CFOtech US - Technology news for CFOs & financial decision-makers
United States
RAVEL & Sandia launch AI workload research project

RAVEL & Sandia launch AI workload research project

Wed, 18th Nov 2026 (Today)
Joseph Gabriel Lagonsin
JOSEPH GABRIEL LAGONSIN News Editor

RAVEL has begun a research collaboration with Sandia National Labouratories on Project Quicksilver, which will study how AI workloads can be managed across geographically distributed computing environments.

The initiative will use RAVEL Orchestrate AI in a multi-phase programme focused on dynamic, power-aware orchestration. The organisations said the research will examine how power availability, operating conditions and workload requirements shape scheduling and placement decisions in a distributed setup.

The work comes as operators face pressure to expand AI computing while managing electricity supply, infrastructure spending and regulatory limits. Project Quicksilver is intended to test whether more flexible workload placement can make better use of existing compute and power resources, rather than relying only on new buildouts.

Phase one will evaluate air-cooled and immersion-cooled computing environments across three Sandia locations linked by a wide-area network. Tests will vary factors including network bandwidth and speed, using Box-Behnken and Plackett-Burman experimental design methods.

The first phase will focus on three areas: energy efficiency, total cost of ownership and workload time-to-completion. This will include measuring energy consumption and operating costs for identical workloads in different cooling environments, alongside analysing capital and operating costs related to power, cooling, networking and operations.

Researchers will also examine how scheduling and placement decisions affect the time needed to execute AI workloads. The goal is to compare trade-offs across different environments rather than assess a single data centre in isolation.

RAVEL said its software acts as a federated orchestration layer for AI workloads running across heterogeneous infrastructure in different locations. In practice, workloads can be scheduled and moved across sites as part of one coordinated system, with decisions informed by factors such as available power, cost and performance requirements.

That approach reflects a broader shift in AI infrastructure planning. As data centre operators spread capacity across multiple facilities and regions, the challenge has moved beyond adding hardware to deciding where workloads should run under changing energy and network conditions.

"The next phase of AI infrastructure leadership won't be won by simply building more. It will be determined by how efficiently and productively we use every megawatt, every GPU, and every dollar invested," said Denise Muyco, Chief Executive Officer and Co-founder of RAVEL.

"The United States has an opportunity to continue to lead and innovate not just in AI capacity, but in how intelligently that capacity is managed. Project Quicksilver will investigate how orchestration can help make AI infrastructure investments more productive while supporting more responsible energy use," Muyco said.

Sandia described the project as a way to explore the operational trade-offs created by multi-site computing. The laboratory said distributed environments can create opportunities for better resource use, but also add complexity when balancing performance, cost and available power.

"Many laboratories and data centers around the world look to Sandia's work in developing and deploying energy-efficient designs and water-saving technologies," said Dave Martinez, Engineering Program and Project Lead for Infrastructure Computing Services at Sandia National Labouratories.

"Managing workloads across multiple computing environments creates opportunities, while also introducing complex trade-offs. 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," Martinez said.

Research focus

The organisations said the results could help inform future data centre design and operating models, particularly where sites differ in cooling systems, energy conditions or network characteristics. The work also aims to produce evidence on whether orchestration policies can reduce waste and improve economic outcomes in distributed AI estates.

Project Quicksilver begins with Sandia as the initial research lead. The partners plan to expand the initiative to additional US Department of Energy national laboratories and investigate other environmental and performance-related use cases.

RAVEL Chief Product Officer Philippa Carroll linked the work to a more federated model of infrastructure management. "AI infrastructure can't reach its full potential if every data center operates as an island," she said.

"As AI infrastructure becomes more distributed, the challenge is no longer just how much computing capacity we have, but how effectively we can use it. Project Quicksilver will investigate the trade-offs between power, cost and performance, and the role orchestration plays in helping us make better and more responsible infrastructure decisions," Carroll said.