Spent yesterday mapping all 363 NVIDIA funding transactions since January 2024. Two very different things are being lumped together — and that matters.
~$14.5B of the $21.5B disclosed is **prepaid capacity**, not funded demand. That's money going upstream to Coherent, Lumentum, Marvell, Synopsys, Corning, Nokia, Intel — optics, custom silicon, EDA tooling. None of it loops back as $NVDA revenue. It's supply-chain bottleneck financing, not customer financing.
The **funded demand** piece — where NVIDIA backs entities that then buy NVIDIA compute — is real vendor financing, and it's a fraction of the headline number. That's the part worth sizing.
The structure has a name. Vendor financing took down Lucent and Nortel between 1998–2001. What's missing now is the measurement.
Here's the concentration risk nobody's talking about: **30 companies took capital from two or more nodes inside the same loop**. NScale got funded by NVIDIA, Nokia, and OpenAI in September 2025, then NVIDIA and Nokia again in October and March 2026. Nokia is itself an NVIDIA investee. The capital reaches NScale twice.
SB Energy took $1.0B from SoftBank and OpenAI in January 2026, then $1.5B from NVIDIA in August.
The problem isn't fraud. The problem is that **nobody can size the exposure**. If one node pulls back, several others are exposed to the same names at once.
We mapped all 305 companies, 363 transactions, 15 nodes. Every overlap, every double-touch. The loop is measurable now.
$NVDA $SPY $QQQ
~$14.5B of the $21.5B disclosed is **prepaid capacity**, not funded demand. That's money going upstream to Coherent, Lumentum, Marvell, Synopsys, Corning, Nokia, Intel — optics, custom silicon, EDA tooling. None of it loops back as $NVDA revenue. It's supply-chain bottleneck financing, not customer financing.
The **funded demand** piece — where NVIDIA backs entities that then buy NVIDIA compute — is real vendor financing, and it's a fraction of the headline number. That's the part worth sizing.
The structure has a name. Vendor financing took down Lucent and Nortel between 1998–2001. What's missing now is the measurement.
Here's the concentration risk nobody's talking about: **30 companies took capital from two or more nodes inside the same loop**. NScale got funded by NVIDIA, Nokia, and OpenAI in September 2025, then NVIDIA and Nokia again in October and March 2026. Nokia is itself an NVIDIA investee. The capital reaches NScale twice.
SB Energy took $1.0B from SoftBank and OpenAI in January 2026, then $1.5B from NVIDIA in August.
The problem isn't fraud. The problem is that **nobody can size the exposure**. If one node pulls back, several others are exposed to the same names at once.
We mapped all 305 companies, 363 transactions, 15 nodes. Every overlap, every double-touch. The loop is measurable now.
$NVDA $SPY $QQQ
