ARAX LABS · Investor / Grant Pitch

AI-Driven Predictive Scaling
for Energy-Efficient Infrastructure

ARAX LABS turns telemetry into action: we predict instability before it happens, then adapt infrastructure and monitoring to cut energy waste and OPEX in AI / HPC environments.

15–30%target OPEX reduction

~4 msLightGBM inference latency

TRL 6validated in realistic environments

01 · Problem

AI infrastructure wastes power when it reacts too late.

Load is volatile, monitoring is expensive, and infrastructure still scales on static assumptions instead of predictive signals.

~30% energy waste

Zombie workloads, static resource allocation, and always-on observability create a persistent efficiency gap.

Grid stress rises with AI scale-up

Operators need early signals to avoid expensive peak power and congestion.

02 · Solution

Predictive scaling replaces guesswork with control.

Predictive

Forecast instability before it reaches production so operators can act early.

Tool-agnostic

Works across Kubernetes, Slurm, MPI, OpenShift, Ansible, and Terraform-managed environments.

Energy-aware

Uses eBPF and Kepler telemetry to reduce monitoring overhead while staying actionable.

LightGBMbest-performing model in the current pipeline

10xbetter RMSE than linear baselines in validation

03 · Product

One control loop: detect, decide, adapt.

94%stability score

4.2 msmodel latency

18%monitoring overhead cut

Signal pressure

CPU

Memory

Energy

Action queue

Adaptive scrape rate

Scale upwhen instability rises

Back offwhen the cluster is stable

Signals in

eBPF, Kepler, Prometheus, OpenTelemetry, cluster telemetry.

Model in the middle

Low-latency inference turns telemetry into a stability decision.

Action out

Adaptive scrape intensity, predictive scaling, and lower-energy operations.

04 · Tech & proof

Built on telemetry, validated by thesis results.

Telemetry

eBPF, Kepler, Prometheus, OpenTelemetry

Models

LightGBM, XGBoost, LSTM, SGD Online

Orchestration

Kubernetes, Slurm, MPI, OpenShift, Terraform

05 · Market

Start with data centers, expand into grid intelligence.

B2Bdata centers, cloud providers, AI infrastructure teams

B2B2Butilities, grid operators, energy market participants

Near-term value is operational savings. Long-term value is a predictive layer for the AI-powered energy stack.

06 · Traction

Traction comes from research, incubators, and deployment intent.

INiTS ScaleUp

Accepted into Vienna’s high-tech incubator with access to startup support and capital pathways.

AWS First Incubator 2026

Pitch video submitted, with infrastructure and mentorship needs clearly defined.

Research pipeline

ML pipeline complete, LightGBM validated on mock data, real-world experiments pending.

07 · Team & ask

Built by one founder who spans energy systems, MLOps, and supercomputing.

Arash Javan Mojarad

Master’s in Data Science at FH Technikum Wien, background in energy systems, distributed systems, and platform engineering at TU Wien’s MUSICA supercomputer.

What we need

Infrastructure credits, design partners, and strategic mentorship to turn research into a product.

AI-Driven Predictive Scalingfor Energy-Efficient Infrastructure