AI’s next leap demands a revolution in data-center wiring: copper’s limits are being crushed by GPUs’ hunger for speed, forcing startups like Point2 and AttoTude to deploy radio tech as the only viable path to connect thousands of chips without melting servers or bankrupting budgets.
The race to train massive AI models isn’t just about faster GPUs—it’s about how fast you can move data between them. And physics is already winning this battle against copper.
Scaling out AI systems means linking thousands of computers together. Scaling up means cramming dozens—even hundreds—of GPUs into each machine. The latter demands denser, faster connections than ever before. Copper, once the backbone of data centers, now faces an existential threat from its own success.
“We call it the copper cliff,” says David Kuo, vice president of product marketing at Point2 Technology. As Nvidia plans to increase GPU counts per system from 72 to 576 by 2027—a jump eightfold—data centers face a brutal reality: copper simply can’t handle terabit-per-second bandwidth over even short distances without becoming exponentially thicker, heavier, and thirstier for power.
Physics fights back with the skin effect. At high frequencies, electrical currents migrate toward the surface of copper wires, increasing resistance dramatically. What takes a thin wire to carry 10 gigabits per second becomes a thick bundle of metal needing twice the voltage—and consuming three times the energy—to carry 100 gigabits per second. That’s not scalable. It’s catastrophic.
Enter two startups betting everything on radio waves—not fiber optics, not fancy silicon photonics—but simple, cheap, reliable radio to bridge the gap between GPUs. Point2 and AttoTude aren’t trying to replace copper entirely. They’re trying to extend its reach beyond what’s physically possible—and do it cheaper, quieter, and with less heat.
Point2’s solution is called the e-Tube. A single chip converts digital data into modulated millimeter-wave signals, then funnels them down eight polymer waveguides. Each waveguide carries 448 gigabits per second using dual frequencies (90 GHz and 225 GHz). Together, they deliver 1.6 terabits per second over distances of 10 to 20 meters—with one-third the power consumption and one-thousandth the latency of optical fiber.
AttoTude operates similarly but targets higher frequencies—terahertz range—and uses flexible dielectric cables. Their prototype demonstrates 224 Gb/s transmission over 4 meters at 970 GHz, a distance ideal for connecting GPUs within racks. Both companies claim their tech beats copper in every critical metric: cost, power, density, and reliability.
But here’s where things get interesting: both startups envision radio replacing copper entirely—not just extending it. “You start with passive copper,” says Don Barnetson of Credo, “and you do everything you can to run in passive copper as long as you can.” That philosophy still dominates data centers today. Liquid cooling exists precisely to keep GPUs cool while running on copper. But when scaling up reaches densities where air cooling fails, radio offers a cleaner, more efficient alternative.
Radio also solves the biggest problem plaguing optical interconnects: manufacturing complexity. Optical fibers require micron-level precision to align lasers with tiny cores. Radio signals operate at much longer wavelengths, making alignment far easier—even hand-assembled in early prototypes.
And if you think radio sounds gimmicky, consider this: Nvidia and Broadcom spent years developing co-packaged optics—tiny optical transceivers embedded directly inside GPUs—to avoid the headaches of fiber-to-chip connections. Point2 and AttoTude argue their radio solutions could cut that engineering time in half—or even eliminate it altogether.
“One of the many challenges is how to attach an optical fiber to a waveguide on a photonic chip with micrometer accuracy,” says Kuo. “By contrast, millimeter-wave and terahertz signals have a much longer wavelength, so you don’t need as much precision to attach the waveguide.” In one demo system, it was done by hand.
Still, skeptics abound. Copper has decades of infrastructure behind it. Companies like Molex and Foxconn Interconnect Technology have already pledged support for Point2’s e-Tubes. That’s no small thing. If hyperscalers adopt this tech, it will redefine how data centers are built—and who controls the future of AI hardware.
For users, this shift means fewer crashes caused by overheating GPUs, lower electricity bills, and faster training runs. For developers, it opens doors to building AI models that previously required entire server rooms just to handle internal communication. For investors, it’s a clean bet on a technology that doesn’t need expensive optics or exotic materials—it just needs better electronics.
“Customers love fiber,” admits Dave Welch, CEO of AttoTude. “But what they hate is the photonics.” Electronics may be inherently more reliable than optics—and radio is pure electronics.
The copper cliff is real. And the only way to climb past it is to stop fighting gravity—and start flying.
Stay ahead of the curve. Read more breaking tech analysis exclusively on onlytrustedinfo.com.
