulate in secondary coverage; those higher figures are inconsistent with what Musk told investors and should be treated with considerable skepticism until Tesla itself confirms them.

When will Optimus actually go to work?

On the Q1 2026 earnings call, Musk gave the clearest timeline yet: Optimus production in Fremont is set to begin around late July or August 2026, roughly four months after the Model S/X line at that same factory wound down production in early May 2026 to free up space. The reveal of the next hardware generation, Optimus V3 — originally teased for Q1 2026 — has also slipped, with Musk now pointing to “roughly mid-year” 2026. A second dedicated Optimus production line is planned for Giga Texas, targeted to start around summer 2027.

On pricing, Tesla has stated a long-term target of $20,000–$30,000 per unit once production reaches scale, though the company has been candid that current build costs sit well above that figure. It’s worth flagging that Tesla’s 2026 volume targets are themselves a moving, contested set of figures: Musk’s original figure was around 50,000 units for 2026, some supply-chain analyst reports have cited a raised internal target closer to 70,000 units of annual capacity, and independent supply-chain analysts have pegged likely actual 2026 deliveries closer to 25,000 units, give or take 10,000 — well under any of the announced targets. Tesla itself has avoided committing to a hard number, with Musk calling forward guidance on Optimus volume “literally impossible to predict.” Given that spread, treat any single Optimus 2026 volume figure you see quoted elsewhere as one analyst’s estimate, not a settled fact.

Figure vs Tesla Optimus: the comparison as of mid-2026

	Figure (Figure 02 → Figure 03)	Tesla Optimus
Verified deployment	11-month pilot completed at BMW Spartanburg, ended Nov 2025	None yet doing “useful work” per Musk, Jan 28, 2026
Documented output	30,000+ BMW X3 vehicles supported, 90,000+ parts loaded, 1,250+ hours	“A few hundred” units, R&D/training only
Cycle time / accuracy	84s cycle (37s load), >99% per-shift accuracy	Not disclosed for production tasks
Current generation	Figure 03, unveiled Oct 9, 2025; now starting BMW sequencing work	Optimus (current gen); V3 reveal pushed to ~mid-2026
Manufacturing plan	BotQ facility: up to 12,000 units/yr initial capacity, 100,000 in 4 years	Fremont line starts ~Jul/Aug 2026; Giga Texas line ~summer 2027
Funding / valuation	Series C: $1B+ raised, $39B post-money valuation (Sep 2025)	Funded internally by Tesla; no standalone round
Target unit price	Not publicly stated	$20,000–$30,000 long-term target
2026 volume outlook	Not yet publicly guided at this granularity	Contested: 50,000–70,000 targets vs ~25,000 analyst estimate

Who’s actually ahead — Figure or Tesla?

If the question is “who has proven a humanoid can hold a real, measured factory job,” Figure is currently ahead, with an 11-month, third-party-corroborated deployment at a major automaker. If the question is “who can eventually build and sell humanoids at the lowest cost and highest volume,” Tesla’s vertically integrated manufacturing base, battery and motor supply chains, and stated sub-$30,000 price target make it a serious long-term threat once its production lines actually start — which, by Tesla’s own account, isn’t until the second half of 2026 at the earliest.

It’s also worth remembering this isn’t a two-company race. Chinese manufacturers pursuing a very different cost and deployment strategy are covered in our look at Unitree and UBTech’s approach to humanoid robots, and warehouse-focused humanoid deployment — a different use case than automotive assembly — is detailed in our piece on Agility Robotics’ Digit rollout. If you’re trying to model whether any of these robots pencil out financially for a given operation, our breakdown of humanoid robot unit economics walks through the cost side that press releases tend to skip. And for the funding and valuation trends shaping which humanoid makers get the capital to reach real deployment scale, see our 2026 physical AI market outlook. For the full picture of where humanoid platforms stand today, our hub for the humanoid robotics race tracks every major player as new deployment data lands. If you’re new to the category altogether, start with our primer on humanoid robots.