Summary

A Bloomberg Tech Summit fireside with Brett Adcock, founder of Figure AI, recorded June 2025 — FIFTEEN MONTHS OLD at the time of filing, and read here as a dated snapshot rather than current fact. The interviewer is openly sceptical and says so, pressing Adcock on whether his targets are credible and whether the BMW deployment is a real commercial relationship or a pilot. Read alongside DK-88, the April 2026 interview with the same founder, this is the 'before' half of a ten-month comparison.

• The market thesis he repeats in both interviews: a little under half of world GDP is paid out as human labour, so a robot that does human work addresses that share. He describes the prize as 'the biggest business we've ever seen by many, many multiples'.
• Manufacturing claim as at June 2025: the BotQ facility has installed capacity of 12,000 robots per year PER LINE, and he stands by a target of 100,000 robots shipped within four years. His stated reason for confidence is a reclassification — 'we're not doing car manufacturing, we're doing consumer electronics manufacturing', which he calls several orders of magnitude easier.
• On customers: BMW is the named deployment, doing body-shop work moving sheet metal for welding. A second logistics customer is signed but unnamed; the interviewer says Bloomberg has reported it as UPS, and Adcock neither confirms nor denies.
• He is categorical that no teleoperation is used in deployed work — it is used only for data collection and testing — and argues that any company that does not set full autonomy as the day-one goal 'will find yourself on a local maximum really quick'.
• The cost claim, unprompted and specific: Figure 3 is '90% cheaper than Figure 2, like 93%'.
• On the OpenAI split, told diplomatically in this interview: 'we were just better at doing it alone', plus a reputational argument that partnering made people assume OpenAI had done the work. He confirms Helix uses an open-source vision-language backbone for semantic grounding, with data collection, training and inference done in-house.
• Asked whether he would rather have 10,000 robots in the field or 10,000 GPUs, he says the former without hesitation: 'we don't have a GPU problem. We have a data problem and a robot problem.'
• His structural prediction, and the most consequential claim in the interview: because robots share a neural network and improve collectively, this could be 'a winner take all market' — whoever fields the most useful robots gets both the cheapest and the smartest, and compounds.
• On funding, as reported at the time: a raise being discussed around $40bn, up from a last known $2.6bn roughly a year earlier. He says Figure holds more cash than the humanoid sector has raised in its history, and deflects the IPO question — 'I still have a lot of PTSD from being a public market CEO'.

Why it matters

Filed principally as a dated baseline. Its value is not what it says about 2026 but that it lets specific claims be checked against the same founder ten months later: capacity per line, the 100,000-in-four-years target, the BMW relationship, and the winner-take-all thesis. The one argument here that stands on its own is the collective-learning claim — if fleets share a policy and improve together, robotics concentrates in a way most hardware markets do not, and that would be a structural fact rather than a forecast. Everything else in this entry should be treated as a position taken in June 2025 by a founder raising money.

zObe3aOz5fw-transcript.txt