Nvidia is reportedly putting another $1 billion into Figure — and critics say the money is moving in a circle.

The report came from The Information on October 7: Nvidia, already a Figure investor, has discussed putting another roughly $1 billion into the company, which is said to be seeking financing at about a $38 billion valuation before the new money. Neither Nvidia nor Figure has confirmed the talks, and no closing date or terms have surfaced — figures that, as the coverage notes, are not directly comparable with the $39 billion post-money mark Figure set in its September 2025 Series C, which Nvidia also joined. On its own, a chip giant topping up a marquee robotics bet is unremarkable. What turned the report into a week of scrutiny is the shape of the money.

Nvidia already holds equity — about $500 million — in Nscale, the GPU-cloud provider; Nscale, in turn, took an undisclosed equity stake in Figure and supplies Figure's training compute; and under a deal announced September 3, Figure committed at least $3.5 billion, with an intent to scale beyond $6 billion, to buy as many as 100,000 of Nvidia's next-generation Vera Rubin GPUs routed through Nscale, with the first chips due at a Barstow, Texas data center in the second half of 2027. A direct Figure stake would add a fourth node, giving Nvidia financial exposure at the chip, cloud and hardware layers of a loop it is simultaneously evangelizing to the world's manufacturers. Short sellers Jim Chanos — who likened the structure to Lucent's vendor financing during the dot-com boom — and Michael Burry have called the pattern circular; a Bernstein analyst flagged 'circular concerns.' Nvidia rejects the label, arguing its partners 'do their own business,' that customers pay invoices within 53 days, and that independent investors co-participate in the same deals.

Underneath the financing question is a valuation one. At $39 billion, Figure is priced above Goldman Sachs' projection for the entire global humanoid robot market in 2035, which the bank put at $38 billion. Figure's claim to that mark rests on the most extensively documented production-scale humanoid trial to date: its Figure 02 ran ten-hour shifts at BMW's Spartanburg plant for eleven months, loading more than 90,000 parts across over 1,250 hours while supporting production of more than 30,000 X3s at above 99% placement accuracy per shift. But that is a single customer and a single task, and the newer Helix 2.5 software hit a 56% task-completion rate across 30 unfamiliar homes in September — real zero-shot progress, and also a reminder that failures remain frequent outside structured settings. Figure has never disclosed revenue; its BotQ factory runs at about one robot an hour toward a 12,000-unit design capacity, and it leases robots at roughly $1,000 a month.

For a buyer weighing a multi-year platform, a vendor's balance sheet is part of the diligence, and Figure is the hardest read in the category. The contrast this week was stark: as Agility opened audited books to go public, Figure — the sector's most valuable private company — was reportedly leaning on another billion dollars from its own chip supplier to hold a valuation that already exceeds the whole market analysts expect to exist in a decade. None of that means Figure 03 will not work; the BMW record is genuine and the Helix numbers are climbing. It does mean the signal a buyer most wants — independent, paying demand at scale — is exactly what a circular chip-maker check can obscure. The things worth waiting for before reading the valuation as a verdict: a confirmed round led by investors who do not also sell Figure its GPUs, a second named production customer beyond BMW, and Helix success rates that climb out of the 'fails often' range.