像极了2008，但数据在说「不」：黄仁勋的5000亿GPU赌局
- 核心观点：英伟达CEO黄仁勋联合六大金融巨头为AI芯片客户筹集超5000亿美元融资，将GPU视为可抵押资产，结构类似2008年次贷产品；但目前需求数据强劲，与崩盘前提相反，风险尚在可控范围。
- 关键要素：
- Apollo、BlackRock、高盛等六家机构共同融资超5000亿美元，为英伟达客户购买芯片提供资金支持。
- GPU被定义为新型资产类别，依据是其能产生收入、服务广泛客户且折旧周期超10年。
- 超级云厂商（谷歌、微软等）承诺未来支出达2.6万亿美元，谷歌上月出现史上首次负现金流，倒逼外部融资。
- 融资结构涉及资产打包、抵押借款、风险分层销售，与2008年房贷金融化操作高度相似。
- 融资条件严格，包括客户偿债能力、GPU盈利验证，且英伟达自担最高25%担保。
- GPU租赁价格自10月以来上涨约40%，明年芯片产能已售罄，Anthropic年收入从100亿涨至470亿美元。
- 崩盘风险核心在于需求消失，目前数据显示需求持续走强，但需观察芯片保值与收入持续性问题。
Original Author: Limitless
Original Translation: TechFlow
TechFlow Editor's Note: Jensen Huang convinced six financial giants to raise over $500 billion for Nvidia's customers, turning GPUs into collateralizable assets—a scenario eerily reminiscent of the 2008 subprime mortgage game. But the demand-side data tells a different story: chip rental prices are rising, and big tech revenues are exploding. This article breaks down the key signals and real risks of this high-stakes bet.
This Doesn't Look Like 2008... Or Does It?
Earlier this week, Jensen Huang announced something unexpected.
He convinced the world's six largest financial institutions—Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR—to collectively raise over $500 billion so his customers could keep buying Nvidia chips.
The fact that he persuaded these institutions to commit that much capital strikes me as insane (AI funding has become incredibly self-referential over the past year), but then I saw his pitch. In his own words:
This is truly the first time technology chips have become an investable asset class.
His argument is that GPUs meet the key characteristics of financial assets:
- Generate substantial income
- Serve a broad customer base
- Have a depreciation lifecycle of 10+ years
- Therefore, GPUs should theoretically be treated as assets that can be used as collateral for borrowing... a bit like houses in 2008...
- So, let's figure out whether this will eventually turn into a famous financial bubble burst like it did back then.
The $500 Billion Raise
Why would the world's richest companies need to raise $500 billion? The answer is simple: AI construction has become so expensive that even the most cash-rich companies in history can't fund it entirely on their own. Google even recorded its first-ever negative cash flow last month.
Hyperscale cloud providers (companies like Microsoft, Google, Amazon) have committed to a record $2.6 trillion in future spending on data centers, chips, and power. Google alone has roughly $900 billion in bills waiting to be paid.
The reason for this massive spending is that these companies expect AI to drive significantly more revenue. So they're investing now to lock in the computing power and GPUs needed to earn money in the future.
But when your own money runs out, where do you go? Wall Street. That's what happened yesterday. Jensen Huang isn't raising $500 billion because things are going badly; it's because spending has exceeded what companies can actually afford.
The Structure (2008 Flashback Warning)
Here's how the play works:
- Take an asset (in this case, GPUs)
- Package them together
- Borrow money using them as collateral
- Slice the debt into different tranches (a fancy way of saying risk levels)
- Sell the shares to yield-seeking investors
- This is almost exactly what Wall Street did with mortgages before 2008. Take a hard-to-value asset, financialize it, keep adding leverage, put everyone in a position where they're holding each other's risk... and pray the collateral doesn't depreciate.
I'll add a disclaimer: the financing Jensen Huang is raising depends on Nvidia customers meeting a host of conditions. In other words, Wall Street isn't just handing over $500 billion for free. They need to see:
- Customers have the money to repay the debt
- Customers are actually making money from using these GPUs
- Huang himself is even providing a guarantee of up to 25% to the lenders.
Okay, so if it looks like 2008, the ending must be the same, right? Honestly, I'm not so sure.
The Data Tells a Different Story
The reason the 2008 crash happened was that the entire system was built on the assumption that housing prices would never fall. Obviously, that assumption didn't hold up.
If we apply the same lens to today's AI demand... it's heading in the complete opposite direction.
GPU rental prices have risen about 40% since October, as market capacity continues to be snapped up. Next year's supply of the latest chips is essentially already sold out.
And these companies are making money. Anthropic's revenue jumped from roughly $10 billion to $47 billion in a year, with rumors they'll hit $100 billion by the end of 2026; Nvidia's quarterly revenue guidance sits around $91 billion.
Most of this is public information: crack open any earnings report from a top cloud provider and you'll see their revenues growing substantially.
And let's be honest, these companies are the smartest capital allocators on the planet. So if I had to bet, I'd wager they've done their analysis and are staring at a demand curve far larger than we expect.
We Must Be Optimistic Yet Cautious
I can't guarantee what this market will look like in 12 months... because no one can. The key things to watch are:
- Whether AI chips hold their value
- Whether they continue to generate income
- Whether the terms of the raised funds aren't excessively harsh
- A 2008-style crash requires demand to disappear, and so far, we're seeing the opposite (at least for now).
