"AI Boom Bill Is Coming": Nvidia AI Server Prices to Rise More Than 15%

Increase to apply to shipments from early next year as memory shortage worsens

Samsung Electronics and SK Hynix gain bargaining power as supply tightens

Nvidia, the world’s largest artificial intelligence chipmaker, has told major customers it may raise prices for AI servers as memory chip costs surge, Bloomberg reported. The move suggests Nvidia has begun passing higher component costs on to customers as investment in AI data centers accelerates.

Bloomberg reported on Aug. 22 that Nvidia recently told key customers prices for servers equipped with its AI chips could rise by more than 15%. The increase is set to apply to systems shipped from early next year. It is also expected to cover Nvidia’s flagship AI server systems, Vera Rubin and Grace Blackwell.

The size of the increase will vary depending on the chip generation and memory configuration. Server manufacturers have also relayed the pricing information to major data-center operators including Microsoft, Google and Oracle.

The main reason is a shortage of memory semiconductors. Nvidia’s AI accelerators are essential for training and running large AI models. To operate at full performance, however, the chips must be paired with memory including high-bandwidth memory, or HBM, DRAM and NAND. As the amount of memory required for each AI server climbs, pricing power has shifted rapidly to memory suppliers.

Bloomberg had previously reported that AI demand is driving a historic memory chip shortage. Data-center DRAM demand rose from 32% of total consumption five years ago to about 50% last year. By 2030, AI servers are forecast to account for more than 60% of global memory consumption.

The market is dominated by a small group of companies including Samsung Electronics, SK Hynix and Micron. Nvidia remains the dominant force in AI chips, but building servers still requires supply from those memory makers. That is why the three companies have gained leverage during the AI boom.

Bloomberg said the latest price increase underscores how much leverage memory chipmakers including Samsung Electronics, SK Hynix and Micron now wield at the negotiating table. Apple and Qualcomm are already facing pressure to raise product prices because of chip shortages, and Nvidia now appears unable to escape the same cost increases.

Higher prices could also affect investment in AI data centers. The Information reported that some server manufacturers told customers prices for Nvidia’s main AI server chip systems could rise by about 17%. If that happens, the cost of building a 1-gigawatt data center could increase by at least $5 billion.

Large technology companies are developing their own AI chips to reduce reliance on Nvidia. Amazon, Microsoft, Google and Meta are all pursuing in-house semiconductors for the same reason. But even with custom chips, reliable supplies of HBM and DRAM remain essential.

AI data-center construction is already running into obstacles including power procurement, permitting delays, community opposition, labor shortages and financing burdens. If prices for key equipment rise as well, the cost burden of big tech’s AI spending will increase further.

Lee Song-ryeol, Hankyung.com reporter yisr0203@hankyung.com