Nvidia raises AI server prices over 15% as memory costs soar

Nvidia has told some of its largest customers that prices for servers containing its AI chips will rise more than 15% in many cases, as soaring memory chip costs push the cost of a flagship Vera Rubin rack toward $7.8 million. The increases take effect on systems shipped early next year, including configurations built around the flagship Vera Rubin and Grace Blackwell chips, according to people familiar with the process.

"The memory industry will face its worst supply shortage in 2027, with demand likely to exceed production capacity beyond 2030," SK Hynix Chief Executive Kwak Noh-jung said in July, according to Reuters. The Korean memory maker is raising $29.4 billion through a Nasdaq listing to fund new chip factories and EUV equipment tied to AI memory demand.

The magnitude of the price adjustments will vary depending on the chip generation and memory configuration ordered, the people said. Contract manufacturers that build servers for Microsoft, Alphabet's Google and Oracle have notified clients of the coming increases. TrendForce data shows conventional DRAM contract prices jumped 90 percent to 95 percent quarter over quarter in the first quarter of 2026, while NAND Flash prices rose 55 percent to 60 percent, with another 58 percent to 63 percent DRAM increase forecast for the second quarter.

Nvidia controls more than 70 percent of the data center AI chip market, and its market value has surpassed $5 trillion. The company reports earnings for the May-July quarter on Aug. 26, with Wall Street expecting revenue to roughly double year over year. The price hikes protect the 75 percent non-GAAP gross margin Nvidia reported for its fiscal first quarter of 2027, shifting the memory cost burden onto hyperscalers expected to spend nearly $700 billion combined this year on AI buildouts, per CNBC.

Memory Is Now a Real Line Item

Nvidia's own hardware shows how much memory costs have moved. A Morgan Stanley estimate reported by Tom's Hardware put the cost of a Vera Rubin-based VR200 NVL72 rack at about $7.8 million for hyperscalers, nearly double the roughly $4 million estimate for a GB300 NVL72 rack. Memory alone accounts for about $2 million of that, up 435 percent from the memory cost inside a GB300 system.

The shortage has handed unusual pricing power to the three companies that dominate DRAM production: Samsung Electronics, SK Hynix and Micron Technology. Micron has said customers including Nvidia committed $22 billion to lock in memory supply, according to Reuters. The effectiveness of Nvidia's AI accelerators depends on how much DRAM they are paired with, making memory allocation a bottleneck in AI server production.

The cost pressure is already reaching the rental market. Nebius raised on-demand GPU rental prices by roughly 30 percent from June 1, with H100 pricing moving from $2.95 to $3.85 per GPU-hour and B200 pricing from $5.50 to $7.15. Amazon followed with an increase of about 20 percent to EC2 Capacity Block pricing for machine-learning GPU instances from July 1, according to AWS pricing documentation tracked by Investing.com.

Who Bears the Cost

Nvidia has little reason to absorb the increase. The company reported a 75.0 percent non-GAAP gross margin for its fiscal first quarter of 2027 and guided to the same level for the second quarter, according to its investor materials. If memory inflation moves through system pricing, Nvidia protects the margin investors watch most; its customers get the other side of that trade.

For hyperscalers, the higher bill is a cost of staying competitive. Alphabet, Microsoft, Meta and Amazon are all pursuing in-house chip programs but remain dependent on Nvidia purchases for their data center build-outs, and their ability to gain independence also hinges on access to memory supply from Samsung, SK Hynix and Micron. Nvidia has also raised prices for its gaming-oriented PC graphics cards, the report said, citing Tom's Hardware.

Nvidia reports fiscal second-quarter results on Aug. 26, with recent Wall Street previews looking for about $92 billion in revenue against Nvidia's own guidance of $91 billion, plus or minus 2 percent. The margin commentary deserves as much attention as the headline number. Investors have spent months asking whether AI capital spending has outrun what the technology can return, and higher rack prices make that question sharper. Nvidia shares, trading near $215, fell 0.98 percent in the latest session.

This article is for informational purposes only and does not constitute investment advice.