Original Article: https://www.datacenterknowledge.com/build-design/pado-vessl-bring-energy-aware-ai-to-data-center-power-crunch
Author: Shane Snider
As AI infrastructure demand accelerates in 2026, a partnership between Pado and Vessl announced in January is drawing renewed attention for its attempt to tackle one of the industry’s biggest constraints: power.
The collaboration combines Pado’s grid-aware compute orchestration with Vessl’s MLOps platform to align AI workloads with real-time energy conditions, pricing, and renewable availability – while preserving performance and revenue goals, the companies say. Together, the approach layers energy intelligence onto workload orchestration across hybrid and multi-cloud environments without requiring major infrastructure changes.
Since January, the effort has offered a window into a broader shift: energy is becoming a critical variable in how AI workloads are scheduled and scaled.
Pado, an LG spinout company, develops software that uses real-time grid data, pricing signals, and infrastructure telemetry to optimize when and where compute runs. Vessl provides a full-stack MLOps platform for deploying and managing AI workloads across on-premises, hybrid, and multi-cloud environments. The combined stack aims to route jobs to times or locations where power is cheaper or more abundant, potentially reducing costs, easing grid strain, and improving renewable utilization as deployments scale.
Pado CEO Wannie Park said the goal is not a traditional efficiency play. “Flexibility and energy savings are not really top of mind for data center operators,” Park told Data Center Knowledge. “The opportunity cost of not using available power is too high. What we’re focused on is maximizing compute – not minimizing consumption.”
Park added: “If you can deliver the same performance more efficiently, that’s where flexibility and efficiency start to align."
Capacity Limits Remain
“I think this partnership is interesting and on paper their product makes sense,” Vladimir Galabov, senior director, enterprise infrastructure, at Omdia, told Data Center Knowledge. “I am worried that in practice, it misses a few aspects.”
The biggest constraint: available compute. “Most GPU clusters today have high utilization,” Galabov noted. “Even if energy is available in one region, there are very few unused GPUs to take advantage of that.”
Park said utilization varies by segment. “We’re not targeting hyperscalers running at near-perfect efficiency,” he said. “In the midmarket, GPU utilization is often closer to 30%-40%. Our goal is to push that toward 60% without impacting SLAs.”
That flexibility, Park added, comes from better workload forecasting and prioritization. “There’s a kind of ‘duck curve’ emerging in AI workloads,” Park said. “We can shift lower-priority jobs or burst them into the cloud while keeping high-value workloads on-premises.”
Analysts Split on Real-World Impact
For Andy Lawrence of Uptime Institute, the concept targets a real problem – but adoption may lag.
“My first instinct is this is absolutely a problem that needs to be addressed,” Lawrence told Data Center Knowledge. “Using analytics to model workloads, grid stability, and energy costs is the right approach – it’s a big data problem.”
But similar efforts have struggled to gain traction.
“Technically, solutions like this can be very effective,” Lawrence said. “The challenge is that business and IT decision-makers don’t usually prioritize energy or grid considerations. ... If I’m running a large AI training cluster, I’m not sure I’m going to start rescheduling workloads based on energy pricing. ... People focus on getting the job done.”
Even when the data exists, it often doesn’t change behavior, he added.
Still, Lawrence sees potential if optimization can happen in the background. “If this works unobtrusively, without impacting performance or users, it becomes compelling,” he said. “But the proof is in how well that actually works.”
Energy Resilience vs. Orchestration
Meanwhile, operators are also tackling power constraints directly.
“The pipeline of battery energy storage systems capable of running data centers off-grid for hours is astounding,” Galabov said.
Park views storage and orchestration as complementary. “Storage, grid interaction, and orchestration all work together,” he said. “We’re optimizing across uptime, sustainability, and cost based on real-time data.”
Sovereignty and Geographic Limits
Regulatory and geopolitical factors further limit flexibility.
“The sovereign AI movement is very strong,” Galabov said. “Many organizations don’t want workloads leaving a country or region.”
“You can restrict workload location today,” Lawrence added. “But that reduces how much flexibility you can actually use.”
Park said the platform is designed to operate within those constraints. “This isn’t about moving everything everywhere,” he said. “It’s about optimizing within existing environments.”
Why It Still Matters
Power availability is increasingly shaping where AI infrastructure can scale.
In key markets, the gating factor for new capacity is access to electricity, not land or capital. That shift is forcing operators to rethink how they use infrastructure.
Energy-aware orchestration introduces flexibility into a system that has historically been rigid.
DCK Analysis
The Pado-Vessl partnership highlights an inflection point: energy is becoming a scheduling input for AI infrastructure.
But the gap between concept and reality remains. Tight GPU supply, growing on-site energy resilience, and operational priorities that favor performance limit how much flexibility operators can exploit today. That tension defines the opportunity.
If capacity expands and automation matures, energy-aware orchestration could unlock meaningful efficiency gains. If not, it risks remaining a marginal optimization layer in a market still defined by a more immediate constraint: access to GPUs.