You are reading the May 2026 issue of EAR’s monthly guide to energy-aware HPC and AI. This month, we address the reality of power-constrained infrastructure and the transition from managing 'compute nodes' to managing 'energy flows.'
It was great connecting with the community at the HPC User Forum in Austin this month. Our GM, Raphaël Brochard, found it an excellent event that highlighted a major shift: we are moving past "growth at all costs" toward an era of disciplined, energy-aware scaling.
A recurring theme was the struggle to balance peak performance with rising energy costs. Manual power capping is no longer enough; the industry now requires automated, runtime optimization.
Success is no longer measured solely by peak FLOPS, but by the "science per watt" a system can sustain. As power envelopes become the ultimate ceiling for innovation, intelligent orchestration has evolved from a niche optimization into a foundational necessity for the modern data center.
Is this a challenge you’re currently navigating? EAR is already proven at scale, optimizing 23+ MW of power across 21,562+ servers for our partners all around the globe. If you'd like a technical brief or start your free 30-day trial of our system software, simply reply to this email.
In the meantime, read on for the latest HPC and AI industry news and team updates we’ve curated for you below.
Watt’s the Plan to Run More Compute?
Infrastructure & Global Strategy News
AI Continent Action Plan
The European Commission’s AI Continent Action Plan reached its first-anniversary milestones, successfully accelerating sovereign AI infrastructure by transforming Europe’s traditional industries into engines of sustainable innovation. (Source: EC)
Europe's AI Factories Hit Power Allocation Hurdles
As EuroHPC expands its network, grid constraints are causing multi-year queues. With shared clusters averaging just 30–70% GPU utilization, the study urges a pivot from buying hardware to runtime optimization to avoid massive energy liabilities. (Source: Interface)
From PUE to AI-per-Watt
For twenty years, PUE measured the "container" (the building). In the era of massive LLMs, the industry is shifting to Tokens per Watt—a metric that connects physics to economics by asking how much "thought" each unit of energy produces. (Source: Oscar Rojas/LinkedIn)
Power Deliverability: The New Constraint for US AI
Power access, not demand, now limits AI growth. Despite 30GW of capacity, grid bottlenecks in Virginia and Ohio are stalling projects regardless of funding. As the race shifts from chips to grid access, on-site orchestration is the only way to bridge the "deliverability gap." (Source: Rabobank)
Catch EAR at these upcoming events!
🇧🇪 EIC Summit 2026 | Brussels, BE | June 3–4, 2026
🇩🇪 ISC High Performance 2026 | Hamburg, DE | June 23–25, 2026
We're exhibiting at ISC 26. Find us at Booth C11 to see EAR in action!
🇮🇪 EuroHPC User Days 2026 | Dublin, IE | Sept 22–24, 2026
Hear from the Team
Meet Marco D'Amico, Computer Scientist at EAR
Marco D'Amico is a Computer Scientist and Engineer collaborating with EAR and contributing to the European-funded DYMAN project. His work focuses on research and development in dynamic power, temperature, and cooling monitoring and control for HPC systems. He supports the development of energy-aware runtime solutions aimed at improving efficiency, reliability, and sustainability in large-scale HPC environments.
Q: How did you first get into HPC/AI/data center technology?
A: When I won an Erasmus position at Universitat Politécnica de Catalunya in Barcelona, I was fascinated by the research-oriented study path, and I learned about Barcelona Supercomputing Center and Marenostrum. My question, which became my PhD research question, was: how to coordinate those thousands and thousands of computing units? I worked at BSC and got my PhD under the wings of Julita Corbalán, CTO at EAR.
Q: What’s the most interesting technical challenge you’ve worked on lately?
A: Modelling and simulating extremely complex problems, like job scheduling, cluster power management, and cooling systems. It is a great challenge which requires skills, knowledge, and a lot of patience.
Q: How do you see EAR making a difference for sustainability in tech?
A: The EAR team mixes knowledge at different layers of the data center infrastructure, from the low-level architecture to the high-level management, and it has great leadership with long experience in the field.
Q: Coffee, tea, or…? What fuels your coding sessions?
A: Shen pu’er tea from the 90’s, spiritual tea which can warm up your body, open up your awareness, favour a calm and present mind.
🎧 Podcast Spotlight:
"Cutting GPU Energy Use by 20%: Raphaël Brochard on How EAR Makes HPC and AI Data Centers Energy-Aware"
Your data center might be wasting 20% of its GPU energy right now. Not because of bad hardware, but because nobody’s watching what actually happens at runtime.
In the latest episode of Scaling Intelligence, Kevin Jackson sits down with Raphaël Brochard, GM of EAR, to explore how EAR—a spinoff from the Barcelona Supercomputing Center—acts as the "central nervous system" for cluster energy.
Real-time Middleware: Vendor-agnostic; works with Slurm and Kubernetes.
The "Invisible" Inefficiencies: From MPI bottlenecks to jobs requesting 12 GPUs when they only need 6.
Hard Limits: How LRZ uses EAR to slash SuperMUC-NG’s peak consumption from 3MW to 2MW.
Testing EAR is entirely frictionless: our system software requires zero hardware replacement, allowing you to easily characterize a targeted subset of nodes against your specific production workloads to generate a quantified, data-backed ROI savings report.
☑️ Stop paying for idle silicon: Reclaim up to 20% of wasted GPU energy by automatically optimizing nodes during runtime bottlenecks.
☑️ Bypass the grid queue: Safely squeeze 10% to 15% more compute out of your existing power envelope, delaying costly infrastructure expansions.
☑️ Automate your compliance: Get instant, application-level power and CO2 analytics to satisfy strict new reporting mandates without wasting engineering hours.
Reply directly to this email to book a brief call with our team, where we can map out your path forward with a Technical Workshop Demo, a Data-based ROI & Review, or a tailored Pilot/Trial Scope designed around your cluster's unique constraints.
