租卡云
Search documents
学习英伟达刺激芯片销售,AMD为“AI云”借款做担保
Hua Er Jie Jian Wen· 2026-02-20 01:53
Core Insights - AMD is accelerating AI chip shipments to catch up with Nvidia's lead in data center AI chips by providing financial backing for a $300 million loan to startup Crusoe [1][2] - The loan, secured by AMD chips and equipment, will be used to purchase and install AI chips in a data center in Ohio, with AMD agreeing to rent the chips if Crusoe cannot find customers [1][2] Transaction Structure - The $300 million loan from Goldman Sachs is backed by AMD chips and related equipment, with Crusoe deploying these chips in a data center built by 5C, supported by Brookfield [2] - AMD's guarantee includes a "last renter" arrangement, where AMD will rent the chips if Crusoe fails to find external customers, helping Crusoe secure a lower loan interest rate of about 6% [2] Replicating Nvidia's Path - Nvidia has previously supported cloud companies to rent its chips, enhancing their financing capabilities through investments and capacity purchase commitments [3] - Similar financing structures are emerging, with companies like Nscale securing significant loans backed by contracts and chips [3] Competitive Pressure - The financing arrangement with Crusoe highlights AMD's use of unconventional methods to capture market share in AI chips, with CEO Lisa Su aiming for annual sales in the "hundreds of billions" and at least 10% market share [4] - This follows AMD's previous agreement with OpenAI to sell chips with a total power capacity of up to 6 gigawatts over several years [4] Controversies and Exposure - Investors criticize AMD and Nvidia's financing and guarantee terms for potentially artificially inflating sales while redistributing the risk of demand fluctuations back to chip manufacturers [5][6] - For Crusoe, this financing provides ammunition for expansion and potential public listing, with an estimated valuation of around $10 billion and projected annual cash burn of $2 billion to $4 billion [6] - AMD's backing for AI cloud loans may accelerate chip deployment in data centers but also increase sensitivity to fluctuations in AI computing demand [6]