Amazon seeks to offload $8 bln of Nvidia chips to investors- FT

Amazon is reportedly seeking to move roughly $8 billion of Nvidia Grace Blackwell chips into a special-purpose vehicle that would finance the assets through external debt investors and lease the chips back to Amazon. The proposed transaction would reduce the balance-sheet burden of Amazon's AI-data-center buildout as the company plans more than $200 billion of 2026 capex, largely for AWS chips and data centers. The structure highlights hyperscalers' increasing use of asset-light financing for AI infrastructure, though the report does not indicate a completed transaction.

Analysis

The financial benefit to AMZN is likely optical before it is economic: moving accelerator purchases into an SPV can reduce reported capex and improve near-term free-cash-flow optics, but fixed lease commitments remain debt-like obligations. The valuation outcome hinges on the implied lease rate, contract tenor, residual-value guarantees, and whether AWS retains capacity-utilization risk; absent favorable terms, this is financing substitution rather than balance-sheet deleveraging. Investors should treat any multiple expansion as premature until those terms are disclosed.

For NVDA, the transaction changes the funding source for deployed systems, not end-customer demand. The more important second-order signal is that hyperscaler AI spend may be reaching a financing constraint: if outside capital requires high yields or credit enhancement, marginal GPU clusters may need materially higher AWS utilization to clear returns. That would be a 6-18 month risk to accelerator order velocity and could pressure lower-tier infrastructure vendors such as SMCI sooner than NVDA, since OEM/server demand is more exposed to project deferrals.

A successful placement could create an asset-backed-finance template for MSFT, GOOGL and META, extending the AI capex cycle by 12-24 months and supporting data-center supply chains. Contrarily, a weak reception would expose a gap between accounting capex capacity and economic funding capacity; credit spreads, not GPU availability, would become the binding constraint. The key near-term catalyst is pricing and investor appetite for the vehicle, rather than the announced notional size.

Key Decisions for Investors

• Maintain or initiate a 1-3 month long AMZN / short GOOGL relative position only if AMZN discloses financing at a cost below its incremental borrowing rate and without meaningful residual-value support. Upside is FCF/multiple relief; exit if lease-adjusted leverage rises or AWS capex guidance is not reduced on a run-rate basis.
• Do not add directional NVDA exposure on this development alone. Set an alert for securitization pricing above roughly 150-200 bps over comparably rated asset-backed debt, or evidence of sponsor guarantees: either would indicate that AI infrastructure economics require subsidy and would be a negative read-through for 6-12 month GPU demand.
• Favor NVDA over SMCI as a defensive AI-infrastructure pair for the next 3-6 months if financing markets tighten. NVDA retains the scarcer compute layer and broader customer base, while SMCI is more exposed to timing slippage in rack-scale deployments; close the pair on accelerating SMCI backlog conversion or a clear decline in GPU lead times.
• Watch AMZN’s next quarterly disclosure for lease commitments, AWS utilization/revenue growth, and capex trajectory. A material increase in commitments without corresponding AWS growth would falsify the asset-light thesis and support reducing AMZN exposure rather than rewarding reported FCF improvement.