⚡ Nvidia wants to use graphics cards as collateral for loans, but Wall Street isn’t buying it
Nvidia plans to turn GPUs into financial assets to unlock $500 billion. From August 10 to 12, it brought in six firms—Apollo, BlackRock, Blackstone, BoFeng, Goldman Sachs, and KKR—signed a memorandum of understanding, and set up a “compute-financing platform.” The deal is simple: AI companies use Nvidia’s GPUs as collateral to borrow money from these firms to buy Nvidia chips.
By early October, not a single one had actually been carried out.
Where’s the bottleneck? Depreciation. Banks price this kind of hardware based on a lifespan of three to four years. Nvidia argues its compute has output and durability, and can generate value for a decade. The gap is more than double, so no one dares to sign first. On October 1, Reuters spelled it out: lenders now want stronger guarantees—either from higher-quality counterparties standing behind the loans, or other forms of assurance. A GPU collateral alone isn’t enough. Nvidia is only willing to guarantee a residual value of 25% for each transaction; it won’t offer anything beyond that.
This kind of setup is familiar to the crypto crowd. From 2021 to 2022, BlockFi, Celsius, and Genesis used mining rigs and crypto assets as collateral to issue loans. When valuations collapsed, the whole chain detonated in sequence. GPU-collateralized loans follow the same template—just swapping the collateral from crypto mining hardware to graphics cards.
There’s another layer: self-financing. The seller puts up the money to help customers buy the seller’s own goods; when demand shifts, the financing pile collapses along with it. It’s the same risk category as FTX using its own tokens as collateral.
Look at Bitcoin the other way around. No depreciation. It trades 24/7. The secondary market is globally liquid. When a GPU has been used for three years, nobody can state its exact value. Quotes for secondhand AI accelerator cards are basically a guess. As a hard asset, the “BTC” label actually fits better.
Debt has already been piling up. By Goldman’s measure, this year, AI-related bonds issued by low-rated companies reached $88 billion, while in the first 11 months last year they were only $20 billion. High-yield bonds for AI infrastructure totaled $40 billion this year, versus $12 billion for the full year last year. The borrowing cost for low-rated borrowers has crept up to around 14% to 15%. Among the data-center joint venture bonds tracked by Goldman, 17 of 23 trades fell below the issue price.
When I wrote this, four BTC quote sources were between $85,332 and $85,348, up about 0.8% over 24 hours. ETH: 2,701. AI concept coins: TAO 301.6, FET 0.2342, RENDER 1.949—its daily percentage moves are three to four times that of Bitcoin.
Whether compute can be used as collateral ultimately comes down to the banks’ risk models. If that model isn’t fixed, $500 billion is just a number on a memorandum. What matters is the first compute-collateralized loan that is actually settled—watch how the terms are written.
$BTC $ETH
#中本聪国际社区Baoluo币商资本 #加密货币 #英伟达
Nvidia plans to turn GPUs into financial assets to unlock $500 billion. From August 10 to 12, it brought in six firms—Apollo, BlackRock, Blackstone, BoFeng, Goldman Sachs, and KKR—signed a memorandum of understanding, and set up a “compute-financing platform.” The deal is simple: AI companies use Nvidia’s GPUs as collateral to borrow money from these firms to buy Nvidia chips.
By early October, not a single one had actually been carried out.
Where’s the bottleneck? Depreciation. Banks price this kind of hardware based on a lifespan of three to four years. Nvidia argues its compute has output and durability, and can generate value for a decade. The gap is more than double, so no one dares to sign first. On October 1, Reuters spelled it out: lenders now want stronger guarantees—either from higher-quality counterparties standing behind the loans, or other forms of assurance. A GPU collateral alone isn’t enough. Nvidia is only willing to guarantee a residual value of 25% for each transaction; it won’t offer anything beyond that.
This kind of setup is familiar to the crypto crowd. From 2021 to 2022, BlockFi, Celsius, and Genesis used mining rigs and crypto assets as collateral to issue loans. When valuations collapsed, the whole chain detonated in sequence. GPU-collateralized loans follow the same template—just swapping the collateral from crypto mining hardware to graphics cards.
There’s another layer: self-financing. The seller puts up the money to help customers buy the seller’s own goods; when demand shifts, the financing pile collapses along with it. It’s the same risk category as FTX using its own tokens as collateral.
Look at Bitcoin the other way around. No depreciation. It trades 24/7. The secondary market is globally liquid. When a GPU has been used for three years, nobody can state its exact value. Quotes for secondhand AI accelerator cards are basically a guess. As a hard asset, the “BTC” label actually fits better.
Debt has already been piling up. By Goldman’s measure, this year, AI-related bonds issued by low-rated companies reached $88 billion, while in the first 11 months last year they were only $20 billion. High-yield bonds for AI infrastructure totaled $40 billion this year, versus $12 billion for the full year last year. The borrowing cost for low-rated borrowers has crept up to around 14% to 15%. Among the data-center joint venture bonds tracked by Goldman, 17 of 23 trades fell below the issue price.
When I wrote this, four BTC quote sources were between $85,332 and $85,348, up about 0.8% over 24 hours. ETH: 2,701. AI concept coins: TAO 301.6, FET 0.2342, RENDER 1.949—its daily percentage moves are three to four times that of Bitcoin.
Whether compute can be used as collateral ultimately comes down to the banks’ risk models. If that model isn’t fixed, $500 billion is just a number on a memorandum. What matters is the first compute-collateralized loan that is actually settled—watch how the terms are written.
$BTC $ETH
#中本聪国际社区Baoluo币商资本 #加密货币 #英伟达
