How NTT Is Building a Low-Power Internet with Silicon Photonics

NTT and Toshiba are pushing the frontiers of silicon photonics to tackle a looming crisis in data-center energy use and network performance.

They want to keep data in the optical domain for as long as possible. Their vision—anchored in NTT’s Innovative Optical and Wireless Network (IOWN) initiative—aims to radically cut power use, boost bandwidth, and slash latency across global networks.

Table of Contents

Why Data-Center Networks Need a Photonic Revolution

Modern data centers rely heavily on fiber optics to move data between facilities. Inside the data center, though, much of that optical information constantly gets converted back into electrical signals for routing and processing.

This process creates a hidden but enormous energy cost. It’s also a real bottleneck for performance.

Each optical–electrical–optical conversion step burns power, generates heat, and adds latency. As AI workloads surge, this inefficiency starts to look unsustainable.

Projections suggest that by 2030, data centers—driven mostly by AI—could use as much electricity as Japan does today. That’s not just a hypothetical; it’s a grid-level challenge.

The Limits of Conventional Architectures

Traditional network architectures weren’t built for the data volumes, model sizes, or real-time responsiveness that today’s AI systems demand. Copper interconnects, even when paired with conventional optics, can’t scale forever without hitting power and thermal walls.

This is where silicon photonics starts to feel indispensable.

Inside NTT’s IOWN Vision: Encoding More in Light

IOWN is NTT’s ambitious blueprint to keep data in the optical domain for as much of its journey as possible. They want to push optics from long-haul transport all the way down to the chip level.

Instead of treating light as just an on/off carrier of bits, IOWN tries to use multiple degrees of freedom in a single optical signal. It’s not just about presence or absence of light anymore.

In this approach, information gets encoded in phase, amplitude, polarization, and photon number as well. This multidimensional encoding can really boost the information density per photonic channel, letting vastly more data flow without a big jump in power.

Projected Performance and Efficiency Gains

NTT claims that a mature IOWN network could:

Cut telecom network power consumption to one-hundredth of current levels

Increase data capacity by a factor of 125

Reduce latency to a small fraction of today’s figures

Photonic-Electronic Convergence: The Role of PEC Hardware

In 2023, NTT moved from vision to deployment by offering early IOWN components. They also created NTT Innovative Devices to focus on photonic-electronic convergence (PEC) hardware.

The goal is to tightly integrate optics and electronics within a single package, not treat them as separate, loosely connected subsystems. This integration is crucial.

Purely all-optical computing isn’t realistic for most applications. Electronics are still essential for logic, memory, and control.

But embedding photonics next to—or even inside—electronic chips can cut out many of the most wasteful conversion steps.

Second-Generation PEC Switches and Real-World Demos

NTT says its second-generation PEC switches can handle 51.2 terabits per second while cutting power use compared with conventional optical computing systems. This isn’t just a lab experiment; the technology has already supported some notable demos:

Factory equipment controlled from 300 kilometers away, showing ultra-low-latency industrial applications

A “Cho‑Kabuki” performance linking Osaka and Taipei with just 17 milliseconds of latency, hinting at the potential for real-time, high-quality remote collaboration

Roadmap: From Boards to Chips to the Future Data Center

NTT’s roadmap for PEC and silicon photonics is all about shrinking the distance over which optical interconnects operate inside computing systems. Each step brings optics closer to the heart of computation.

The planned progression includes:

Board-to-board optical links as the current focus, improving communication between separate modules in servers and switches

Interchip connections by around 2028, letting multiple chips on the same board talk optically with minimal conversion overhead

Intrachip optical links in the early 2030s, bringing photonics right into the chip package and maybe even onto the silicon die itself

Commercialization of new PEC hardware is targeted for 2026, in collaboration with partners like Broadcom. That timing lines up with the next wave of AI and HPC demand, where traditional electronics alone probably won’t cut it.

Balancing Realism and Ambition

Experts mostly agree on one thing: fully all-optical systems aren’t coming anytime soon. Electronics will still play a big role for quite a while.

Still, a lot of folks see IOWN as a real and meaningful step forward for silicon photonics. If AI keeps driving up demand for data and computation, we might see architectures that really blend photonics—like what NTT and Toshiba are working on—start to change how we build data centers, telecom networks, and even supercomputers.

IOWN isn’t just a minor upgrade. It’s more like a foundational shift, nudging digital infrastructure toward a future that’s centered on photonics and a bit more mindful about energy. Hard to say exactly when, but the momentum’s there.

Here is the source article for this story: NTT’s Vision for a Low-Power Internet Future

Additional Reading: