The memory shortage is reportedly making Nvidia-powered AI infrastructure even more expensive.
Some of Nvidia’s largest customers have reportedly been told that prices for servers containing its AI chips will rise by more than 15% in many cases, according to Bloomberg, which cited people familiar with the matter. Nvidia has not publicly confirmed the reported increases.
The increases are expected to take effect for systems shipped early next year and will include servers built around Nvidia’s latest Vera Rubin and Grace Blackwell platforms. The size of the increase will vary depending on the chip generation and memory configuration, according to Bloomberg.
Contract manufacturers that build servers for major data center operators, including Microsoft, Google and Oracle, have reportedly notified customers about the coming increases.
The pressure is coming largely from the memory market. Samsung Electronics, SK hynix and Micron produce most of the world’s DRAM, but demand from AI data centers has grown faster than supply.
Nvidia’s AI accelerators rely heavily on high-performance memory to move data quickly. As AI systems become more demanding, data center operators are buying more accelerators and memory, pushing prices higher.
Korea JoongAng Daily, citing market researcher TrendForce, reported that contract prices for server DRAM jumped 53% to 58% in the second quarter from the previous quarter and were expected to rise another 13% to 18% in the third quarter.
The pressure is notable because Nvidia has a gross margin of about 75%, according to Bloomberg. Rather than absorbing the higher component costs, the company appears to be passing at least some of them down the supply chain.
The cost could reach data center plans
The reported price increases would add another expense to already costly AI data center expansion plans.
That creates a dilemma for operators: continue expanding AI capacity at higher costs or slow projects and wait for hardware and memory economics to improve. The increase could also strengthen the case for companies to develop their own AI accelerators. Amazon, Microsoft, Google and Meta are all working on in-house chips, although they remain dependent on Nvidia hardware for much of their current infrastructure.
The AI boom faces a new cost test
The bigger issue is that rising memory prices could begin testing how much companies are willing to spend on AI infrastructure.
As Shin Joong-ho, head of LS Securities’ research center, told Korea JoongAng Daily, “If the cost of building AI infrastructure continues to rise, even major tech companies may find it difficult to maintain their current pace of investment.”
For Nvidia, the upcoming price increases may protect margins in the face of higher component costs. For its customers, however, they raise the cost of expanding computing capacity and could make competing AI chips more attractive.
Nvidia is scheduled to report its fiscal second-quarter results Wednesday, Aug. 26, giving investors an opportunity to assess whether demand remains strong enough to absorb rising infrastructure costs. For channel partners and data center operators, the reported increases mean AI projects may require larger hardware budgets, revised customer quotes, or a closer look at alternative accelerators and system configurations.
Read more: Rising memory and storage prices are creating new planning challenges for channel partners as AI infrastructure demand puts additional pressure on hardware budgets.