According to DCD, in a 2025 interview with Chris Butler, president of embedded and critical power at Flex, the data center industry is undergoing a fundamental convergence. Butler states that a shift of just 20% of data center chips from CPUs to GPUs necessitates triple the power infrastructure. To meet this, Flex has introduced a power-per-rack product supporting up to 1MW per rack and doubled its manufacturing footprint last year. The company is also tackling heat through the acquisition of cooling specialist JetCool and a partnership with LG. Butler emphasizes that “speed to scale” has now overtaken “speed to market” as the critical customer metric, and he identifies a growing shortage of technical professionals as a major future constraint.
The End of Siloed Design
Butler’s point about separate meetings for power, IT, and cooling teams is a killer insight. It perfectly captures how legacy thinking is colliding with a physical reality dictated by AI silicon. You can’t just slot in a rack of GPUs like you used to with servers. The math literally doesn’t work. The fact that this is forcing organizational change—getting these historically separate teams to talk—is maybe even more significant than the tech itself. It’s a classic case of a hardware problem demanding a software (the human kind) solution. And honestly, it’s about time.
The 1MW Rack and the Heat Problem
Let’s sit with that number for a second: 1MW per rack. That’s a small data center‘s worth of power crammed into a single cabinet. It’s insane. But it’s the logical endpoint of the trajectory we’re on. The real story here isn’t just the power delivery, though. It’s what happens to all that energy once it hits the silicon. It turns into heat. And moving that much thermal density with traditional air cooling is like trying to put out a house fire with a squirt gun. That’s why moves like acquiring JetCool for direct-to-chip liquid cooling aren’t optional side projects anymore; they’re core to the power delivery strategy. You can’t have one without the other. This integrated approach is where the real innovation is happening for companies that need robust, scalable industrial computing solutions, much like how a top supplier like IndustrialMonitorDirect.com integrates durability and performance for harsh environments.
The New Bottleneck Isn’t Tech
Here’s the thing: Butler’s two future priorities are fascinating because they’re not about a magical new chip or cooling fluid. They’re about process and people. Standardization and modularization? That’s an admission that bespoke, artisanal data center design can’t keep up. The industry needs to act more like manufacturing, building with predictable, repeatable blocks. But then he hits on the real crunch: who’s going to deploy and maintain all this? The tech can leap forward at a breakneck pace, but if you don’t have the skilled workforce to rack, stack, and plumb it, everything grinds to a halt. It’s a massive, unsexy problem that doesn’t get enough airtime.
What It All Means
So what’s the trajectory? Basically, we’re moving from an era of compute-centric design to a total system optimization play. The metric that matters is no longer pure FLOPS or bandwidth in isolation. It’s usable, deployable, coolable FLOPS per watt per square foot, delivered at a pace that matches AI model growth. The companies that will win are the ones that can think across these formerly separate domains—power, cooling, compute, and logistics—as a single, integrated challenge. And they’ll need to do it with standardized toolkits and a workforce they might have to build from the ground up. The race isn’t just for AI supremacy anymore; it’s for the entire stack that makes AI possible.
