Predict risk node by node.
A live risk preview of every node — hours before thresholds are crossed. Built on the telemetry your facility already produces.
From reacting to alerts to seeing them coming.
Operators already sit on years of telemetry for every node. That history is the signal. We turn it into an early view of which nodes are behaving unlike themselves — so action happens before an incident, not after.
Alerts fire after the threshold is crossed.
Drift accumulates silently in the telemetry. The first signal an operator gets is a page — at which point the incident is already in motion and history is consulted retrospectively.
Drift surfaces against the node's own past.
Each node is modelled against its own baseline. When live telemetry drifts away from that pattern, the node is flagged and ranked — hours before any threshold is crossed.
Three steps. One signal.
No new sensors. No new pipelines. We read what's already there, model each node against its own history, and surface the ones drifting away from it.
Read — what's already there.
We ingest rolling telemetry directly from existing infrastructure — inlet and outlet temperatures, CPU clock history, power draw, fan curves, workload patterns. No new sensors. No new data pipelines. Everything OrionLinks needs is already being produced by the facility.
Analyse — each node against itself.
Every node is modelled against its own 90-day history — not vendor specs, not cluster averages. When thermal behaviour, power draw, or fan response drifts from what that specific node has historically produced under the same workload, the deviation is detected and scored. The baseline is the node's own past.
Surface — before any threshold is crossed.
Nodes whose risk score rises above their historical pattern appear on the operator's view, ranked and explained. The alert hasn't fired. There is time to plan, act, and fix — not just respond. The causal chain is always one click away.
One mission, three layers.
OrionLinks exists to make HPC facilities calmer, cheaper, and more sustainable. The node-level risk platform is the entry point today. Change management and waste-heat recovery extend the same philosophy deeper into the stack.
Risk preview platform
A proactive intelligence layer that reads each node's historical telemetry and surfaces the ones drifting toward failure before any alert fires. It turns the data the facility already produces into an early-warning signal that changes how operators start their day.
Change management & justification
When a larger operational change is on the table — a cooling upgrade, a setpoint shift, a capacity move — operators can simulate the impact, evaluate the trade-offs across six risk domains, and generate an auditable decision brief for leadership and compliance. Evidence before commitment.
Waste heat → reusable energy
A future hardware layer that captures the heat produced at chip level inside HPC and data-center environments and converts a portion of it back into usable energy for those same chips. Closing part of the energy loop at the source — without redesigning the facility.
Technical leadership.
Sebastian Irinciuc
Enjoys dissecting complex technical challenges and transforming ambitious ideas into working systems. Leads product strategy, customer discovery, and investor relations.
Jorge Iglesias
Has the technical depth to build something out of nothing and the engineering judgement to know what matters. Leads platform architecture and hardware R&D.
Backed by Europe's best.
OrionLinks is built within the ecosystem of Europe's leading innovation and academic institutions — accelerating our path from prototype to market.
European Institute ofInnovation & Technology
Universitat Politècnicade Catalunya
ESADE Business& Law School
Latest from OrionLinks.
Let's make HPC quieter to operate.
We're speaking with HPC operators, data-center managers, and early-stage investors who want to help shift operations from reactive to proactive.
→ sebastian@orion-links.com