Nvidia's upcoming Rubin Ultra GPUs will push data center racks to a staggering 600 kilowatts by 2027 - nearly twice the power of today's fastest EV chargers. As AI infrastructure hits this thermal wall, Alloy Enterprises thinks the answer lies in revolutionary metal stacking technology that turns copper sheets into seamless cooling plates capable of handling extreme pressures and tight spaces.
The AI industry just hit a cooling reality check. When Nvidia announced its Rubin series GPUs in March, the company quietly dropped a bombshell: racks built with the Ultra version, expected in 2027, will consume up to 600 kilowatts of electricity. That's more power than most EV fast chargers can deliver.
The physics are unforgiving - all that electricity becomes heat, and traditional air cooling can't handle it. Enter Alloy Enterprises, a startup betting that stacks of precisely bonded metal sheets can solve AI's thermal crisis.
"We didn't care too much about that 20% when racks were 120 kilowatts," Ali Forsyth, co-founder and CEO of Alloy Enterprises, told TechCrunch. She's talking about the peripheral components - memory modules, networking chips, and support hardware that account for a fifth of a server's heat load. "But now, as racks have hit 480 kilowatts on their way to 600 kilowatts, engineers have to figure out how to liquid cool everything from RAM to networking chips, parts for which there are no solutions available today."
The startup's breakthrough centers on what it calls "stack forging" - a process that transforms sheets of copper into solid cooling plates without the traditional weaknesses of machined or 3D-printed alternatives. Unlike conventional cold plates that require machining two halves separately and fusing them together (creating potential leak points), Alloy's technique bonds metal sheets using precise heat and pressure to create seamless, solid copper blocks.
The technical advantages are significant. Stack forging can create cooling channels as narrow as 50 microns - about half the width of human hair - allowing more coolant to flow through the metal. According to Forsyth, this precision translates to 35% better thermal performance than competing solutions.
"We hit raw material properties," Forsyth explained to TechCrunch. "The copper is just as strong as if you had machined it," but without the structural compromises of traditional manufacturing.
The process itself resembles high-tech origami. In Alloy's factory, copper sheets are laser-cut with precise features, then coated with inhibitors in areas where bonding isn't wanted. Each slice gets registered and stacked before entering a diffusion bonding machine that applies controlled heat and pressure, fusing the layers into a single piece of metal.
This matters because the industry is scrambling to keep pace with AI's exponential power demands. Data center operators who once worried about 120-kilowatt racks now face the prospect of 600-kilowatt monsters that generate enough heat to melt traditional cooling infrastructure. Every component - not just the headline-grabbing GPUs - needs liquid cooling solutions that simply don't exist today.
The timing couldn't be better for Alloy. Initially, the company designed its technology around aluminum alloys, but pivoting to copper proved prescient as data centers demanded better heat conductivity and corrosion resistance. When Alloy announced its copper-based product in June, "things just blew up," Forsyth said.
While Forsyth won't name specific customers, she confirms the company is working with "all the big names" in the data center world. That likely includes the usual suspects - Amazon, Microsoft, Google, and Meta - all racing to build AI infrastructure that can handle next-generation workloads without melting down.
The broader implications extend beyond just keeping servers cool. As AI models grow larger and more complex, the infrastructure supporting them becomes increasingly critical. Companies investing billions in AI capabilities need assurance their hardware won't throttle or fail when pushed to the thermal limits.
Alloy's approach is more expensive than traditional machining but cheaper than 3D printing, positioning it in a sweet spot for high-performance applications where reliability matters more than cost. The startup handles most design work internally, using proprietary software to translate customer specifications into shapes optimized for stack forging.
This represents a fundamental shift in how the industry thinks about cooling. Rather than treating thermal management as an afterthought, companies like Alloy are making it central to AI infrastructure design. The race isn't just about building faster chips - it's about building systems that can actually run those chips at full capacity without overheating.
As AI infrastructure races toward 600-kilowatt power consumption, thermal management has become the industry's next critical bottleneck. Alloy Enterprises' stack forging technology represents more than just better cooling - it's a glimpse into how manufacturing innovation will determine which companies can actually deliver on AI's computational promises. The startups and tech giants that solve these fundamental infrastructure challenges won't just keep the lights on; they'll define who leads the next phase of the AI revolution.
