This article explores how the explosive rise of artificial intelligence is reshaping the optical connectivity landscape. From data center architectures and investment cycles to the photonic technologies that move information, nothing’s untouched.
Drawing on insights from the Silicon Photonics Workshop at ECOC 2025 and industry forecasts, we’ll look at why AI traffic is pushing optics to its limits. Which technologies stand to benefit, and how might supply-chain constraints slow—but not stop—the next wave of optical innovation?
AI’s Acceleration of Optical Connectivity Demand
Over the past decade, AI has gone from a niche workload to a dominant force in data center design. What started as a subtle shift in network usage is now a major reconfiguration of cloud infrastructure, with optical links at the core.
From Early Signals to Market Surge
The first clear sign of this transformation showed up about seven years ago, when Google dramatically increased its use of optics in scale-up/”>AI clusters. Back then, most folks saw it as a bold experiment in high-bandwidth, low-latency interconnects.
Looking back, it was an early warning of the AI supercycle. But the steep expansion in optical demand from 2024 onward has blown past what most people expected.
Cloud and hyperscale companies are now spending heavily on optical interconnects to support larger models, more GPUs per cluster, and tighter coupling between compute nodes.
Investment Patterns Through 2030
Forecasts from the workshop say cloud investment in optics should maintain single-digit annual growth through 2030. That might sound modest after the recent spike, but it’s actually a more stable, structural phase—not just a one-off bubble.
The growth is getting more nuanced, too. The breakneck scaling of 2023–2025, fueled by urgent AI buildouts, is shifting toward a more careful balance between performance, power, and cost in network design.
Scale-Out vs. Scale-Up: The Next Network Evolution
AI networks are evolving in two overlapping phases: scale-out and scale-up. Each phase brings its own demands on optical tech—and different opportunities for vendors and operators.
How Scale-Out Fueled the 2023–2025 Spike
Between 2023 and 2025, scale-out networks drove most of the demand. Companies rapidly added more nodes, racks, and clusters to handle growing AI workloads.
This led to a surge in demand for 400G and 800G optical modules, and a big push toward 1.6T solutions.
The Shift to Scale-Up: LPO and CPO in Focus
The next phase is about scale-up—making each system, rack, and cluster more efficient and tightly integrated. Two technologies stand out here: linear pluggable optics (LPO) and co-packaged optics (CPO).
LPO is already moving into real-world deployment. Companies like Oracle and Google are using LPO to cut power consumption and cost in high-radix switches and AI back-end networks.
By simplifying the analog signal chain, LPO can deliver real efficiency gains at scale. CPO is emerging as the long-term integration path, bringing optics closer to the switch ASIC to minimize electrical reach and power loss.
Early Nvidia and Broadcom CPO products are already in the pipeline. Volume shipments are expected within about two years.
As AI clusters get denser, the benefits of co-packaging—lower power, reduced latency, and higher bandwidth density—start to look pretty irresistible.
Market Outlook for AI Optics
The AI optics segment is growing fast, driven by training clusters, inference fabrics, and storage interconnects. But this growth won’t be a perfectly smooth line.
AI Optics Revenue Projections
LightCounting’s forecast says AI-related optical components and modules should grow from around $5 billion in 2024 to over $10 billion by 2026. That’s a doubling in just two years, showing how intense the AI infrastructure buildout is.
After 2026, growth is supposed to cool off a bit, then return to double-digit rates after 2028 once short-term corrections and capacity issues get sorted. It’s a pretty typical pattern for a maturing, but still fast-moving, tech segment.
Silicon Photonics and the Push to 400G per Lane
Silicon photonics has really gone mainstream, just as AI workloads are demanding more bandwidth and integration in optical systems than ever before.
Silicon Photonics Today and Tomorrow
Silicon photonics tech is behind many of today’s high-speed transceivers, and 1.6T modules are already shipping. Work on 3.2T optics is underway, often paired with new form factors built for AI and high-performance computing.
Sales of silicon photonics chips look set to grow from about $0.8 billion in 2023 to over $3 billion by 2029. That’s not just AI—it’s broader adoption in cloud, telecom, and edge infrastructure, too.
Pushing data rates to 400G per lane is really testing current device physics and modulation schemes. The industry’s exploring new modulator tech to hit these speeds in a way that’s manufacturable and power-efficient.
The Rise of Thin-Film Lithium Niobate (TFLN)
Thin-film lithium niobate (TFLN) photonic integrated circuits (PICs) are showing up as a strong option for next-gen high-speed modulators. These devices offer excellent electro-optic properties and a clear path to integration and volume manufacturing.
By 2029, TFLN PICs could reach about $0.75 billion in sales. They’ll likely replace older modulation technologies that just can’t scale to 400G per lane and beyond.
This transition will be key for keeping up the pace toward even higher per-lane speeds.
Supply Constraints and Near-Term Market Risks
Despite strong demand and a promising technology pipeline, the optics industry faces some tough supply-side challenges right now.
Indium Phosphide Bottlenecks and Market Correction
The biggest constraint is indium phosphide (InP) capacity, which is essential for many high-performance optical components. Manufacturing capacity is tight, and ramping up new lines takes time and effort.
Industry insiders think real capacity relief is about a year away. Until then, supply constraints could mean higher prices, longer lead times, or delayed deployments.
Some projects might get pushed back or redesigned to use more available components. As capacity expands and new materials like TFLN ramp up, growth should pick up again in 2027–2028.
Conclusion: AI as the Catalyst for the Next Optical Era
AI isn’t just another workload anymore. It’s the main force reshaping optical connectivity, from pluggable modules and co-packaged optics to choices about modulator material and even how foundries plan their capacity.
The industry feels like it’s shifting gears. We’re moving from a wild, frantic scale-out phase to a more thoughtful scale-up era.
Technologies like LPO, CPO, silicon photonics, and TFLN—those are the ones that will set companies apart.
Short-term supply constraints might cause some bumps in the road. Still, if you look at the bigger picture, optics is quickly becoming the central nervous system of AI infrastructure.
Innovation in photonics will probably decide how fast, and honestly how far, AI can really go over the next decade.
Here is the source article for this story: AI Demand Reshapes Optical Connectivity and Photonics Roadmaps