ARAX LABS
AI-Driven Predictive Scaling for Energy-Efficient Infrastructure.
Reduce data center energy costs by 15–30% through intelligent predictive scaling.
The Problem
Data centers waste billions in energy every year. The AI boom is making it worse.
of data center energy is wasted
- →Zombie workloads running with no real purpose
- →Static resource allocation that doesn't adapt to actual load
- →Observability systems scraping constantly — even when nothing happens
AI boom is overloading energy grids
- →Single AI training run: hundreds of megawatts
- →Grid operators can't predict data center load
- →Emergency peak power is expensive & dirty
The Solution
AI-driven predictive scaling that reduces energy consumption by 15–30%.
OPEX reduction through predictive scaling
Predictive
Forecast workload stability before problems occur. Not reactive autoscaling — proactive optimization.
Tool-Agnostic
Works across Kubernetes, Slurm, MPI, OpenStack, Terraform. No vendor lock-in.
Energy-Aware
Minimizes monitoring overhead while maintaining SRE visibility. Kepler-validated energy savings.
Core: eBPF + Kepler telemetry · LightGBM models · ~4ms inference · Model-Predictive Control
The Vision: Grid-SaaS
From data center optimization to energy utility intelligence.
Data Center Optimization
Data centers reduce their energy costs by 15–30% through intelligent predictive scaling.
Grid-SaaS for Energy Utilities
Energy utilities predict data center load to manage grid congestion caused by the AI boom.
"Our customer is no longer just the data center trying to save power. Our customer is the energy utility provider. We sell them predictive load forecasting to manage grid congestion caused by the AI boom."
Dashboard Preview
Live monitoring: chaos experiments, CPU metrics, stability timeline.
Technology Stack
Full-stack: eBPF telemetry → Kepler energy measurement → LightGBM prediction → MPC control.
Telemetry
eBPF, Kepler, Prometheus, OpenTelemetry
ML Models
LightGBM, XGBoost, LSTM, SGD Online
Orchestration
Kubernetes, Slurm, MPI, OpenShift, Terraform
Cloud
Azure, GCP, AWS, OpenStack, Oracle Cloud
Feature importance analysis — top 20 metrics for stability prediction
SHAP values — interpretable model predictions
Results
Empirical validation: LightGBM wins on accuracy, speed, and energy footprint.
Stability Timeline
Adaptive scaling maintains stability during workload transitions
Detection Delay
Fault detection delays under 5% vs. static configurations
Model Comparison
LightGBM: 10x better RMSE than baselines, 2x faster than LSTM
ROC Curves
~4ms inference latency — suitable for real-time control loops
Market Opportunity
Two revenue streams: direct data center savings + energy utility SaaS.
Data Centers
annual energy costs per mid-size facility
Energy Utilities
Predictive load forecasting for grid management
ARR potential by Year 2 (5–10 customers)
of customer savings = ARAX revenue
Traction
Real validation from day one.
Accepted into INiTS ScaleUp Program
Vienna's premier high-tech incubator
AWS First Incubator — pitch submitted
Infrastructure credits & mentorship
Technology demonstrated in relevant environment
Validated on real Kubernetes clusters
Master's thesis = direct R&D pipeline
FH Technikum Wien, MSc Data Science
Watch our pitch:
▶ @AraX-Labs on YouTubeTeam
Unique combination of energy domain expertise, cloud infrastructure, and AI/ML.
Arash Javan Mojarad
Founder & CTO
- →MSc Electrical Engineering (Energy Systems)
- →MSc Data Science — FH Technikum Wien
- →Platform Engineer — TU Wien VSC (MUSICA Supercomputer)
- →95+ certifications (Azure, Red Hat, LPIC-1, Oracle, IBM)
Unique Positioning
Energy Systems
MLOps
Distributed Systems
Most tools optimize one. ARAX connects all three.
Ready to reduce your energy costs?
Open to investors, grants, pilots, and partnerships.