Figure AI Bets on 16 Million Videos, Not the Robot Alone

Figure AI's real bet is not the robot. Figure's bet is data and compute — 16 million crowdsourced human videos and up to 100,000 GPUs aimed at teaching a machine your chores.

The US humanoid race became a data-and-compute race

For a decade the American humanoid contest looked like an engineering problem: build a bipedal robot that walks, balances and grips. In the six weeks to 11 September 2026, Figure AI redrew the contest into something sharper — a race for data and capital. Figure did not ship a new robot. Figure shipped a crowdsourcing app and signed a compute deal, and together the two moves say more about where humanoid AI is heading than any hardware demo.

Start with the scoreboard. Figure has raised roughly $1.9 billion in total, according to Forbes. In the same window Figure committed $3.5 billion to compute — more than the company has ever raised — and built a data engine that has pulled 16 million videos from ordinary people in 108 countries. Figure bets the winner of the humanoid race will be decided by data and compute, not by the robot alone.

Index: paying the world to film its own labour

On 25 August 2026, Figure came out of stealth with Index, a crowdsourced data-collection app built quietly over about four months. Reporting from Humanoids Daily and Figure's own release put the launch numbers at 264,000 downloads across 108 countries, 44,000 weekly active users, and more than 16 million videos uploaded. Figure says Index processes about 30 minutes of footage every second.

The economics are the striking part. Figure has paid contributors $15 million so far and committed more than $1 billion to data and compute over the next 12 months. Every uploaded clip of a person folding laundry or loading a dishwasher becomes training fuel for Helix, Figure's vision-language-action model. Figure later reported Index passing 69,000 weekly active users, evidence the pipeline is still growing.

Figure's ambition is not modest. Founder Brett Adcock told TIME at the Figure 03 reveal in October 2025 that the goal is a humanoid in every household.

Every home will have a humanoid.

— — Brett Adcock, CEO, Figure

Nscale: a $3.5 billion bet on 100,000 GPUs

On 3 September 2026, Figure signed a multi-year infrastructure partnership with Nscale, confirmed in the Nscale release and on PR Newswire. The terms are unusually large. Nscale will supply up to 100,000 GPUs on the NVIDIA Vera Rubin platform, starting with an initial compute commitment of $3.5 billion and a stated intent to scale past $6 billion. First deployment lands in the second half of 2027 in Barstow, Texas.

Independent reporting confirms the scale. Data Center Dynamics and Forbes report the Barstow site as a 234-megawatt data centre, with Nscale taking a strategic equity stake in Figure and becoming the preferred compute provider for training future generations of Helix. Analysis from Forbes notes the oddity plainly: Figure is spending more on compute than the company has ever raised.

Our AI model, Helix, becomes more capable the same way every learned system does: with more data and compute.

— — Brett Adcock, CEO, Figure

Nscale's founder framed the logic as an industry turn toward the physical.

We're excited to partner with Figure as physical intelligence becomes AI's next frontier.

— — Josh Payne, CEO, Nscale

Meanwhile, Tesla is still installing the line

Set Figure's sprint against Tesla Optimus. In Tesla's Q2 2026 results on 22 July 2026, the shareholder letter said the first Optimus production lines were still being installed, with output anticipated "later this year" and early units reserved for training-data collection rather than customers. Tesla disclosed no Optimus unit count. As of mid-August 2026, Tesla had not officially revealed the Optimus V3 design.

Tesla's stated targets remain vast — a Fremont line aimed at 1,000,000 units a year and a longer-term second line at Giga Texas — yet the in-window reality is a slipping ramp, not a shipping product. Elon Musk was candid about the difficulty on the call.

This is going to be the hardest product to scale manufacturing that we've ever made at Tesla, because everything on the robot is new.

— — Elon Musk, CEO, Tesla

Atlas, Apollo and the shape of the American field

Two other American contenders round out the field. Boston Dynamics, with Hyundai, unveiled a production-ready all-electric Atlas at CES 2026, its reasoning supplied by Google DeepMind's Gemini Robotics; Forbes reported the Hyundai factory targeting about 30,000 robots a year. Apptronik raised $520 million at a roughly $5 billion valuation for the Apollo robot and unveiled Apollo 2 as a data-collection platform, with a targeted $80,000-a-year price at volume.

Read across the field, one pattern shows up. Boston Dynamics leans on DeepMind's models, Tesla builds an in-house stack slowly, and Figure has chosen to buy data and compute at unusual speed. The shared American bet is blunt: embodied intelligence now scales with data and silicon, the way large language models did.

The dignity question Emergent Intelligence forces

Here the picture turns human. The intelligence set to let a robot load your dishwasher is being distilled from hundreds of thousands of ordinary people filming their own labour. A new, planet-scale market in embodied human knowledge is forming — and the market deserves the dignity-first questions I group under Emergent Intelligence (EI), the frame I use for what is more commonly called AI.

Three questions matter. Whose embodied knowledge trains the machines? Figure's Index answer is 264,000 contributors across 108 countries. Who gets paid, and how much? Figure reports $15 million paid so far, though per-task economics stay undisclosed. And who is displaced by the fleet the videos build? The honest reading: the people filming their chores may be training their own replacements.

None of the above is a reason to look away. A $3.5 billion compute deal and a 16-million-video data engine are historically large regardless of any ethical frame. The dignity-first point is simpler: as embodied AI scales on human labour, the humans supplying the data deserve to be named, paid fairly and counted — not treated as raw material. Ubuntu asks for clear eyes, not cheerleading.

Frequently Asked Questions

These are the questions people are asking about Figure AI, Index and the humanoid data race. Short answers follow, drawn from Figure, Nscale and Tesla's own releases.

What is Figure Index?

In short, Figure Index is a crowdsourced app collecting real-world human video to train Figure's Helix robot model. Data from the August 2026 launch shows 264,000 downloads across 108 countries and more than 16 million videos uploaded.

How does Figure's Helix model learn?

Simply put, Helix learns from scale. According to Figure, the model becomes more capable with more data and compute, which is why Figure paired the 16-million-video Index pipeline with a $3.5 billion Nscale compute deal. Research on large models shows the same pattern of capability rising with data.

Why is the Nscale compute deal significant?

The key is scale and signal. The Nscale partnership commits up to 100,000 NVIDIA Vera Rubin GPUs and $3.5 billion, rising past $6 billion — analysis from Forbes notes Figure is spending more than the company has raised. Evidence points to compute becoming the humanoid industry's true bottleneck.

Who is filming the videos behind Figure's robots?

In other words, ordinary people are. Figure's Index data reveals 264,000 contributors across 108 countries recording everyday tasks, with $15 million paid to date. Research into the pipeline shows footage arriving at roughly 30 minutes every second.

What are the risks of the humanoid data race?

The answer is concentration and displacement. Analysis of the deals shows compute and data pooling inside a few well-funded firms, while the people filming their own labour may be training their replacements. Evidence on pay stays thin: Figure discloses a $15 million total, not per-task rates.

Sources:

Figure — introducing Index · Humanoids Daily — Figure unveils Index · Humanoids Daily — Index passes 69,000 weekly users · Figure — Figure and Nscale partnership · Nscale — Nscale and Figure · PR Newswire — Nscale and Figure sign partnership · Data Center Dynamics — $3.5bn compute agreement · Forbes — how Figure committed $3.5bn · Forbes — up to 100,000 NVIDIA GPUs · CNBC — Tesla Q2 2026 results · Engadget — Boston Dynamics electric Atlas at CES 2026 · Forbes — Atlas factory 30,000 per year · CNBC — Apptronik $520M raise · The Robot Report — Apptronik Apollo 2 · TIME — Figure 03 reveal · Related on this site: Figure 03 and the 119-hour livestream · Unitree, China and the humanoid IPO · Foundation models for embodied AI