AI-server suppliers seek up to 45% more after contracts are signed, FT reports
Rising memory costs are reaching customers at delivery. An ongoing Dell–XTX dispute shows how even an agreement described as fixed-price can become contested.
Loading page…
Rising memory costs are reaching customers at delivery. An ongoing Dell–XTX dispute shows how even an agreement described as fixed-price can become contested.
Listen to this story
AI-server buyers are facing a risk that signed prices will change before delivery, as suppliers seek to pass on sharply higher memory costs amid scarce equipment. JPMorgan research cited by the Financial Times puts memory-cost increases above 400% since early 2024, while reported server increases vary widely by type. The contract terms permitting repricing remain unclear; buyers’ willingness to pay reflects the value of securing computing capacity, not necessarily agreement that suppliers can change prices.
One person cited by the FT said GPU-server prices rose at least 20–30%, while CPU-node prices tripled or more.
Jane Street, Optiver and Hudson River Trading were among affected trading firms; the FT reported that firms largely accepted higher bills to obtain the equipment.
XTX Markets alleges Dell sought another $62–70 million on a $549 million server deal, despite a separate $5 million payment for price certainty.
Signing an AI-server contract is not necessarily settling the bill. Dell, HPE, Lenovo and Supermicro are seeking higher prices months after agreements were signed, the Financial Times reported on November 22. One person described increases reaching 45%, with demands arriving when equipment was ready for delivery—putting buyers’ budgets against their need for computing power.
The suppliers assemble servers containing graphics processing units, or GPUs—the main AI processors—alongside memory and networking equipment. Customers building their own data centers typically agree these purchases well before delivery. That gap is where the reported repricing is happening: the equipment is ready, but the supplier wants more than the figure agreed months earlier.
The FT attributes the increases largely to suppliers passing on soaring memory costs. JPMorgan research cited in the article puts the rise in those costs at more than 400% since the start of 2024. The explanation is a component-cost squeeze, rather than simply a higher price for the AI processors inside each machine.
The effect varies by equipment type. One person told the FT that GPU-server prices had risen at least 20–30%. CPU nodes—servers built around central processors—had tripled or more in price, the person said, noting that memory accounts for a large share of their cost. Those figures describe different equipment categories, not a uniform increase across every order.
The standard now is we’ll sign a contract with an indicative price and the price could be arbitrarily high by the time they deliver it
A person familiar with the matter, speaking to the Financial Times
Jane Street, Optiver and Hudson River Trading are among the financial trading firms affected, according to people cited by the FT. These companies buy specialized computing equipment for data centers that support research behind their trading strategies. They are among the largest computing-equipment customers after the biggest cloud operators, yet are still facing suppliers’ demands.
Trading firms are largely accepting the increases because they need the equipment, the FT reported. The person who described contracts with indicative prices said they had never refused a higher bill, explaining that the computing power remained valuable enough to pay for. Demand for AI equipment is outstripping supply, strengthening suppliers’ position.
Whether those contracts permit price changes remains unclear: buyers approached by the FT did not confirm their terms. HPE and Supermicro declined to comment; Dell and Lenovo did not respond to repeated requests. Optiver and Hudson River Trading also declined to comment, while Jane Street did not respond.
A dispute already in court shows the contrasting position of a buyer claiming explicit protection. XTX Markets sued Dell in an Irish court in June. According to court filings described by the FT, the companies agreed a fixed server price of $549 million, and XTX paid another $5 million to “ensure price certainty”.
XTX alleges Dell later demanded at least an additional $62–70 million to supply the servers. Its filing says Dell cited higher component costs and said it could not absorb rising memory prices. XTX also alleges Dell told its executives that other customers had explored alternatives, then returned and accepted price adjustments.
XTX declined to comment, and Dell did not respond to the FT’s requests about the lawsuit. The case remains ongoing, leaving that claimed guarantee of price certainty contested in court.
Loading discussion...
Join the conversation
Explain when securing the equipment outweighs holding the supplier to the original price.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.