Wall Street has placed trillions of dollars in wagers on artificial intelligence, and now some investors are rushing to position for the opposite outcome. Until now, this trade has been a consistent winner. Optimism about AI's potential to transform the economy has driven up share prices for tech giants, utilities, and chipmakers, while sparking speculative frenzies in other corners of the market. Recently, many bullish expectations have run into a reality check, with companies large and small failing to meet market forecasts. Now, investors are studying how to guard against a sharp reversal in the market. Stephen Solaka, a senior vice president at Gateway Investment Advisers, said: "A large number of clients want to hedge downside risk."
Anxiety has surfaced on multiple fronts: lenders are demanding higher yields for financing large data center projects and for purchasing chips; shares and bonds of heavily indebted AI "neocloud" companies like CoreWeave have tumbled; and tech favorites such as Alphabet and Intel have suffered violent sell-offs, with some individual stocks posting their largest single-day declines on record. The tech-heavy Nasdaq Composite Index fell 2.9% on Friday, capping its worst monthly performance since March 2025. Solaka manages client portfolios that are often heavily concentrated in single stocks, including the tech giants that have driven the broader market in recent years. He said a growing number of clients 鈥?some of them former employees of these very tech companies 鈥?want something else: insurance. "These people have made a lot of money, and now they are setting aside funds specifically to hedge the stocks they hold," Solaka added, noting that many companies impose restrictions on employee trading.
Investors and tech executives are turning to options and credit default swaps to hedge. Put options, which give investors the right to sell shares at a specified price at a future date, have seen persistently rising trading volumes. Data from Cboe Global Markets shows that the scale of option contracts tied to neocloud provider CoreWeave surged to a record high this month and has since remained elevated. The most actively traded CoreWeave options are puts expiring in 2028: if the stock falls to $35, the contract becomes profitable to exercise. This week, the shares traded around $55.
Meanwhile, data from bond data service provider Solve shows that over the past month, the cost of debt default protection for hyperscale cloud providers such as Amazon, Alphabet, Microsoft, Meta, and Oracle has risen, touching the highest level since 2021. Eugene Greenberg, co-founder of Solve, said: "There is not a single exception, which shows the market is worried about the entire sector, not an independent credit event at any individual company." The change in Oracle's credit default swap spread has been particularly notable, widening by more than 1 percentage point.
As the market expects the scale of AI infrastructure spending to keep expanding, traders have developed other tools to hedge related risks. Quantitative trading firm Jump Trading recently set up a compute trading desk. This new business unit is arbitraging the huge price gaps between different service providers, opportunities that have become hard to find in mainstream markets such as stocks and options. The firm is also designing over-the-counter derivatives for clients who want to hedge against a sharp drop in compute prices. Jump President Dave Olsen said: "Market participants are starting to carefully distinguish who the winners and losers are, asking tougher questions, and everyone is digging deeper into business models."
The logic behind these trades works as follows: a neocloud or AI infrastructure company that rents out Nvidia chips can buy put options to hedge against the risk that, if demand declines, the hourly compute rent it charges will fall sharply. The option gives it the right to sell compute at an agreed price by an agreed date, cushioning the hit to profits, with Jump taking the other side of the position. Jump can also negotiate cash-settled contracts, paying cash directly to the service provider; or the trading firm can take on compute leases and sublease them to others to earn the spread. According to the Silicon Data index, Nvidia B300 chip rental prices have jumped about 56% this year to $7.56 per hour. For example, Jump could agree to buy when compute prices fall to $4 per hour. Olsen and the trading team carefully review counterparty credit conditions and assess the likelihood of default.
Major exchanges and trading institutions also plan to roll out a series of compute derivatives. Nick Volbi, who heads institutional prediction markets at major Wall Street trading firm Susquehanna International Group, said multiple data centers and unlisted neocloud companies have approached the firm directly or through brokers, hoping to hedge chip exposure. If compute prices fall because of weakening demand or rising supply, such contracts would pay out, offsetting losses in the buyer's other business. He said these include put options linked to a compute index, products that amount to insurance and can be used to prove to bank lenders that risk is manageable. Volbi said: "I will go to a potential lender and say, if I buy this structure and get this insurance, can you lend to me? What are the terms?"
A small amount of trading has also emerged in prediction markets on states pausing new data center construction. The Kalshi platform has listed related contracts betting on which states will halt new data center projects before 2028, including New Hampshire, Michigan, and Maine. However, such trading is currently very thin. Olsen said the overall scale of hedging strategies remains small compared with the main AI trade, and many investors are still bullish. "I don't think my house will catch fire, but I still buy insurance," Olsen said. "Wanting to buy insurance doesn't mean you have to be bearish on the market."
Comments