According to the FT these GPUs are now being used as collateral for loans, despite being depreciating assets with limited shelf life.
CoreWeave, the most hyped neocloud unicorn, has morphed from a crypto mining minnow into a hyperscale GPU landlord. Backed by Nvidia and VCs, it’s sitting on 250,000 high-end chips and expanding like a bad spreadsheet formula with 32 data centres and counting. Its valuation rocketed from $2 billion to $19 billion in just 18 months, and it's eyeing a 2025 IPO that could inflate that number even further.
The firm has hoovered up over $10 billion in debt over the past year, including a $650 million lifeline this month from JPMorgan, Goldman Sachs, and Morgan Stanley. Not bad for a company that, as of last August, was still boasting revenues of just $25 million with negative EBITDA of $8 million.
A person close to the deal said that getting a Microsoft contract was vital because once they won it the contract then said that $2 billion of GPUs were required which could then be financed.
CoreWeave’s play is a poster child for a broader GPU economy. Wall Street types are increasingly drawn in by the promise of Nvidia-sourced hardware and long-term leasing deals with AI firms. But it's starting to smell like a silicon-fuelled feedback loop, with loans buying chips that back more loans to buy more chips.
Orso Partners short seller Nate Koppikar slammed the premise: “The lenders all coming in push the story that you can borrow against these chips and add to the frenzy that you need to get in now. But chips are a depreciating, not appreciating, asset.”
Crusoe chief executive Chase Lochmiller disagrees: “The scale of the investment makes a lot more sense once people start recognising this is the biggest capital investment in human history,” he said.
His firm just bagged a $3.4 billion deal from Blue Owl Capital to build a data centre that’ll rent GPU grunt to Oracle and OpenAI.
Still, trouble may be brewing under the hood. GPU compute prices have crashed from $8 per hour to $2. And with AMD racing to close the gap and tech giants like Google rolling out their own silicon, the party might not last. Supply has improved, demand's looking flimsier, and contracts anchoring these debt deals will begin expiring soon.
Sequoia Capital's Sequoia Capital said: “There’s a $500 billion gap between the revenue expectations implied by technology companies’ AI infrastructure buildout, and actual revenue growth in the AI ecosystem."
But lenders like Magnetar Capital are still charging ahead. The outfit's head of strategy Eric Falk said: “Predicting demand is incredibly challenging, but historically forecasts have underestimated future demand in many areas. The urgency here is immense, and it seems that tech companies are prepared to invest heavily to be at the forefront.”