Tech Products intermediate 3 min read Mar 18, 2026 · Updated Mar 20, 2026
Public Preview Sign in free for the full digest →

30% of data center power sits unused — this startup just raised $12M to unlock it

“Data centers waste 30% of their power capacity because GPU spikes are too fast to predict — Niv-AI just raised $12M to fix it.”

30% of data center power sits unused — this startup just raised $12M to unlock it
8 Views
0 Likes
0 Bookmarks
Source · techcrunch.com

“The grid is actually afraid of the data center consuming too much power at a specific time. The problem we're looking at is a problem with two sides of the rope. One is to try to help the data centers utilize more GPUs, and hopefully make more of the power that they're already p...”

You know that feeling when you provision expensive GPU clusters but your data center tells you to throttle back because of 'power constraints'? Here's what's actually happening: modern GPUs create violent, millisecond-level power surges as they switch between computation and communication. Data centers can't predict these spikes, so they assume the worst-case scenario and heavily buffer power usage. Before: you pay for 100MW but can only safely use 70MW because the grid can't handle surprise surges. Now: Niv-AI maps your workload's electrical fingerprint and orchestrates power in real-time, unlocking that stranded 30%.

ai-infrastructuredata-centergpuenergyhardwarestartupsisrael

Think of it like a heart monitor for your data center. Niv-AI deploys high-resolution sensors at the rack level that capture power usage at millisecond granularity — standard facility meters completely miss these rapid transients. This data reveals the unique 'electrical fingerprint' of different AI workloads: training GPT-4 looks different from running inference on Llama. An AI model then learns to predict these patterns and synchronize power loads across the facility. Instead of buffering for worst-case spikes, the system actively orchestrates compute to smooth out demand — like a conductor coordinating instruments so the orchestra never gets too loud for the concert hall.

01
Millisecond-level power sensing — why YOU care: Standard meters sample every few seconds and miss the violent transients that trip circuits. Niv-AI's rack-level sensors capture data 1,000+ times per second, revealing patterns invisible to ...
02
Electrical fingerprinting — why YOU care: Different AI workloads have distinct power signatures. Training creates different spikes than inference. Knowing these patterns lets you schedule workloads to avoid simultaneous surges.
03
Real-time orchestration — why YOU care: Instead of passively monitoring, the system actively coordinates when workloads draw power. Think load balancing, but for electricity instead of network traffic.
04
AI-powered prediction — why YOU care: The system builds models on collected data to forecast power demand. This moves you from reactive (buffer for worst case) to proactive (schedule around predicted patterns).
05
No hardware addition required — why YOU care: The value comes from intelligence, not infrastructure. You're unlocking capacity you already paid for, not buying new equipment.
06
Grid-friendly power profiles — why YOU care: Utilities hate unpredictable loads. Niv-AI creates smoother demand curves, which could mean better rates and fewer grid interconnection headaches.
07
Copilot for data center engineers — why YOU care: The founders envision an AI assistant that helps engineers make real-time decisions about workload placement and power management.
Who it’s for

If you're a data center operator running GPU clusters for AI training or inference — this is directly for you. Also relevant for ML infrastructure leads at companies building their own AI capacity, or anyone whose GPU utilization is bottlenecked by power constraints rather than compute. Not useful yet if you're running small-scale experiments in the cloud where power management is abstracted away.

Worth exploring

Yes — this addresses a real, expensive problem that every hyperscaler faces. The 30% stranded capacity number comes from actual operational constraints, not marketing fluff. The team combines low-level kernel developers, electrical engineers, and algorithm experts — exactly the mix needed for this problem. The main caveat: they're 6-8 months from operational systems in US data centers, so this is early-stage. But if you're planning data center buildouts or frustrated by power constraints, get on their radar now.

Developer playbook
Tech stack, code snippet, sentiment, alternatives.
PM playbook
Adoption angles, user fit, positioning.
CEO playbook
Traction signals, ROI, build vs buy.
Deep-dive insight
Full long-form analysis, no fluff.
Easy mode
Core idea, fast — when you need the gist.
Pro mode
Technical nuance, edge cases, tradeoffs.
Read the full digest
Go beyond the preview

Deep-dive insight, Easy and Pro modes, plus action playbooks — the full breakdown is one tap away.

Underrated tools. Unfiltered takes.

Read the full digest in the Snaplyze app for deep-dive insight, Easy and Pro modes, and the playbooks you can actually use.

Install Snaplyze →