In AI lending, the customer's credit rating — not the GPUs — sets the price of money.
Lenders price GPU-backed facilities off the off-taker's standalone credit rather than the collateral, producing spread gaps of hundreds of basis points between identical hardware anchored by investment-grade versus junk customers.
The same conclusion keeps arriving from across the workspace's research — 2 topics independently instantiate this theme. Filter the evidence by where it came from:
The same GPU collateral prices 300bps inside CoreWeave's latest loan purely because the off-taker is investment-grade — customer credit sets the cost of money.
Lenders would only stand behind the anchor-tenant lease stream, confirming GPU collateral itself cannot set the price of money — the off-taker's credit does.
Structurally identical GPU-backed SPVs price 225+ bps apart purely on the off-taker's standalone credit — the hardware is a recovery floor, not the price-setter.
An unrated borrower with Nvidia inside its cap table clears at double-digit junk pricing despite GPU collateral — the loan is priced on creditworthiness, not the chips securing it.