Nvidia is raising AI server prices by more than 15% as memory costs soar

Free daily briefing on global business news.

The increases will apply to systems containing Vera Rubin and Grace Blackwell chips and take effect on shipments early next year

Ying Tang/NurPhoto via Getty Images

Nvidia $NVDA plans to raise prices on servers containing its artificial intelligence chips by more than 15% in many cases, with the increases set to take effect on systems shipped early next year, according to Bloomberg.

Vera Rubin and Grace Blackwell chip-based systems are among those subject to the increases, with the exact percentage varying by chip generation and how much memory a given configuration includes. The company did not respond to requests for comment.

Contract server manufacturers that supply major data center operators — Microsoft $MSFT Corp., Alphabet $GOOGL Inc.'s Google, and Oracle $ORCL Corp. among them — have recently passed word of the upcoming price adjustments to their customers, Bloomberg reported, citing people familiar with the matter.

The driver behind the hikes is the rising cost of memory chips, which are a core component of Nvidia's GPUs and server systems. Those three companies collectively dominate global DRAM production, but even as each has ramped up manufacturing capacity, the explosive growth in AI infrastructure spending has outrun their ability to supply the market, sending prices upward.

Nvidia posts a gross margin of 75%, and its chips command prices in the tens of thousands of dollars apiece — a reflection of persistent shortfalls in supply from contract manufacturer Taiwan Semiconductor Manufacturing Co. relative to what buyers need. That Nvidia — the market's dominant force — is passing these costs on rather than absorbing them signals how commanding a position memory chip suppliers have carved out as AI infrastructure spending surges.

Nvidia is also scheduled to report fiscal second-quarter earnings next week.

The price increases come as Nvidia's largest customers are pursuing their own custom AI chip programs in an effort to reduce their dependence on Nvidia hardware. Google is already shipping its seventh-generation Ironwood chip, and Amazon $AMZN's Trainium3 has been available since late 2025. Microsoft's Maia 200 inference chip began reaching U.S. data centers earlier this year. Despite those efforts, all of these companies remain dependent on Nvidia purchases for the bulk of their data center capacity.

How quickly customers move toward rival chips in response to the hikes may ultimately hinge on whether they can lock in adequate memory supply from the same trio of manufacturers whose pricing is squeezing Nvidia.

Join 500,000+ readers who start their day with Quartz.

By subscribing, you agree to our Terms of Service and Privacy Policy.

Global business news for a smarter world