The AI industry has a power problem. As inference computing , the work AI models do every time they answer a question or complete a task , scales into the stratosphere, the copper wiring inside server racks is quietly becoming the bottleneck no one talks about publicly.
Thatâs changing fast.
In early 2025, Ayar Labs raised $500 million at a $3.8 billion valuation to accelerate commercial deployment of its silicon photonics interconnect technology. Backed by Nvidia, AMD, Sequoia Capital, and others, the company has spent 15 years solving one of the most stubborn physics problems in semiconductor design. Around the same time, Nvidia separately committed $2 billion each to Coherent and Lumentum , two other leaders in optical interconnect technology.
This isnât a coincidence. The industry is signaling, loudly, that silicon photonics is moving from research lab to production floor.
What Silicon Photonics Actually Solves
To understand the stakes, it helps to understand the problem.
Inside an AI server rack, chips need to talk to each other , constantly, at enormous scale. Traditionally, that communication happens over copper wires. Copper is cheap, well-understood, and everywhere. But it has physical limits. The more data you push through copper at speed, the more heat it generates, the more signal degrades, and the more power it consumes.
For training large language models, this was already a concern. For inference , where millions of queries are processed simultaneously, in real time , itâs becoming a hard ceiling.
Silicon photonics replaces copper interconnections with fiber-optic ones. Instead of electrons, data travels on photons , particles of light , which move faster, lose less signal over distance, and consume a fraction of the energy. Ayar Labs claims its co-packaged optics technology delivers between 4x and 20x the computing throughput per watt compared to copper-based alternatives.
Thatâs not an incremental improvement. Thatâs a generational one.
Co-Packaged Optics: The Manufacturing Challenge Nobody Mentions
The concept of using light to move data between chips isnât new. Whatâs new is doing it reliably, at scale, integrated directly into the chip packaging.
Co-packaged optics , where the optical engine is packaged alongside the silicon die rather than somewhere further down the signal path , has been theorized for decades. Former Intel CEO Pat Gelsinger, now on Ayar Labsâ board, noted that he launched a silicon photonics research unit inside Intel more than 23 years ago. The technology simply wasnât manufacturable at the yield and cost levels required for volume production.
The barriers were real: thermal management inside tightly packed chip assemblies, the precision alignment required to couple light into microscopic optical waveguides, and the sheer complexity of qualifying a new interconnect approach with hyperscale customers who have zero tolerance for downtime.
Solving those problems requires deep collaboration across the semiconductor supply chain , from advanced packaging foundries and photonic component suppliers to the firms sourcing and qualifying every sub-component in the bill of materials.
What This Means for Semiconductor Sourcing and Supply Chain
For procurement managers and electronics executives, the rise of co-packaged optics and fiber optic semiconductor interconnects introduces a new category of components that many organizations have limited experience sourcing.
A few realities worth understanding now:
The component ecosystem is thin. Silicon photonics devices , laser sources, photodetectors, optical modulators, fiber coupling assemblies , are currently manufactured by a small number of suppliers. Unlike standard copper transceivers, where a broad market of qualified vendors exists, optical components for co-packaged architectures remain specialized. Supply constraints are a near-certainty as demand accelerates.
Qualification cycles are longer. Optical components in high-reliability applications require extensive testing: thermal cycling, fiber pull strength, long-term reliability under operating conditions. Procurement teams accustomed to standard semiconductor lead times should expect extended qualification windows for next-generation interconnect components.
Second-source strategy matters more, not less. With hyperscalers and AI accelerator OEMs placing enormous bets on specific photonic architectures, the risk of single-source dependency is elevated. Identifying and qualifying alternate supply paths early , before a production ramp , is essential risk management.
Inventory strategy may need recalibration. As AI chip architectures evolve to incorporate optical interconnects, the effective lifecycle of some legacy copper-based interconnect components may compress faster than anticipated. Organizations holding significant inventory of older transceiver technology should monitor design win activity in next-generation AI server platforms closely.
The Broader Semiconductor Innovation Cycle
Whatâs unfolding in silicon photonics is part of a larger wave of semiconductor innovation driven directly by AI infrastructure demands. Advanced packaging , chiplets, 2.5D and 3D integration, hybrid bonding , is reshaping how chips are designed and manufactured. Each of these technologies introduces new component categories, new supplier relationships, and new sourcing challenges.
The companies that navigate this transition successfully will be those that treat component sourcing and supply chain strategy as a competitive advantage, not an afterthought. Understanding which suppliers are qualified for production, where capacity constraints are emerging, and how to build resilient multi-source supply networks is increasingly central to product roadmap execution in AI hardware.
Vyrian works with OEMs, EMS companies, and procurement teams across the electronics industry to source hard-to-find, allocated, and specialized semiconductor components. As AI infrastructure demands drive adoption of technologies like silicon photonics, our team is actively tracking supplier qualification status, lead time trends, and inventory availability across the optical and advanced packaging component ecosystem.
The Bottom Line
Silicon photonics is graduating from an interesting research area to a foundational technology in AI infrastructure. The capital flowing into this space , from Nvidia, AMD, and major institutional investors , reflects a growing consensus that copperâs limitations are real and the timeline for optical interconnects moving into production is now, not later.
For electronics companies building or supplying AI infrastructure, the component sourcing implications are significant and the window to build supply chain readiness is shorter than most teams realize.
Working on a design that incorporates advanced optical interconnect or AI accelerator components? Vyrian specializes in sourcing specialized and allocated electronic components for OEMs and EMS providers in complex technology markets. Contact our team to discuss your component sourcing needs, or explore our capabilities to learn how we support procurement teams navigating fast-moving semiconductor technology cycles.