Hyperion: AI server revenue up 19.9% as Dell takes HPE's top spot

The figures matter to anyone buying AI or scientific servers. Revenue is climbing much faster than unit deliveries, so buyers pay more for roughly the same number of machines. Dell's lead over HPE changes which vendor sales teams call first. Neither Hyperion nor Intel has said how much of the growth is new AI work rather than replacement of ageing machines.

Hyperion puts six-month server revenue at $16.5bn

On-premises servers sold for HPC and AI-for-science work brought in $16.5bn in the first half of 2026, Hyperion Research said on 3 November. The comparison period came in at $13.8bn, a 19.9% increase. Unit deliveries rose just 4.4% over the same stretch. Server node volume grew 14.1%, so the average system carried more nodes than before.

The two published accounts of the figures disagree on the comparison period. Hyperion's own release describes the $13.8bn baseline as the second half of 2025, while HPCwire's report on the same numbers calls it the first half of 2025.

Dell ends HPE's 25-year run at the top

Dell took first place in on-prem HPC and AI-for-science sales with 24% of the market on $4.2bn of revenue in the first half of 2026. HPE recorded 21%, down one percentage point from where it finished 2025. HPE had held the top spot for over 25 years, according to Hyperion chief executive Earl Joseph.

HPE recognized significant revenue in 2025 from sales of multiple exascale-class leadership supercomputers.

Hyperion said Dell's growth reflected several large procurements and a growing group of customers buying mid-range AI-focused machines to pair with scientific and engineering work. Joseph said HPE's slower 2026 was only "slightly" slower.

Why the money is growing faster than the boxes

Hyperion put the gap between revenue and unit growth down to rising system costs. More GPU components inside servers and higher memory prices are behind the increase.

Intel chief executive Lip-Bu Tan told the company's earnings call the industry "continues to navigate one of the most significant supply shortages in history", driven by the "unprecedented demand for Super Intelligence".

The density shows in the hardware. A single Nvidia Vera Rubin NVL72 rack carries 36 Vera CPUs and 72 Rubin GPUs, and ASRock Rack showed one at OCP Global Summit 2026.

ASRock Rack's 2U HGX Rubin NVL8 system runs on a dual-socket Intel Xeon 6 platform, a reminder that accelerators still need host CPUs.

What to watch into 2027

Tan said Intel expects server CPU demand to stay strong through 2027 and 2028 as operators deploy more CPU-intensive agentic and inference workloads. He also put the total addressable market for server CPUs at about $40bn today, with a path to more than $200bn within five years.

Intel has not said how much capacity it is moving from client to data centre products.

For the market as a whole, Joseph said the long-term average growth rate was around six to eight percent, against 16.9% for the full year 2025. Hyperion recorded 23.5% growth in 2024, which Joseph called the hottest year in HPC in three decades. Hyperion groups traditional modelling and simulation systems with large AI training clusters under one AI-for-science label.