The IOWN initiative from NTT (among others) wants to tackle a big challenge in modern data centers: the exponential growth in power consumption and latency caused by converting optical signals to electrical and back again. By keeping data transmission purely photonic from end to end, All-Photonics Networks (APN) can help eliminate bottlenecks in traditional infrastructure.
Of course, optical fiber is nothing new. It reaches most buildings today. However, data converts between light and electricity throughout network infrastructure. In our conversation with him, Marco Provolo, who heads NTT’s IOWN (Innovative Optical and Wireless Network) promotional office, explains that this conversion process creates inefficiency through jitter, which translates to packet loss and retransmission. The cumulative effect of all of this is a significant bottleneck in bandwidth, latency, and power consumption.
An All-Photonics Network (APN) wants to build networks that use only light, with minimal to zero electronic switching along the entire path. In an era where data centers are scaling to 500 kilowatts and even 1 megawatt per rack, the efficiency gains become increasingly valuable. IOWN as a whole, of which APN is part, is something that NTT has been focusing on for quite some time. It is one of the founding members of the IOWN Global Forum. This community has attracted hundreds of members since its foundation in 2020.
Also read a story we published last year, on IOWN in general, the IOWN Global Forum, and about a demonstration of an all-photonics network between Taiwan and Japan.
The four-stage miniaturization roadmap
NTT’s strategy follows a progressive miniaturization approach, moving photonic connections closer to where computation happens. Currently, the technology connects at the switch level between data centers. The PEC-2 (Photonic Electronics Convergence) device represents the next step, connecting optical fiber at the board level and eliminating copper wires on motherboards.
The third PEC stage targets chip-level connections, while the ultimate goal and fourth PEC stage involves photonic connections inside the silicon die itself. Each step down this roadmap increases the difficulty but also multiplies the impact, as the quantity of devices affected grows exponentially when moving from data center-wide to chip-internal connections.
Silicon photonics manufacturing advantages
Unlike some photonic technologies that require entirely new manufacturing processes, the silicon photonics we talk about here leverages existing semiconductor fabrication techniques. The industry has decades of experience miniaturizing electronic components, and many of those processes apply to photonic devices. One of the big challenges is coupling lasers to optical fibers, but once achieved, the connections prove reliable, Provolo says.
Current commercial devices from NTT reach 1.8 terabits per second in a single optical fiber, exceeding the capacity of the fastest individual chips. This bandwidth enables entirely new architectures where multiple data centers pool resources across greater distances.
Data center interconnect transforms infrastructure planning
One of the most interesting applications we heard about involves connecting data centers separated by up to 100 kilometers while maintaining performance as if they were a single facility. The benchmark for this capability is round-trip latency under one millisecond. Traditional IP connectivity over optical fiber cannot achieve this performance level due to the conversion delays.
This capability solves multiple infrastructure challenges simultaneously. Financial services companies struggle to separate data centers by more than 10 kilometers due to bandwidth and latency limitations, Provolo gives as an example. With APN technology, they gain resilience and scaling options previously unavailable. Similarly, older data centers designed for 5-6 kilowatts per rack cannot support modern 500-kilowatt requirements without connecting to new facilities.
Location flexibility and renewable energy
The ability to separate data centers by significant distances while maintaining performance opens new possibilities for renewable energy integration, Provolo argues. Organizations can balance workloads based on energy availability, shifting computation to locations with active wind or solar generation. This vision includes micro data centers in container form factors that leverage local renewable sources while maintaining connectivity to centralized facilities. We heard about that approach to building infrastructure from other players in the market too. Specifically, there’s the MISD (Modular Integrated Sustainable Datacenter) project we covered recently.
Ultra-low latency applications
Some applications require performance levels that simply aren’t possible without photonic networks. NTT is working on remote machinery control for mining operations, where the vision involves operating 100 machines simultaneously from a remote location. This requires receiving multiple camera feeds and providing control feedback in under 70 milliseconds.
The project combines APN with 5G wireless connectivity, though Provolo notes that the majority of latency occurs on the wireless portion. Similar experimentation is underway for construction equipment and remote surgery applications.
PEC-2 timeline
Despite the clear technical advantages, APN remains relatively unknown in the market. Provolo attributes this to the paradigm shift required in network design thinking. Once engineers understand the approach, the value becomes immediately apparent, but explaining the new model takes effort.
The PEC-1 devices available today already enable use cases that were previously impossible or significantly improve existing applications. PEC-2, which operates at the board level, is scheduled for commercialization within one year (so somewhere at the beginning of 2027), Provolo says. This is an important step in making the technology more accessible and demonstrating its value across broader use cases, according to him.
NTT maintains an official roadmap through the IOWN Global Forum, though specific timelines for PEC-3 and PEC-4 devices remain flexible pending market response to PEC-2 deployment. The technology’s success will ultimately depend on demonstrating return on investment across diverse applications, across the entire spectrum of the market.