While stock investors celebrate booming demand for chips and servers, credit markets are flagging risky contracts and rising debt costs Investors who’ve been in on the artificial-intelligence build-out have reaped handsome stock-market gains over the past few years. But as the companies behind the build-out seek to keep the momentum going, the funding for the AI boom is getting increasingly risky, according to Rothschild & Co Redburn. The “compute” layer of the AI supply chain — specialized cloud providers that rent out the chips and processing power to those building and running AI models — is coming under pressure, Rothschild & Co analyst Alex Haissl wrote in a note to clients.
Bullish investors will point to the high prices that AI processing power is fetching in the market: Due to infrastructure bottlenecks, buyers are paying, on average, over $20 billion per gigawatt annually. Haissl believes these prices are artificially inflated by well-funded AI startups called neolabs that are using funding from venture capitalists or large technology companies “to sign compute contracts at prices materially above those supported by broader market economics.” Don’t Short Yourself offers weekly money tips to help you earn it, stack it and grow it. I would like to receive updates and special offers from Dow Jones and affiliates.
I can unsubscribe at any time. He gives each of their stocks a sell rating. While equity investors are focused on the soaring demand for chips and servers, the credit markets are beginning to price in the structural fragility of the AI trade, according to Haissl.
Free cash flow from Big Tech companies like Amazon.com , Microsoft and Alphabet has dwindled as the result of heavy AI spending, leading these companies to take on debt. **More:**Oracle’s stock falls for the fifth day in a row — missing out on the AI bounce But Haissl believes the “the true extent of leverage is materially higher than reported figures suggest,” as conventional spending metrics fail to capture off-balance-sheet infrastructure commitments through leases and financial backstop guarantees these companies extend to smaller players. Hyperscalers such as Amazon, Microsoft, Meta and Oracle have over $1 trillion of future lease commitments on their balance sheets, according to Haissl. Specialized cloud providers like CoreWeave and Nebius have incurred higher borrowing costs for “less-established counterparties,” Haissl said, adding that neolabs are “signing higher priced compute contracts for training workloads that remain dependent on continued access to external capital.” As a result, lenders are wondering if these expensive contracts can be renewed when they expire.
When Nebius raised $5.75 billion in convertible debt in August 2026, the effective interest rate it incurred increased significantly when the duration of the debt extended from 2030 to 2034. CoreWeave’s March debt facility, which supports a contract with Meta, carried a funding cost of roughly 5.9%. The company’s latest August debt facility carried an all-in funding cost of roughly 9%, which Haissl attributes to the credit markets assigning a higher risk premium to contracts backed by AI startups instead of hyperscale cloud companies.
Representatives from CoreWeave and Nebius didn’t immediately respond to requests from MarketWatch for comment. Additionally, Haissl points out that macroeconomic factors such as interest-rate hikes could further increase the financing costs for CoreWeave and Nebius. **Also**: An AI bubble is no longer Wall Street’s biggest fear. This stock-market risk just took its place.
Extract — continue reading at the source.
