3.2 Tb/s AI Data Center Optical Interconnects [Explained]

Ever try to push a golf ball through a garden hose? That is essentially what modern AI data centers are attempting to do right now.

As artificial intelligence models grow exponentially, the traditional copper wiring connecting massive clusters of compute power is choking on the sheer volume of information. The result? A big data bottleneck that threatens to slow down the entire AI buildout. But instead of accepting the limits of copper, the biggest names in the tech industry are teaming up to swap out that hose for a high-speed optical pipeline.

What is the Optical Compute Interconnect Multi-Source Agreement (OCI MSA)?

So, who is stepping up to solve this massive infrastructure headache? According to reports, an alliance of industry heavyweights has officially formed the Optical Compute Interconnect Multi-Source Agreement, or OCI MSA for short.

This coalition includes some of the fiercest competitors and biggest spenders in the tech world: AMD, Nvidia, Microsoft, Broadcom, and Meta. Instead of working in isolated silos, these titans are joining forces to develop a standardized optical interconnect specifically designed for the unique demands of modern AI data centers.

Why are AI data centers moving away from copper wiring?

You might be wondering, what exactly is wrong with the copper cables we have used for decades? Simply put, copper has hard physical limitations. When it comes to the blistering speeds required to train and run massive AI workloads, copper creates a severe data bottleneck.

It simply cannot move information fast enough between components without degrading the signal or generating too much heat. By shifting from electrical signals over copper to light signals over optical connections, data centers can push vastly more data over longer distances with incredible efficiency.

How fast will the new optical PHY standard be?

The ultimate aspiration of the OCI MSA is incredibly ambitious. The group has set its sights on building a new PHY—the physical layer of the network connection—that is capable of handling staggering speeds of up to 3.2 Tb/s (terabits per second).

Hitting this 3.2 Tb/s milestone would fundamentally break through the current limitations of copper, giving AI models the rapid-fire data delivery they need to scale up without constantly hitting a wall.

The Bigger Picture

This rare collaboration between fierce rivals like Nvidia and AMD highlights just how severe the AI bandwidth crisis has become. By standardizing optical interconnects, the entire industry avoids a fragmented ecosystem of proprietary cables, ultimately accelerating data center buildouts for cloud giants like Microsoft and Meta. Broadcom stands to win big as a primary supplier of these new standardized optical components, while traditional copper networking vendors will need to pivot their entire business models immediately or face rapid obsolescence.

Related Articles

AI & Machine LearningAdani’s $100B AI Data Center Investment: The Plan [Analysis]

General TechEquinix Potters Bar Data Center: £3.9B Bet [Analysis]

General TechWhy upGrad is Acquiring Unacademy: $3.5B Drop [Analysis]