Rising memory chip costs are driving up the prices of servers powered by Nvidia Corp’s artificial intelligence chips, with some of the chipmaker’s biggest customers facing increases of more than 15%, Bloomberg News has reported.
The higher prices are expected to affect systems shipped early next year, including servers equipped with Nvidia’s flagship Vera Rubin and Grace Blackwell chips, according to people familiar with the matter. The scale of the increase will depend on the generation of Nvidia’s chips and the amount and type of memory configured in each system.
Server manufacturers that build systems under contract for major data centre operators, including Microsoft Corp, Alphabet Inc’s Google and Oracle Corp, have recently notified customers of the expected increases, Bloomberg reported. The sources requested anonymity because the communications have not been made public.
The development highlights the growing pressure that surging demand for AI infrastructure is placing on the global memory chip market.
Samsung Electronics Co, SK Hynix Inc and Micron Technology Inc have gained significant pricing power as demand for memory used in AI systems has surged. The three companies account for the bulk of global DRAM production, giving them greater leverage as technology companies compete for limited supplies.
Memory capacity is particularly important to Nvidia’s AI accelerators because the performance of the processors depends partly on how much high-speed DRAM they are paired with. As AI models become more demanding, data centre operators are deploying systems with increasingly sophisticated memory configurations.
The higher server costs come as technology companies already face broader supply pressures across the semiconductor industry. Apple Inc and Qualcomm Inc have recently indicated that chip shortages have contributed to higher prices.
Nvidia, meanwhile, continues to command substantial pricing power in the AI accelerator market. The company has a gross margin of about 75%, allowing it to charge tens of thousands of dollars for some of its processors amid supply constraints at Taiwan Semiconductor Manufacturing Co, which produces Nvidia’s chips.
Nvidia has also increased prices for some of its gaming-focused PC graphics cards, according to industry publication Tom’s Hardware.
The pressure is particularly significant for major cloud and technology companies. Amazon, Microsoft, Google and Meta Platforms Inc are developing their own AI chips to reduce their reliance on Nvidia, but they remain heavily dependent on the company’s accelerators for their expanding data centre operations.
Their efforts to build greater independence will still depend on access to memory supplies from Samsung, SK Hynix and Micron, underscoring the strategic importance of the memory market to the broader AI industry.
The rising cost of AI servers could also complicate the technology sector’s ambitious data centre expansion plans. Projects are already facing delays caused by labour shortages, tighter capital markets, community opposition and other infrastructure constraints.
With AI companies and cloud providers racing to expand computing capacity, higher memory prices could add another layer of cost and uncertainty to the industry’s massive investment cycle.
