AI-Powered Digital Twin Paves Way for Wireless Data Centers

Data centers are the warehouses of the internet, full of servers that power everything from streaming to AI chatbots. Right now, they rely on piles of cables to move data between machines. But cables are rigid, expensive to install, and hard to reroute when workloads change. This paper explores a clever alternative: using terahertz waves—extremely fast wireless signals—to connect servers in a room, no cables needed.

To test whether this could work, the researchers created a "digital twin." That's a virtual copy of a real data center where engineers can experiment without touching physical hardware. The twin was built from actual measurements taken at three super-high frequencies (140, 220, and 300 gigahertz). They then trained an AI model to instantly predict signal strength and whether a clear line of sight exists between two points. This means planners can test different layouts and antenna placements in seconds, not weeks.

The results are promising. The AI model could predict signal levels faster and more accurately than earlier methods. In a simulated data center with access points mounted on the ceiling, over 90% of the room received a strong enough signal for reliable communication—even with interference from other equipment. That is a key finding, because it suggests you don't need a transmitter for every rack; a few well-placed ceiling units could cover an entire aisle.

Why does this matter to you? Data centers are the backbone of cloud computing and AI. Making them wireless could cut construction costs, reduce downtime when expanding, and let operators rearrange servers on the fly to match demand. That could mean cheaper cloud services and faster rollouts of new AI features. The catch: this is early research. Terahertz signals are easily blocked by walls and even people, so real-world deployment will need careful design. Still, this digital twin approach is a big step toward making wireless data centers a practical reality.