Revenue from on-premises HPC servers rose by almost twenty percent in the first half of 2026. The number of systems sold increased by only 4.4 percent, however.
Revenue from on-premises servers for high-performance computing (HPC) and AI for scientific research reached 16.5 billion dollars in the first half of 2026, market researcher Hyperion Research reports. In the second half of 2025, it was still 13.8 billion dollars. The market is growing, then, and by a solid 19.9 percent.
Prices of RAM and GPUs
The number of systems sold rose by only 4.4 percent in the same period. Most of the market’s growth is therefore caused by higher prices per server.
Those prices follow the sharp rise in component costs. RAM is up to four times more expensive than a year ago, because builders of AI data centres are buying up the available supply. For the same reason, demand for GPUs is very high. Powerful HPC servers contain more and more GPUs, or more recent ones, which pushes their price up further.
The fact that more systems are being sold despite those prices indicates that interest in HPC servers remains high. According to Earl Joseph, CEO of Hyperion Research, organisations are replacing outdated machines and want to build AI into their workflows.
Dell is the largest
Dell is the new market leader, with server revenue of 4.2 billion dollars in the first half of 2026. That is just over a quarter of the total market. HPE was the largest player for more than 25 years.
In 2025, HPE booked a lot of revenue from the sale of several exascale-class supercomputers. Such big purchases do not recur every year. Dell benefited from a few large orders and from a fast-growing group of customers buying mid-range AI systems to combine AI with scientific and technical applications. “Dell has made full use of the demand for AI infrastructure,” says Joseph.
According to Hyperion Research, AI, large language models and the purchase of large systems with GPUs are lifting the entire HPC market to a stronger growth level. The extent to which the trend continues will be determined in part by the availability of the necessary hardware and the stability of its prices.