
各大银行希望以这些显卡作为抵押品发放贷款,保险公司则希望为其提供保障,投资者也渴望参与相关交易。目前,围绕英伟达的图形处理器(GPU)正在形成一个完整的金融生态系统;华尔街正试图将这类AI芯片打造为一种全新的资产类别,类似于房地产、石油及农产品。AI产业的蓬勃发展引发了市场对英伟达尖端芯片的抢购热潮;由于这些芯片价格高昂、难以一次性全额购买,一个专门的金融市场也应运而生。
Banks want to lend against them. Insurers want to protect them. Investors want to trade them. A financial ecosystem is forming around Nvidia's graphics processing units (GPUs), as Wall Street tries to turn the AI chips into a new asset class, like real estate, oil, and agricultural commodities. The AI boom sparked a scramble for Nvidia's cutting-edge chips — and a financial market has emerged because they're too expensive to buy outright.
倘若企业能够以GPU本身作为抵押品来融资,那么数百亿美元的资金便有望被用于建设新的AI基础设施,从而助力英伟达实现其“提供充足算力”的愿景。一套全新的金融运作模式已然形成:英伟达为相关交易提供担保;保险公司则负责防范GPU价值下跌带来的风险;而亚马逊等企业则选择将GPU交由独立实体管理,而非直接持有。
But if companies can borrow money against the chips themselves, it could unlock hundreds of billions of dollars for new AI infrastructure — and help realize Nvidia's vision of abundant compute. A new financial playbook has taken shape. Nvidia is acting as a guarantor for deals. Insurers are protecting against falling GPU values. And companies like Amazon are putting GPUs into separate entities instead of owning them directly.
随着市场的逐步成熟,这一领域还催生了大量新就业岗位。据Luxor Technology的首席运营官伊桑·维拉介绍,部分AI云服务提供商如今正招聘专职的算力交易员,负责买卖那些能够锁定未来价格的GPU合约。不过,这一切能否顺利实施,仍取决于两个关键因素:这些芯片的长期价值如何?它们的二次流通性又如何?英伟达持续快速推出新一代产品,这也引发了市场担忧:如今的GPU可能很快就会贬值。
As the market matures, it's also creating new jobs. Some AI cloud providers are hiring full-time compute traders to buy and sell GPU contracts that lock in future prices, said Ethan Vera, chief operating officer of Luxor Technology. Whether any of it works hinges on these questions: how much will chips be worth over time, and how resellable are they? Nvidia keeps releasing new generations at a rapid cadence, raising fears that today's chips could quickly lose value.
GPU融资业务兴起于2020年代初,当时多家银行及私人信贷机构开始向AI云服务提供商提供贷款。投资机构Upper90的联合创始人兼首席执行官比利·利比指出,2023年CoreWeave与微软签署的重要合约,让放贷机构确信:用于购买GPU的贷款终将得到偿还。如今,GPU融资初创企业American Compute的首席执行官伯尼·马古利斯表示,放贷机构正逐渐不再单纯依赖商业合约,而是开始认真考虑将GPU本身作为抵押品的可能性。目前来看,客户合约仍发挥着主要作用;不过,华尔街对这一领域的兴趣正与日俱增。
GPU financing took off in the early 2020s, when banks and private credit firms began lending money to AI cloud providers. A major contract between CoreWeave and Microsoft in 2023 gave lenders confidence that their loans to buy GPUs would be repaid, said Billy Libby, cofounder and CEO of investment firm Upper90. Now, instead of relying on contracts, lenders are increasingly considering whether GPUs themselves can serve as collateral, said Bernie Margulies, CEO of GPU financing startup American Compute. For now, customer contracts do most of the heavy lifting. Still, Wall Street's interest is growing.
英伟达已聘请贝莱德、阿波罗管理公司和高盛协助筹集超过5000亿美元的资金,用于建设人工智能基础设施。融资只是第一步,接下来需要打造一个市场平台,让企业能够锁定未来的GPU价格,同时也让投机者能对该资产的价格走势进行押注。Silicon Data的首席执行官卡门·李指出,要想让期货市场交易顺利运行,就必须确立一个可信的GPU价格标准。
Nvidia has enlisted BlackRock, Apollo, and Goldman Sachs to help raise more than $500 billion to finance AI infrastructure. If financing is the first step, the next is building a marketplace where companies can lock in future GPU prices — and speculators can take bets for and against this asset. For a futures market to work, GPUs need a trusted price, said Carmen Li, CEO of Silicon Data.
正因如此,Silicon Data、Compute Desk以及Ornn等公司正在着手制定相关指数,以统一芯片租赁价格的计算标准。与原油不同,计算能力目前尚无统一的价格标准,其价格会因芯片型号、地理位置及其他因素而有所差异。监管机构也担忧一些新出现的GPU指数是否可能被人为操纵。倘若这些指数得以广泛采用,便有望为GPU期货市场的发展奠定基础,投资者届时便可借此对GPU价格走势进行预测与押注。
That's why companies like Silicon Data, Compute Desk, and Ornn are creating indexes that standardize chip rental prices. Unlike a barrel of oil, there's no universal price for compute, which depends on the model of the chip, where it's located, and other factors. Regulators have also raised concerns about whether some nascent GPU indexes might be manipulated. If these indexes gain traction, they could pave the way for a GPU futures market, which investors could use to bet on where prices are headed.
从某种程度上来说,这样的未来已然到来:在Kalshi和Polymarket等预测平台上,人们已经开始对GPU价格进行押注;不少用户甚至预测了截至10月底英伟达B200芯片的租赁价格。,可参阅我们的相关指南。
In some ways, that future is already here. People are already betting on GPU prices on prediction markets like Kalshi and Polymarket, where users have made bets on how much Nvidia's B200 chips will cost to rent by the end of October.