HPC Is Not Just Riding the Coattails of AI (www.nextplatform.com)

At SC25, Hyperion Research laid out fresh market numbers showing AI workloads are turbocharging datacenter and HPC spending — but not just as a passenger. In 2024 the combined HPC-AI market reached about $59.9B (on‑prem $50.4B, up 22.9%; cloud $9.5B, up 4.9%), with cloud now representing ~16% of spend. Hyperion teases apart true “AI-augmented” scientific computing from general GenAI and finds that AI funding is fueling upgrades to traditional simulation and modeling stacks, driving above‑trend growth that’s projected to remain roughly double the historical 7–8% annual pace through the rest of the decade. Key technical and market takeaways: on‑prem systems are more compute‑heavy (compute $25.3B vs storage $6.68B; compute:storage ~3.77:1) while cloud spending allocates a larger share to storage (30% vs 21.7% on‑prem). Services remain a major line item; software is ~5% of spend. Hyperion predicts AI‑centric hardware will outsell HPC‑centric iron around mid‑2027, blurring distinctions between architectures. Original design manufacturers (ODMs) now rival top OEMs like HPE in on‑prem HPC‑AI server revenue, and hyperscaler/model‑builder capex (hundreds of billions, multi‑GW scale) is reshaping price bands, power budgets and procurement models — including DOE moves toward cloud-hosted supercomputer outposts that could smooth revenue volatility.