What TradingKey 美股財經 said in this video:
- CRWVneutrallong-termCoreWeave (CRWV) has about $100 billion in orders and has nearly sold out its computing capacity through 2027, but its model of receiving orders before cash means larger orders require more upfront cash and debt. The bond market doubts whether it can deliver data centers and power on time so that cash generated by its chips covers high interest and financing costs; the company has almost no room for error.Source
- Bonds & Ratesneutrallong-termThe bond market has sent a warning to CoreWeave and other AI neoclouds: unsecured bond yields have surged to about 15% and spreads have exceeded 1,000 basis points, showing investors demand extremely high risk compensation. Once interest rates rise or the financing door narrows and cheap capital disappears, the expansion model that relies on borrowing and upfront cash will turn from an accelerator into a pressure cooker, and the $100 billion orders will instead become a bill requiring huge prepaid cash.Source
- NVDAneutralNVIDIA (NVDA) will most likely not let CoreWeave suddenly collapse and will continue to support its chip supply, computing orders and market confidence; however, NVIDIA cannot pay CoreWeave's interest, data center rent and construction costs over the long term, so it can only reduce risk but cannot eliminate the risk on CoreWeave's entire balance sheet.Source
- Oil & Energyneutrallong-termFor AI neoclouds, the scarcest resource in the future will no longer be just chips, but the land, power and data centers that can make the chips run immediately. There is a multi-year gap between contracted power and actual energization; as long as power connection or delivery is delayed, revenue is postponed while interest and depreciation keep accruing, so only companies that lock in large amounts of power early can turn orders into cash.Source
- AI Capexneutrallong-termThe AI neocloud industry will not collectively collapse, but there will be a brutal shakeout: only companies that can obtain low-cost capital, lock in power early, and deliver data centers on time will survive and grow with AI demand; those with only chips and orders but lacking power, funding and execution will be forced to shrink, sell assets, or be acquired.Source