Nvidia Sits on Both Sides of the Compute Deal — and Wall Street Now Wants a Different Multiple for Each Side

On Aug 10, 2026, Nvidia signed memorandums of understanding with Apollo Global Management, BlackRock, Blackstone, Brookfield Asset Management, Goldman Sachs and KKR to \"mobilize over $500 billion of third-party capital\" for AI compute infrastructure. The framing — first circulated by the FT and confirmed in Nvidia's own press release — was that Nvidia chips, treated as long-duration revenue-generating assets, would become the collateral behind a new asset class. Nvidia capped its own skin in the game at a 25% residual-value guaranty on each deal.

On Oct 1, 2026, Reuters reported that the bankers and asset managers behind the syndicate want more. Per the report and follow-on coverage, lenders are pushing for either universal Nvidia guarantees on every deal, or for loans to be wrapped around revenue contracts from investment-grade customers (the Meta Platforms-style anchor). For now, S&P's view is that GPUs may not justify long-term residual-value underwriting without more historical data; TCW's Brian Gelfand was more direct — creditors don't subscribe to the long average lives Nvidia is selling.

Nvidia argues, and points to third-party studies to support, that hyperscaler server depreciation has extended from three to five or six years, and that its top-end GB300 NVL72 systems could earn revenue for nine to ten years. Bank underwriting desks are not buying it. The default valuation horizon for GPU collateral — what a CoreWeave-style lender actually books — remains a 3–4 year depreciation schedule.

The gap between those two numbers is the spread the Nvidia platform is asking investors to absorb. If lenders price GPU residual value at 30–40% of face value after four years and Nvidia only offers a 25% residual guaranty, the un-guaranteed tranche has to clear at materially wider spreads than the implied AAA rhetoric.

Three months of disclosures put the credit side of Nvidia in sharper — and much larger — focus than the chip side wants investors to see. The disclosure set is now layered:

Nvidia sits in the middle of a cap stack where it now plays supplier, lender, anchor equity investor and residual guarantor simultaneously. Three industry vectors matter most:

Stripped to its two halves, the Nvidia trade is two trades. The chip-vendor half — Q2 FY27 Data Center revenue $89.0B (up 117% Y/Y), 75.0% gross margin, $96.2B total revenue, $108B Q3 guidance that excludes any China contribution — is being priced like a hyper-growth semiconductor with a trailing P/E around 30x on $5.67T of market cap. The credit-book half is being priced like an investment-grade industrial with junior lender optionality, but with disclosure lag, residual-value tail risk, and a 3–4 year depreciation horizon that lenders won't waive.

The reason the Oct 1, 2026 Reuters story matters is that it is the first time the sell-side has named the bifurcation explicitly. Lenders are saying: if Nvidia wants its chips to be the collateral for Nvidia-papered credit, the credit terms have to clear a lender's underwriting standard, not Nvidia's. If the terms don't, the gap between announced financing and drawn financing — the difference between a $500B MOU and $500B of funded paper — will widen.

Short-term (days to quarters), the catalysts are concrete. The next Nvidia Q3 FY27 print (expected late November 2026) will reset the residual-guaranty and equity-investment disclosures in a 10-Q filed days later; any downward revision to that $108B guide would be the first hard signal that the circular structure is bending. Look also for the first finalized term sheets out of the $500B MOU — those will reveal whether the 25% backstop survives contact with bank counsel or whether it migrates closer to 100% on the IG-tenant-wrapped tranches.

Long-term (one to three years), the structural question is whether the chip-financing stack becomes an investable asset class or remains a vendor-credit narrative with a chip-vendor paint job. Three milestones will tell: (i) the first AAA-rated, GPU-collateralized CLO or ABS deal (no such instrument yet trades as of Oct 2026); (ii) the first neocloud with $20B+ of GPU-backed debt restructuring or paying in kind — CoreWeave's TTM $1.9B net loss and negative operating income is the canary; (iii) a Broadcom follow-on that markets cleanly as a standalone AI-XPV platform, proving the template isn't Nvidia-specific.