RAVEL to Test Power-Aware AI Scheduling at Three Sandia Sites

By Grace Kim

RAVEL announced on November 12, 2026, that Phase One of Project Quicksilver will test power-aware scheduling of AI workloads across three Sandia National Laboratories locations, comparing air-cooled and immersion-cooled systems connected by a wide-area network.

RAVEL said the multi-phase research collaboration with Sandia will examine whether power availability, operating conditions and workload requirements can guide scheduling and placement decisions to address energy, capacity and cost constraints across geographically distributed computing environments. The goal, according to the announcement, is to inform data-center design and operations by making power and computing capacity more productive, rather than assuming extra capacity automatically creates more useful infrastructure.

The announcement describes planned measurements, not completed findings or demonstrated savings.

What will Phase One compare?

The three-site experiment will evaluate two cooling environments across Sandia locations joined by a wide-area network. RAVEL and Sandia will use a Box–Behnken experimental design for the air-cooled environments and a Plackett–Burman design for the immersion-cooled environments. Network bandwidth and speed will vary alongside other experimental factors.

The energy-efficiency analysis will compare identical workloads across the cooling environments. It will account for compute, IT infrastructure and data-center operations, rather than measuring computing equipment in isolation. Measurements will include:

• Energy use
• TUE
• PUE-adjusted efficiency
• Potential heat-reuse benefits

RAVEL CEO and co-founder Denise Muyco said the project will investigate how orchestration can make AI infrastructure investments more productive while supporting more responsible energy use.

The practical takeaway is that Phase One will judge workload placement against facility and cooling overhead, not computing equipment in isolation.

How will Orchestrate AI manage distributed workloads?

RAVEL describes Orchestrate AI as a federated orchestration layer for managing AI workloads across heterogeneous, geographically distributed resources. Its intended model treats multiple locations as one system rather than optimizing each environment independently.

Power-aware scheduling and policy-driven management will consider:

• Available power
• Cost
• Performance requirements
• Infrastructure conditions
• Workload needs
• Operational priorities

In short, Orchestrate AI is intended to make placement decisions against shared infrastructure constraints, not isolated site-level goals.

How will the study measure cost and completion time?\n

A separate total-cost-of-ownership analysis will assess how scheduling and placement affect computing costs across distributed environments. It will cover capital and operating expenditures tied to infrastructure, power, cooling, networking and operations.

A third workstream will measure AI workload time-to-completion. Researchers will place execution time alongside energy consumption and cost to examine the trade-offs among those objectives. Moving or rescheduling a workload may affect more than its power draw: the test also tracks networking and operating costs, cooling environments and execution time.

Dave Martinez, engineering program and project lead for Infrastructure Computing Services at Sandia, said: “Under a Cooperative Research and Development Agreement with RAVEL, we’ll explore whether dynamic workload placement can make better use of available power and computing resources while balancing performance and cost.”

The combined measurement plan turns Project Quicksilver into a test of distributed computing economics, rather than a power-efficiency study alone. It will show whether coordinated placement can improve resource use without weakening workload performance.

What comes after the three-site test?

Sandia is the initial research lead. RAVEL and Sandia plan to expand Project Quicksilver to additional U.S. Department of Energy national laboratories and explore further environmental and performance-optimization use cases.

RAVEL and Sandia are scheduled to discuss the experimental approach, planned test phases and anticipated research outputs at the Department of Energy booth during SC26 on November 17, 2026.

According to the project announcement, Project Quicksilver’s central question is whether separate data centers can operate as a coordinated system to improve energy use and computing economics without undermining workload performance. The November 17 session is the next scheduled checkpoint, and any claim of savings or performance gains will still require the study’s results.