Skip to content
World clockEU--:--UK--:--USA--:--CN--:--PLDEFRIT中文EN

portal about AI and technologyevents · analysis · interviews · technical background

Search
LIVE
›

Figure AI freezes Figure 04 design and starts component validation

Figure AI completed design lock on its fourth-generation humanoid, Figure 04, in May and has begun validating components, Brett Adcock said in an all-hands video published on 22 October. The company is pairing that hardware work with an initial $3.5 billion compute commitment.

AI & modelsAnalysisGrace OkonkwoPublished: 9 November 20264 min readSources 35
Figure AI freezes Figure 04 design and starts component validation

A frozen design and a new hand

The all-hands video ran about an hour and contained no prototype demonstration.

Adcock spent much of it on where the hardware stands. Design lock on Figure 04 was completed in May, and components have entered the delivery and validation cycle, according to a BigGo Finance account of the meeting. A new generation of actuators, batteries and dexterous hands is in reliability testing in parallel. The team assembled ten complete battery packs the week before the meeting and will run load and thermal tests next. Adcock called the machine the largest improvement in any Figure product to date.

The hand absorbed the most engineering effort. Figure went through four major design revisions, and the latest targets near-millimeter precision in fingertip and contact-point tracking. Adcock likened the unreleased part to the first-generation iPhone; no specifications or launch date have been published. The Maeil Business Newspaper reported that Figure has developed a 16 degree of freedom hand, but that its limitations remain.

What Figure has shipped

Figure 02 completed an eleven-month deployment at a BMW plant in South Carolina, TechTimes reported. Two robots assisted in assembling more than 30,000 cars and handled more than 90,000 sheet metal parts. Built In reported that Figure 02 worked nearly around the clock, seven days a week, during a beta test at a BMW factory.

Figure 03, launched in October 2025, has taken on logistics sorting on BMW production lines, according to BigGo Finance. BigGo put its cumulative output at 1,000 units this summer and its cost more than 90% below the previous generation; eWeek, citing Figure's manufacturing update, reported that the company had delivered more than 350 third-generation robots, without a breakdown of commercial sales. Published specifications list a height of about 173 cm, a weight of 61 kg, a 20 kg payload, five hours of endurance and a top speed of 1.2 m/s.

At the company's BotQ plant, Figure reported in April that takt time improved from one robot per day to one robot per hour in under 120 days, a 24-fold gain. The announcement did not establish that the facility could sustain continuous hourly production.

Figure has said it plans to deploy 100,000 units within four years, the Maeil Business Newspaper reported.

Compute and an unconfirmed Nvidia investment

The compute side is where the money is committed. Figure's September announcement of a partnership with AI cloud provider Nscale carries an initial $3.5 billion commitment for computing capacity, with plans to expand beyond $6 billion. The arrangement targets access to up to 100,000 Nvidia Vera Rubin GPUs, with first deployments planned for the second half of 2027. The infrastructure is meant to support Figure's Helix AI system. Nscale said it would make a strategic equity investment in Figure and serve as its preferred compute provider; the size was not disclosed.

Nvidia's own money is a separate question. The Information reported that Nvidia has discussed another $1 billion in Figure at a pre-money valuation of about $38 billion, and no agreement has been announced. Nvidia was among the investors in Figure's September 2025 Series C, which raised more than $1 billion at a $39 billion post-money valuation.

Nvidia's role goes past writing cheques. Jensen Huang said the partnership lets Figure train models on Nvidia Vera Rubin hardware and then run them on Nvidia GPUs inside its robots, The News reported.

Data is the other half of the bet. Figure reported more than 16 million videos uploaded by contributors through its Index initiative in August, and the effect of that dataset on real-world performance remains unproven. TechTimes reported that sim-to-real training, in which behaviour is learned in a physics simulator and transferred to hardware, helped move the commercial-grade price band from $150,000 and above in 2023 to roughly $40,000 to $60,000 in 2026.

Figure has not published a Figure 04 price, specification or launch date. The next disclosed step is load and thermal testing of the new battery packs.

Comments 0

Sources

35
  1. 01Figure AI Sets Course for Next Leg: Figure 04 Design Frozen, $3.5 Billion Compute Bet on a "Robot Version of the Scaling Law"EN
  2. 02Nvidia Reportedly Eyes Another $1B Investment in Figure AI's Humanoid RobotsEN
  3. 03Only $5.71 an hour. It's the labor cost of the robot. Compared to the $28 an hour for U.S. warehouse..EN
  4. 04Humanoid Robots Investment Race Heats Up: Goldman 6x Forecast, China Leads With Spy LawEN
  5. 05Nvidia eyes major investment in Figure AI robotics: Here's what to knowEN
  6. 06What Is Figure AI? - Built InEN
  7. 07General Purpose Humanoid Robotics Market Report Now Available — Profiles Tesla, Unitree, Figure AI, Boston Dynamics, Apptronik & 22 Other Key PlayersEN
  8. 08Nvidia investigates CEO Jensen Huang's relationship with alleged Chinese spy EN
  9. 09Nvidia’s GR00T 2 teaches robots delicate skills from human videoEN
  10. 10Who is the suspected Chinese spy at centre of the Nvidia investigation?EN
  11. 11Nvidia opens investigation after report CEO Jensen Huang was target of foreign intelligenceEN
  12. 12NVIDIA Announces New Planned Investments in Korean Startups at the MSIT-NVIDIA AI Startup Partnership ForumEN
  13. 13The Download: empty AI climate claims, and an Nvidia investigationEN
  14. 14Practice Makes Precise: At CoRL, NVIDIA and Partners Reveal How Robots Rehearse Delicate Work in SimulationEN
  15. 15Nvidia CEO Jensen Huang admits to affair; board orders investigationEN
  16. 16Nvidia’s board needs to do some succession planning – sharpish | Nils PratleyEN
  17. 17Silicon Valley and Trump rally behind Nvidia CEO after his affair with alleged Chinese spyEN
  18. 18Nvidia espionage scandal weakens Huang's lobby against tighter AI chip export controlsEN
  19. 19Nvidia will investigate CEO Jensen Huang’s relationship with possible Chinese spyEN
  20. 20Xiaomi’s humanoid robots work a full 8-hour shift in its EV factoryEN
  21. 21Nvidia Launches Investigation Into Huang’s Affair With Suspected Chinese SpyEN
  22. 22Nvidia's CEO turmoil gives newly empowered Democrats fresh ammunition in pursuit of oversightEN
  23. 23What is Megaspeed? The Nvidia chip buyer in the spotlight after Jensen Huang’s affairEN
  24. 24How to minimize Wi-Fi roaming interruptions for autonomous mobile robotsEN
  25. 25Nvidia CEO's affair becomes a national security problemEN
  26. 26 Lawyer builds private AI server with eight used Nvidia V100 GPUs EN
  27. 27Google ships four TPU variants in one year as it ramps up competition with NvidiaEN
  28. 28IT-Infrastruktur: Hohe Nachfrage: Telekom investiert eine weitere Viertelmilliarde in KI-Fabrik in MünchenEN
  29. 29Nasdaq defies Nvidia impact as Huang faces espionage affair probeEN
  30. 30Fall um Nvidia-Chef: China weist Spionagevorwurf zurückEN
  31. 31DHS Has Used This Controversial Investigative Tool 185,000 Times Since Trump Took OfficeEN
  32. 32Read Huang's email to his staff and statement from Nvidia's board of directorsEN
  33. 33 Nvidia CEO Jensen Huang admits to having an affair with suspected Chinese spy EN
  34. 34Nvidia fallout: chipmaker faces extensive risks after CEO affair, analysts and lawyers sayEN
  35. 35Powered by GPT-6, Unitree humanoid robots master painting, chess without training, says CEOEN

All figures and quotations in this text come from the sources listed below.

Content prepared by the editorial team with AI assistance.

Grace Okonkwo

Grace Okonkwo

AI, models and technology

Grace Okonkwo covers AI, models and technology for FLASH24, working from primary sources such as model cards, API documentation and benchmark papers rather than vendor summaries. She checks training data provenance, evaluation conditions and reported scores against the underlying datasets before any figure reaches print. She interviews researchers and engineers directly, tracks release calendars from major labs, and compares successive model versions on the same tests. Her own self-hosting, home-network and documentation-reading habits feed straight into that desk, since she tests tools on her own hardware first. She does not publish benchmark claims without a reproducible method.

Newsroom →

Comments

0
  1. No comments yet — be the first.

Write a comment

Comments are public. We do not publish abuse, spam or advertising.