3.2 Pb/s Fiber Record: 20,000 4K Movies Every Second

What the 3.2 Pb/s fiber record actually means

The number is real, but read the fine print: this is a single link in a controlled test, not a broadband plan.

A 3.2 Pb/s figure is 3,200 terabits every second through one strand of glass. The Hubei provincial government's write-up compared it to 3.2 million households each pulling a separate 1 Gbps line at the same moment — a useful mental picture, not a promise of what lands in your router.

The work came from the National Key Laboratory of Optical Communication Technology, China Information and Communication Technologies Group, Pengcheng Laboratory and Fiberhome Fujikura Fiber Optic Technology. They combined two tricks network engineers have chased for years: spatial division multiplexing, which splits the fiber into 24 separate cores each running its own stream, and S+C+L transmission, which spreads traffic across three wavelength bands instead of the conventional C band alone.

On their own, neither idea is new. Running them together, at speed, without the channels drowning each other out, is the hard part.

Why 'commercially compliant hardware' is the interesting phrase

Plenty of fiber records are set with equipment that will never leave a lab bench. This team says its architecture meets commercial requirements and ran continuously in real time, which separates it from demonstrations that lean on one-off signal processing.

That matters because the bottleneck in real networks is rarely the fiber itself. It is the amplifiers, the transceivers, the cost per bit and the willingness of an operator to rip up a duct. A record that assumes deployable parts is closer to a product than a record that does not.

Closer is not the same as close. There is no deployment schedule, no pricing and no independent assessment of how the link behaves over months of real traffic. Interference between neighbouring cores and mismatched specs across the three bands were problems the team had to manage; managing them in a test is not the same as managing them in a live national backbone.

The 5.29 Pb/s problem

The new record is not the fastest ever reported on 24-core fiber.

Lab experiments on comparable multi-core fiber have reached 5.29 Pb/s — over two petabits more than this result. Those runs tend to rely on specialist processing and equipment that would be impractical to deploy at scale.

So the claim is not 'fastest fiber ever'. It is 'fastest real-time transmission on hardware built to commercial specs'. That is a narrower title, and a more useful one for anyone who cares about networks rather than press releases.

Who actually benefits from petabit fiber

Not home broadband, at least not soon. The near-term customers are data center interconnects, subsea routes and the long-haul links between cities where a single fiber is expensive to lay and capacity is precious.

This is the same pressure driving the rest of the hardware world. AI clusters are starving for memory bandwidth and data center RAM demand is reshaping consumer prices. More compute in fewer places means more traffic on fewer routes.

If multi-core fiber can be spliced, amplified and maintained at sane cost, it buys operators capacity without new trenches. That is the prize. Nobody has yet shown the cost.

What it means for you

Nothing changes for US, UK or EU broadband customers in the near term. No carrier has announced trials of this system, and the 3.2 Pb/s figure describes one link, not an access network.

Realistically, multi-core fiber reaches commercial long-haul routes before it reaches homes, and that timeline depends on standards work, splicing costs and amplifier availability that the researchers have not addressed publicly.

The record is a real step, and the team deserves credit for running it live on parts meant for the field. But fiber records have a habit of looking bigger than the networks they eventually change. The next number that matters will not be petabits — it will be cost per bit.