Figure AI teases Figure 04 as humanoid robotics' "iPhone 1"

Figure AI CEO Brett Adcock just dropped the footage from the company’s latest all-hands meeting — kicked off by Arnold Schwarzenegger, no less — and made a massive claim: their next-generation humanoid, Figure 04, is "the iPhone 1 of humanoids."

While the robot itself was heavily blurred out in the public video, Adcock revealed that Figure 04 is now in "full design lock," with parts already shipping. The biggest upgrade seems to be its hands, which Adcock claims approach human-level dexterity and pack more actuators than the entire rest of the robot combined.

Burning $8.1M a week on data

Building super-dexterous hands is only half the battle; the real bottleneck is teaching them what to do. To train its humanoid brain, Figure has turned to sheer scale.

The company’s Index app is currently pulling in 81 minutes of human video every single second, totaling over 37.5 million uploads. That pipeline is shockingly expensive: Figure reportedly spent roughly $8.1 million on data in just a single week.

Adcock claimed Figure will spend more money on data over the next 12 months than most robotics startups have ever raised in total. Backing up that bet is an enormous war chest: over $2 billion in cash and a $3.5 billion compute deal.

All of this data is feeding a new model called Helix 3.0, which Adcock refers to as "baby AGI." It has consumed more compute and training data than any previous Figure system, and an even larger version is slated to drop within weeks, though Figure has not yet released any official benchmark results.

Why it matters

Figure is essentially running the OpenAI and Anthropic playbook inside a physical chassis. Instead of obsessing over niche control algorithms, Adcock is betting that brute-forcing compute, capital, and massive datasets will solve embodied intelligence.

If this scaling law holds true in robotics, it will completely reset the playing field. Smaller robotics startups relying on tens of millions in funding will find it nearly impossible to compete with a player spending billions just on video ingestion and GPU clusters.

That said, we think some healthy skepticism is warranted. Announcing the "iPhone 1 of robotics" while completely blurring the hardware and withholding benchmark data looks more like a Silicon Valley hype sprint than a proven breakthrough. Cramming more actuators into a hand than the rest of the body creates enormous mechanical reliability and thermal headaches. Until Figure shows Figure 04 operating reliably in real-world environments without video edits, we will treat the "baby AGI" claims as an ambitious bet rather than a finished reality.