The Right Approach: The Pros and Cons of PCB Vertical Integration
American Made Advocacy: PCBAA’s Fifth Annual Meeting—Growing Membership and Valuable Relationships
It's Only Common Sense: Why the Best Years of Electronics Are Still Ahead of Us
AI is pushing power infrastructure to its limits, and many of those same challenges, particularly around power delivery and heat management, parallel what the electronics industry faces in building and scaling advanced microelectronics. Yet when it comes to AI and hardware, we typically only see a small piece of the total picture. So, when Schneider Electric invited me to tour a modern AI data center still under construction, I jumped at the opportunity. This became a fascinating technical journey I’m excited to share.
Who Is Schneider Electric?
Schneider Electric is a French multinational company with nearly 200 years of history that began in steel production and heavy machinery. Today, it is a global leader in energy management, industrial automation, and digital infrastructure, operating in more than 100 countries. Its technologies support factories, utilities, buildings, and data centers worldwide, with a growing emphasis on electrification, automation, and sustainability. Schneider Electric has also become a major player in the AI boom, investing heavily in power distribution, automation software, and advanced liquid cooling systems designed for high-performance computing.
While data centers themselves are not new, next-generation AI facilities present entirely new challenges. “AI factories are not traditional data centers; they’re purpose-built to produce digital insight across the AI lifecycle,” said Steve Carlini, chief advocate of data centers and AI. “With 60% of deployments moving to AI and power densities jumping from 20 to 150 kilowatts per rack, this is a fundamental shift in how we design and operate infrastructure.”
Schneider’s portfolio is extensive and its market reach substantial, with the company positioning itself as the world’s largest data center solutions provider across power, cooling, software and services, covering roughly 90% of the end-to-end portfolio needed to support modern data center infrastructure, including AI deployments. Schneider software is used in nine out of 10 data centers worldwide, with much of its business currently concentrated in North America. The company also provides cooling technology for six of the world’s 10 fastest supercomputers, while 95% of hyperscale data centers use Schneider solutions in their facilities and server racks.
Underscoring Schneider Electric’s influence in the AI infrastructure market is its close partnership with NVIDIA, the dominant force in AI computing. Together, the companies have developed AI rack “reference architectures”—detailed, pre-validated blueprints designed to maximize power efficiency, cooling performance, and compute density in next-generation data centers. Schneider makes these designs freely available to companies building AI infrastructure, effectively creating a standardized deployment framework for high-density computing environments. The approach reflects a broader industry shift from managing individual servers to designing entire racks as unified computing platforms, driving fundamentally new approaches to power delivery, cooling, and networking to support dense GPU clusters.
A New Era of Data Centers and AI Factories
I first learned about Schneider Electric at SEMICON last year, where I met the company’s then president of commercial and industrial segments (click here to read the original interview). During our conversation, he discussed the semiconductor industry’s growing power and cooling demands and Schneider’s efforts to address them through technologies such as microgrids and small modular reactors (SMRs). At the time, it served as an introduction not only to Schneider Electric, but also to the broader infrastructure challenges now reshaping AI data centers and high-performance computing.
Pushing Toward Bigger and Better AI Infrastructure
I arrived in Buffalo in time for Schneider’s welcome reception in the company of 31 journalists from around the world. Buffalo’s historic architecture remains remarkably intact, and the Buffalo Marriott at LECOM Harborcenter near Canalside offered a beautiful view of Lake Erie and Buffalo Naval Park. I was pleasantly surprised by the city’s character and industrial heritage.
After a restful night, our press tour began with a full day of presentations. Kevin Brown, senior vice president and chief marketing officer for Schneider’s Secure Power and Data Centers business, emphasized the company’s role in AI data center design, rack architecture, and lifecycle management, specifically citing the addition of APC and AVEVA software solutions for facility operations and rack system maintenance. “There is no other company in our industry that has this end-to-end lifecycle approach,” he said.
Manish Kumar, executive vice president of Schneider’s Secure Power and Data Centers, opened the presentations with an oft-asked question: “How long is this AI thing going to last?” Though the audience laughed, it reflected what many in tech continue to ask.
Kumar recounted a moment with his daughter when he asked her why she had stopped asking him for help with her math homework. “She said, ‘I have ChatGPT now. It is with me all the time. It helps me with my math.’” This was an aha moment for him. The anecdote illustrated how quickly AI has become integrated into our daily lives.
“We are at the beginning of a revolution, an inflection point, and that revolution is intelligence,” Kumar said. “We believe data centers are becoming AI factories.”
Because of this shift, data center design must be approached holistically and much earlier in the process. This discussion echoed similar trends in electronics manufacturing, where earlier collaboration and system co-design increasingly drive better product outcomes and faster development cycles.
Next, Carlini discussed power and grid considerations. Rack density is growing rapidly, he said, roughly doubling every year for high-compute applications: from 3 kW in 2000 to 15 kW today, with projections reaching 150 kW by 2050. This requires specialized intelligent power distribution units (PDUs) and direct high-voltage DC delivery systems to avoid massive electrical distribution losses.
Given that all of this power must be generated from somewhere, the central question remained: Where will all this power come from?
Carlini said the answer lies in two primary schools of thought. One argues that all projected capacity must be built immediately to support future AI growth. Large technology companies, including Amazon, Meta, and Google, are pursuing this strategy aggressively, along with investors hoping to capitalize on the AI boom. Many companies are requesting excess future power capacity to prepare for anticipated AI expansion, leading to inflated projections of total demand.
The second school of thought believes there is still significant untapped capacity within the existing grid and that it should be utilized more efficiently before building entirely new infrastructure—something the company had emphasized in an earlier conversation. Interestingly, Schneider appears to be comfortably operating in both camps. The company works with utilities and industry stakeholders through initiatives such as its Catalyze Program to expand renewable energy adoption, while also supporting new AI data center projects that generate at least some of their own power onsite.
Anuja Ratnayake, an emerging technologies executive at the Electric Power Research Institute (EPRI), described how the organization is working with data center operators as AI-driven power demand accelerates. EPRI’s 70 stakeholders span the full data center ecosystem, including hyperscalers. Today’s data centers typically depend on three primary power sources—the utility grid, localized generation such as natural gas plants or microgrids, and battery backup systems—and EPRI’s DC Flex initiative is aimed at helping operators balance those resources more dynamically. Rather than treating flexible power management as a risk, EPRI is encouraging operators to see it as a path to greater reliability and efficiency.
Now That’s Hot!
The balance of the morning’s presentations focused on managing the thermal challenges inside modern AI server racks. The Buffalo facilities are dedicated entirely to Motivair liquid cooling technology, best known for its ChilledDoor® rear-door heat exchangers and expanding direct-to-chip cooling portfolio. Schneider also operates cooling manufacturing facilities in Italy, India, China, and the United States. Richard “Rich” Whitmore, CEO of Motivair by Schneider Electric, explained some of the company’s technology in detail, stating, “The unprecedented growth of AI is pushing computing power to new limits, and air cooling simply can’t keep up. Liquid cooling is the only way to efficiently, reliably, and sustainably run the chip and server technologies required for AI.”
One surprising takeaway was that liquid cooling is both more energy- and water-efficient than traditional air cooling. It was equally surprising to learn how much water conventional air-cooling infrastructure consumes. In closed-loop designs, liquid cooling can sharply reduce facility water use compared with some traditional air-cooling approaches, depending on the site and cooling architecture.
Of course, this prompted an obvious question: Why wasn’t liquid cooling adopted sooner?
Until about four years ago, traditional air cooling was sufficient for conventional CPU server heat loads, with existing data centers designed around those systems. Those facilities represent enormous infrastructure investments, leaving little incentive to absorb the cost and complexity of transitioning to liquid cooling—at least until the AI boom dramatically increased power densities and thermal demands.
But even now, air cooling continues to play an important role, and this is not likely to change. In many AI data centers, liquid cooling systems are designed to remove roughly 70–80% of the heat load, while the remaining thermal management is handled through conventional air-cooling infrastructure. As AI workloads continue to grow, operators now rely on hybrid cooling strategies that combine both technologies for optimum efficiency and performance.
After a full day of presentations and facility tours, the evening concluded with interviews, including a conversation with Andrew Bradner, senior vice president of Schneider Electric’s cooling business, followed by dinner at a well-known downtown restaurant. The day was full and the experience rich. Everyone enjoyed the evening’s networking and camaraderie, excitedly anticipating the AI data center tour of TeraWulf the following day.
I look forward to sharing my interview with Andrew Bradner, along with additional insights from my AI data center tour experience in the July issue of I-Connect007 Magazine.
