Prysmian signed a $6.29 billion agreement to supply Koch Industries-owned Molex with optical cable for artificial intelligence data centers, according to The Wall Street Journal's coverage of the deal. The contract runs up to ten years. Molex pays about €550 million, roughly $630 million, upfront to secure the supply. Prysmian says this deal and the broader hyperscaler push behind it could add €10 billion in cumulative revenue through 2035, with as much as €1.1 billion landing every year starting in 2031.
This cabling runs inside the data hall: the dense optical plant linking GPU servers to switching gear a few feet away, separate from the long-haul networks connecting one site to the next. Molex builds connectors and electronic components under Koch's ownership. Prysmian builds the fiber. Prysmian chief executive Massimo Battaini called the deal "a transformative moment for Prysmian's Digital Solutions." Molex chief executive Joe Nelligan said the arrangement lets the two companies "better support our mutual customers." Behind the language: Prysmian is committing €1.25 billion through 2031 to more than double its U.S. optical fiber capacity, adding over 1,000 jobs globally, about 600 of them in the U.S.
A ten-year contract with money paid upfront reads like a capacity story more than a fiber story, and it mirrors the liquid cooling supply chain over the past eighteen months. A handful of vendors now control most of the cooling hardware queue, and hyperscalers have learned to buy years out or go without. Molex just ran the fiber-optic version of that math: pay now, lock a factory's output, skip the line everyone else is standing in.
The reason this cabling fits in the rack at all comes down to heat management. Dense copper wiring used to choke airflow around servers, so every added cable taxed the air cooling budget. Direct-to-chip liquid cooling removed that constraint by moving the heat load off the air path, freeing physical space for exactly the denser optical plant Molex just bought ten years of. Coolant manifolds, quick disconnects, and busbars already compete for that same rack space, and cable trunks are the next line item in the argument over who gets the U.
Fiber stays cool. The transceivers on each end run hot. Networking speeds are climbing from 400G toward 800G and 1.6T to keep pace with GPU-to-GPU traffic, and each module becomes a fixed heat source packed into switch gear that was never built to dissipate it. That heat load scales directly with the fiber count Molex just locked in for the next decade. Vertiv already bet the real bottleneck was never at the chip. This contract is early evidence the network layer is where that bet gets tested next.