It is not only access to computing power, but also its cost, that is becoming an increasingly significant constraint on the development of artificial intelligence infrastructure.
Some of Nvidia’s largest customers have been informed that prices for servers equipped with the company’s chips will rise by more than 15 per cent across many configurations. The price rises are set to apply to systems delivered from early 2027, including solutions based on the Grace Blackwell and Vera Rubin platforms. The information was reported by Bloomberg, citing people familiar with the matter.
The main reason is said to be rising memory prices. High-performance AI systems require ever-increasing amounts of DRAM and HBM, which is why memory manufacturers – primarily SK hynix, Samsung and Micron – are gaining an increasingly strong position in the supply chain. The pressure is such that, in July, Nvidia entered into a long-term partnership with SK hynix concerning, amongst other things, the next generations of HBM memory.
These price rises will increase the cost of the next wave of investment in data centres. For Microsoft, Google, Oracle and other operators, this means more expensive infrastructure expansions, and in the longer term may be reflected in the prices of cloud services and AI computing.
However, the scale of demand remains high. In the last quarter, Nvidia achieved a record $75.2 billion in revenue from data centres, 92 per cent more than a year earlier. The market will therefore be watching not only the pace of GPU sales, but also, with increasing attention, the cost of the entire ecosystem required to utilise them. Nvidia is due to report its next results on 26 August.
