AI server suppliers raise prices for Nvidia systems as data center costs climb
Data center equipment vendors are increasing prices for Nvidia-powered AI servers months after contracts are signed, pushing higher infrastructure costs on to customers. The increases, which reach as much as 45%, are affecting major trading firms as demand for AI computing capacity continues to outstrip supply.
Highlights
- Dell, HPE, Lenovo, and Supermicro are demanding 25% to 45% higher payments for Nvidia-based AI servers at delivery versus agreed contract prices.
- Memory cost inflation—up over 400% since early 2024 per JPMorgan—has driven GPU server prices up 20% to 30% and tripled CPU node costs.
- Financial trading firms like Jane Street, Optiver, and Hudson River Trading face these surcharges, while XTX sued Dell over a $62 million to $70 million post-contract price hike.
Contract repricing hits AI server buyers
As first reported by the Financial Times, suppliers including Dell, HPE, Lenovo and Supermicro are seeking substantially higher payments when AI server orders are ready for delivery than the prices originally agreed. The companies sell systems built with graphics processing units from Nvidia and AMD, alongside memory and networking components, to customers building their own data centers.People familiar with the matter said the repricing comes months after deals are signed, even though such purchases are typically negotiated well before delivery. One person familiar with the matter said buyers are seeing increases of 25%, 30% and in some cases 45%, highlighting a widening gap between indicative contract prices and the final amounts paid.
Another person familiar with the matter said the current norm is to sign contracts with only indicative pricing, leaving customers exposed to sharply higher charges by the time equipment ships. They added that buyers keep accepting the terms because the computing capacity remains strategically valuable.
Memory inflation drives wider industry impact
The price increases are landing on some of the largest financial trading firms, including Jane Street, Optiver and Hudson River Trading, as they expand data center capacity for research supporting their trading operations. Financial firms are among the biggest purchasers of advanced computing equipment after large technology groups, but are still facing suppliers with strong pricing power in a market where AI hardware demand exceeds supply.The higher charges are largely tied to rising memory costs, which JPMorgan research says have climbed more than 400% since the start of 2024. One person familiar with the matter said GPU server prices have risen at least 20% to 30%, while CPU nodes, where memory makes up a large share of costs, have tripled in price or more.
One dispute has already moved into the courts. In June, XTX sued Dell in an Irish court, saying the companies had agreed a fixed price of $549 million for servers and that XTX had paid an additional $5 million to secure price certainty, but Dell later said XTX would need to pay at least another $62 million to $70 million because component costs had risen, according to court filings. XTX said it was shocked by the request, while the filings also said Dell told the firm it could not absorb higher memory costs. The case is ongoing.
Previously, we examined CoreWeave debt risks as heavy borrowing raised concerns about AI datacentre expansion sustainability.