
Dive Brief:
- Schneider Electric and power system design and operations firm ETAP on Tuesday released digital twins that can design and simulate the power needs of AI factories, the companies said in a release shared with Facilities Dive. A digital twin is a system replica that enables users to simulate operations to improve performance in real time.
- By leveraging ETAP’s integration of NVIDIA Omniverse technologies, the companies said, they were able to develop digital twins that bring together multiple inputs for mechanical, thermal, networking and electrical systems to mirror how an AI factory operates.
- NVIDIA introduced its Omniverse Blueprint for AI factory design and operations Tuesday at its annual conference, GTC. The companies say the collaboration will provide enhanced insight and control over the electrical systems and power requirements at AI factories. That presents “an opportunity for significant efficiency, reliability and sustainability gains,” Schneider Electric said.
Dive Insight:
AI workloads such as large-scale training clusters and edge inference servers are driving a significant increase in data center power consumption, requiring substantial computational power that has led to higher rack power densities, the companies say. This has led data center capacity proposals to balloon, with the average proposed center doubling in size from about 150 MW in early 2023 to 300 MW in mid-2024, according to an Oct. 24 report from Wood Mackenzie.
The demand is giving AI-related occupiers increasing influence over data center development decisions like site selection, design and operational requirements, CBRE said in its North America Data Center Trends H2 2024 report, released Feb. 26. These occupiers are “prioritizing markets with scalable power capacity and advanced connectivity solutions,” the commercial real estate services firm said.
Power availability remains the top priority for data center developers looking at greenfield sites, according to CBRE’s report.
“Startups, enterprises, colocation providers, and internet giants must rethink data center design and management to address the growing need for power efficiency,” the companies said. The ETAP and NVIDIA collaboration introduces a “grid to chip” approach that addresses the critical challenges of power management, performance optimization and energy efficiency associated with AI, they said.
While basic visualization of electrical systems was previously possible, ETAP said, its integration with NVIDIA’s Omniverse technologies enables the digital twin model where “multiple dynamics interact seamlessly.” As a result, instead of only estimating average power consumption at the rack levels, data center operators will be able to increase precision on modeling dynamic load behavior at the chip level to improve power system design and optimize energy efficiency, per the release.
Through this sophisticated modeling, ETAP can create a virtual replica of a data center’s electrical infrastructure that can be combined with real-time power system data, advanced analytics and insights to provide “unprecedented insights.”
Schneider Electric and ETAP said the insights span:
- Advanced electrical design and simulation
- Dynamic “what-if” scenario analysis
- Real-time electrical infrastructure performance tracking
- Advanced energy efficiency optimization
- Predictive maintenance and system reliability assessment
- Infrastructure needs based on power usage that can help reduce total cost of ownership
While the January announcement of highly efficient, open-source AI platform DeepSeek led some to question just how much infrastructure — like high-performance chips, new power plants and electrical equipment — the AI boom actually requires, many experts believe that AI efficiency gains will be a net positive for data center developers, operators and customers in the long run.
A reduction in the computing power needed to train new models could eventually tip the balance of new data center deployments toward smaller “inference” data centers, which serve user requests rather than model development and refinement, experts told Facilities Dive.
The news adds to ongoing AI-focused research and development by Schneider Electric, which earlier this month officially opened a data center and microgrid laboratory at its global research and development center in Andover, Mass. The facilities include a 6,000 square-foot, three-bay lab to test high-voltage power systems for AI data centers and will help the company meet consumer demand for data for digital services, Schneider Electric Executive Vice President for Data Centers & Networks Pankaj Sharma said in a statement.
In December, Schneider Electric also introduced a data center reference design in partnership with NVIDIA that is optimized for the chipmaker’s high-computing-power chips and can support liquid-cooled, high-density AI clusters using up to 132 kilowatts of power per rack, the company said when the partnership was announced. The design offers options for both liquid-to-liquid coolant distribution units and direct-to-chip liquid cooling systems amid expectations that both methods will play a role in future data center operations, experts say.