#亚马逊拟售80亿美元英伟达芯片 The key isn’t that “Amazon is done selling NVIDIA chips,” but rather the SPV (special purpose vehicle) + sale-and-leaseback approach—using it to move off-balance-sheet financing for roughly $8 billion worth of GPU assets
1️⃣What exactly is happening
1.Amazon plans to put thousands of deployed high-end NVIDIA Grace Blackwell AI chips from data centers across the United States into a newly set up SPV
2.That SPV raises funds by issuing debt to external investors; Amazon then leases the chips back from the SPV to keep running AWS and its AI business
3.Amazon may give investors up to about 10% equity in the carrier (the SPV), while not consolidating it—or consolidating it as little as possible
2️⃣Why do this
1.AI infrastructure is extremely expensive: Amazon’s capital expenditures this year are expected to exceed $200 billion, with the bulk going to buying GPUs for AWS and building data centers
2.Lighter-asset structure: turning the chips from “owned fixed assets” into “lease-use rights” reduces the on-balance-sheet capital intensity and improves perceptions of ROA and free cash flow
3.Lock in compute capacity without giving up control: the chips physically remain in Amazon’s data centers and are still used by Amazon—the change is in ownership and the financing structure
4.Following the GPU-collateral financing playbook: compute-power companies like CoreWeave have long used GPUs as collateral to borrow; big firms are now starting to financial-engineer as well
3️⃣A few points that are easy to misread
❌ Not that “Amazon is bearish on NVIDIA / clearing out and selling chips”
✅ It’s “use the chips while shifting pressure on the asset side to investors”
❌ Not that being off-balance-sheet means zero burden
✅ Lease obligations will still show up as future cash outflows; credit analysts will adjust their models by treating it like a form of debt
Deal not finalized: no bond issuance, no official announcement, no SEC filings confirming anything—this is only exploratory discussions
4️⃣What it means for each party
1.Amazon: in the short term, it improves the pacing of capital expenditures, while preserving its ability to expand AI capacity; in the long term, it adds another lease-liability
2.NVIDIA: neutral to slightly positive—Amazon is still deploying Blackwell at scale, and demand on the customer side hasn’t cooled
3.Investors / bond market: it could create a new category of asset-backed securities—“GPU-collateral + Amazon credit”—but the bet is on the chips’ residual value; if Blackwell is replaced by a newer generation in 2–3 years, the collateral haircut becomes a key risk
4.Industry signal: mega-scale cloud providers are shifting from “absorbing Capex themselves” toward “securitizing compute assets” $BTC $ETH
1️⃣What exactly is happening
1.Amazon plans to put thousands of deployed high-end NVIDIA Grace Blackwell AI chips from data centers across the United States into a newly set up SPV
2.That SPV raises funds by issuing debt to external investors; Amazon then leases the chips back from the SPV to keep running AWS and its AI business
3.Amazon may give investors up to about 10% equity in the carrier (the SPV), while not consolidating it—or consolidating it as little as possible
2️⃣Why do this
1.AI infrastructure is extremely expensive: Amazon’s capital expenditures this year are expected to exceed $200 billion, with the bulk going to buying GPUs for AWS and building data centers
2.Lighter-asset structure: turning the chips from “owned fixed assets” into “lease-use rights” reduces the on-balance-sheet capital intensity and improves perceptions of ROA and free cash flow
3.Lock in compute capacity without giving up control: the chips physically remain in Amazon’s data centers and are still used by Amazon—the change is in ownership and the financing structure
4.Following the GPU-collateral financing playbook: compute-power companies like CoreWeave have long used GPUs as collateral to borrow; big firms are now starting to financial-engineer as well
3️⃣A few points that are easy to misread
❌ Not that “Amazon is bearish on NVIDIA / clearing out and selling chips”
✅ It’s “use the chips while shifting pressure on the asset side to investors”
❌ Not that being off-balance-sheet means zero burden
✅ Lease obligations will still show up as future cash outflows; credit analysts will adjust their models by treating it like a form of debt
Deal not finalized: no bond issuance, no official announcement, no SEC filings confirming anything—this is only exploratory discussions
4️⃣What it means for each party
1.Amazon: in the short term, it improves the pacing of capital expenditures, while preserving its ability to expand AI capacity; in the long term, it adds another lease-liability
2.NVIDIA: neutral to slightly positive—Amazon is still deploying Blackwell at scale, and demand on the customer side hasn’t cooled
3.Investors / bond market: it could create a new category of asset-backed securities—“GPU-collateral + Amazon credit”—but the bet is on the chips’ residual value; if Blackwell is replaced by a newer generation in 2–3 years, the collateral haircut becomes a key risk
4.Industry signal: mega-scale cloud providers are shifting from “absorbing Capex themselves” toward “securitizing compute assets” $BTC $ETH