Lambda has closed a $1.008 billion investment-grade loan at 6.78% and Sharon AI a $356 million facility at 9.95%, both secured on GPU servers. Lenders are now pricing GPUs as collateral, which makes residual value and lifecycle planning a balance-sheet question.
What was announced
- Lambda announced on 1 October 2026 a $1.008 billion delayed-draw term loan at a 6.78% fixed rate, rated A (low) by Morningstar DBRS and Baa1 by Moody's. It was marketed to insurance companies and fixed-income investors and was oversubscribed.
- The Lambda facility is secured by the GPU servers and related infrastructure it funds plus contracted cash flows, supports three committed deployments with two investment-grade customers, and is fully amortising with final maturity on 30 May 2033. Money is drawn only as clusters are commissioned.
- It is Lambda's second institutional credit facility, after a broadly syndicated loan that closed on 27 August 2026. J.P. Morgan was sole coordinating lead arranger.
- Sharon AI announced on 1 October 2026 its first GPU-backed SPV facility: $356 million of committed senior secured debt at a fixed 9.95% excluding fees, secured against the GPUs and associated cash flows.
- Sharon AI describes it as the first of a series of GPU financings behind a planned build-out of more than 68,000 NVIDIA GPUs by mid-2027, says it has raised over $2.6 billion of debt and equity in the past 10 months, and puts its contracted customer offtake at over $8.8 billion of total contract value.
- DCD notes that GPU-backed lending is still a young market because hardware has traditionally lost value quickly, and that CoreWeave and Nscale have taken similar loans.
The ETON view
The gap between these two deals is the story. The same kind of collateral, GPU servers plus contracted revenue, priced at 6.78% for Lambda and 9.95% for Sharon AI. Lenders are not really pricing the chips, they are pricing who has signed for the capacity and for how long. A loan amortising to 2033 against hardware with a much shorter front-line life only works if the contracts carry it, and if the GPUs still hold value as they move down from frontier training into inference, fine-tuning and secondary markets.
That second point matters to everyone running GPUs, not just neoclouds raising debt. If financiers are treating used accelerators as assets with a real residual value, operators should manage them that way: track configuration and service history, keep the full server rather than stripping it for parts, and plan the exit before the refresh rather than after. A complete, documented HGX or PCIe GPU server is worth more on resale and redeployment than a pile of cards, and the difference shows up in the total cost of ownership.
For hosting providers and enterprises building smaller clusters, this capital will mostly go to large contracted builds, which keeps pressure on new allocation. Previous-generation GPU servers, sourced and supported independently of any one OEM, are often the faster route to inference capacity. We supply and buy back GPU platforms across Dell, HPE, Supermicro, ASUS and GIGABYTE, and can put a realistic value on hardware you plan to retire.
Related infrastructure
Category: Market · Vendor: NVIDIA · Technology: GPU servers, GPU financing, neocloud infrastructure · Last verified: 04 Oct 2026 · ~2 min read
Sourcing this kind of infrastructure? Talk to ETON about availability, lead time and pricing.
