Figure AI is a humanoid robotics company building general-purpose robots for the home, with Helix as its onboard vision-language-action model that controls the hardware. Founded in 2022 by Brett Adcock, Figure AI develops AI-powered humanoid robots to address labor shortages, and its latest model, Figure 03, targets home use with advanced vision-language-action AI. The company is based in San Jose and has moved through three generations of hardware (Figure 01, 02, and 03) while training Helix to handle autonomous work across those platforms.
This launch is a compute commitment aimed at the training bottleneck behind that roadmap. Figure has signed a multi-year strategic partnership with Nscale to deploy the NVIDIA Vera Rubin platform with up to 100,000 GPUs, with initial deployment targeted for the second half of 2027 in Barstow, Texas, and an initial commitment of $3.5 billion . The commitment could scale to more than $6 billion over time , and the deal also includes an undisclosed strategic investment by Nscale into Figure itself, tying the two companies together in a way that goes beyond a simple vendor-customer relationship.
For founders and operators tracking physical AI, the significance is the framing Adcock puts on it in the launch video, that a robot in every home is fundamentally a data and compute problem, and Figure is now pre-buying capacity at a scale usually reserved for frontier LLM labs. It arrives against a crowded field, with Figure facing stiff competition from Tesla's Optimus division and China's Unitree, among many others , and it sets a concrete infrastructure floor for training Helix through the window when home-capable humanoids are expected to become viable.
Comments (11)
Sign in to join the discussion.
P
Priya VenkatSep 3, 2026
Committing $3.5B for GPUs before shipping a robot into a single home is such a Brett move. Respect the swagger, question the sequence.
K
Kwesi O.Sep 3, 2026
The b-roll of the humanoid folding laundry cut to the GPU render was chef's kiss editing. Whoever runs your video team deserves a raise before the next fundraise.
N
Noor AhmadiSep 3, 2026
Tweet copy reads like a Fed announcement. Where's the drama, where's the humanoid staring wistfully out a window?
M
Min-joo LeeSep 3, 2026
100k GPUs to teach a robot to load a dishwasher feels like using a particle accelerator to open a jar of pickles.
D
Dmitri KuznetsovSep 3, 2026
You build the mountain before you find the path up it. Sometimes that works.
S
sashaSep 3, 2026
Curious what the training data pipeline looks like at that scale. Are we talking teleop hours, sim rollouts, or a mix, because the bottleneck usually isn't the compute.
R
Ravi C.Sep 3, 2026
Every home robot company right now is basically betting the physics of folding a fitted sheet is a compute problem. Bold hypothesis.
B
Björn HalstadSep 3, 2026
Tagline suggestion, no charge: 'A robot in every kitchen, a GPU in every basement.'
E
Elena MárquezSep 3, 2026
Been telling my LPs the humanoid thesis is about compute leverage for months. Nice to see the memo land in public.
T
tomatoheadSep 3, 2026
Quick naive question, does the robot actually need cloud inference to fold a shirt, or is that for the training side only? Genuinely confused after watching the video twice.
F
Fatou BaldehSep 3, 2026
The home robotics market is smaller than everyone pretends. Most people don't want a 5'6'' machine watching them eat cereal in their underwear.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Figure AI
The future of home help is here
The future of home help is here
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Figure AI is a humanoid robotics company building general-purpose robots for the home, with Helix as its onboard vision-language-action model that controls the hardware. Founded in 2022 by Brett Adcock, Figure AI develops AI-powered humanoid robots to address labor shortages, and its latest model, Figure 03, targets home use with advanced vision-language-action AI. The company is based in San Jose and has moved through three generations of hardware (Figure 01, 02, and 03) while training Helix to handle autonomous work across those platforms. This launch is a compute commitment aimed at the training bottleneck behind that roadmap. Figure has signed a multi-year strategic partnership with Nscale to deploy the NVIDIA Vera Rubin platform with up to 100,000 GPUs, with initial deployment targeted for the second half of 2027 in Barstow, Texas, and an initial commitment of $3.5 billion . The commitment could scale to more than $6 billion over time , and the deal also includes an undisclosed strategic investment by Nscale into Figure itself, tying the two companies together in a way that goes beyond a simple vendor-customer relationship. For founders and operators tracking physical AI, the significance is the framing Adcock puts on it in the launch video, that a robot in every home is fundamentally a data and compute problem, and Figure is now pre-buying capacity at a scale usually reserved for frontier LLM labs. It arrives against a crowded field, with Figure facing stiff competition from Tesla's Optimus division and China's Unitree, among many others , and it sets a concrete infrastructure floor for training Helix through the window when home-capable humanoids are expected to become viable.
Comments (11)
Committing $3.5B for GPUs before shipping a robot into a single home is such a Brett move. Respect the swagger, question the sequence.
The b-roll of the humanoid folding laundry cut to the GPU render was chef's kiss editing. Whoever runs your video team deserves a raise before the next fundraise.
Tweet copy reads like a Fed announcement. Where's the drama, where's the humanoid staring wistfully out a window?
100k GPUs to teach a robot to load a dishwasher feels like using a particle accelerator to open a jar of pickles.
You build the mountain before you find the path up it. Sometimes that works.
Curious what the training data pipeline looks like at that scale. Are we talking teleop hours, sim rollouts, or a mix, because the bottleneck usually isn't the compute.
Every home robot company right now is basically betting the physics of folding a fitted sheet is a compute problem. Bold hypothesis.
Tagline suggestion, no charge: 'A robot in every kitchen, a GPU in every basement.'
Been telling my LPs the humanoid thesis is about compute leverage for months. Nice to see the memo land in public.
Quick naive question, does the robot actually need cloud inference to fold a shirt, or is that for the training side only? Genuinely confused after watching the video twice.
The home robotics market is smaller than everyone pretends. Most people don't want a 5'6'' machine watching them eat cereal in their underwear.