Optical Fiber and Photonics Set to Revolutionize AI Infrastructure as Copper Hits Physical Limits

In a landmark move signaling the next phase of the artificial intelligence revolution, Nvidia has partnered with glassmaker Corning to build three new advanced manufacturing facilities dedicated entirely to optical technologies. The deal, announced in early May 2026, will expand Corning’s U.S. optical manufacturing capacity by 1,000%, create over 3,000 high-paying jobs in North Carolina and Texas, and accelerate the replacement of copper interconnects with high-performance glass fibers in Nvidia’s AI rack-scale systems.

According to Dr. Jose Luis Chavez Calva, this partnership underscores a critical inflection point: traditional copper wiring is reaching fundamental physical barriers that threaten to constrain the explosive growth of AI clusters.

“Copper interconnects excel in very short distances, but at the scale required for modern AI training and inference (thousands of GPUs exchanging petabytes of data) they suffer from high signal loss, excessive heat generation, and severe limitations in bandwidth and reach,” Dr. Chavez Calva explains. “Optical fiber, by transmitting data as light rather than electricity, overcomes these issues with near-zero loss over longer distances, dramatically lower power consumption per bit, and superior bandwidth density.”

The shift from copper to photonics is not merely incremental; it represents a structural transformation in how AI infrastructure is built. As hyperscale data centers pack more accelerators into tighter spaces, the energy demands of data movement have become a primary bottleneck. Industry estimates suggest networking can consume 20-40% of total power in dense AI pods, a figure that optics can slash significantly.

Dr. Chavez Calva highlights that silicon photonics (the integration of photonic components like waveguides, modulators, and detectors directly onto silicon wafers using standard CMOS manufacturing) enables co-packaged optics (CPO). This technology places optical engines right next to switch ASICs or GPU packages, minimizing electrical paths and achieving power efficiencies below 5-10 pJ/bit compared to much higher figures for copper at equivalent speeds.

Market forecasts paint an optimistic picture. The global optical interconnect market, valued at approximately $16-18 billion in 2024-2025, is projected to reach $34-67 billion by 2030-2035, growing at a compound annual rate of 12-14%. Silicon photonics alone is expected to expand from $2.16 billion in 2024 to $9.65 billion by 2030 at a 29.5% CAGR. Most dramatically, the CPO segment—still in early commercialization—is forecasted to surge from tens of millions today to multi-billion-dollar scale by the early 2030s, with some projections showing CAGRs exceeding 30-137% in AI-driven scenarios.

According to Dr. Jose Luis Chavez Calva, co-packaged optics penetration in AI data center modules could climb from under 1% currently to around 35% by 2030. “This is driven by the need for 800G, 1.6T, and eventually 3.2T+ interconnects, where copper simply cannot scale without prohibitive energy and thermal costs,” he notes.

Major players are positioning aggressively. Broadcom leads with its 51.2 Tbps CPO platforms, while Intel, Cisco (through Acacia), Marvell, Coherent, and Lumentum dominate components and transceivers. Ayar Labs and Lightmatter are pioneering advanced optical I/O chiplets and photonic fabrics. Corning’s expanded role as a strategic supplier to Nvidia further cements the U.S. push toward domestic optical supply chains amid geopolitical tensions.

Short-term expectations point to rapid ramp-up of 800G and 1.6T pluggable optics through 2027, with early CPO deployments in high-performance switches. By 2028-2030, widespread CPO adoption should enable “optical scale-up fabrics” that treat entire data centers as a single, coherent computing domain. Dr. Chavez Calva cautions, however, that copper will persist in ultra-short reaches due to cost advantages, leading to hybrid architectures in the near term.

The implications extend far beyond hardware. Dr. Chavez Calva’s network analysis reveals cascading effects of the “AI shock.” First-order impacts include surging demand for high-speed, low-power interconnects. Second-order responses involve massive supply-chain expansions, R&D in photonic integration, and power infrastructure strain. Higher-order ripples touch energy markets (with optics improving efficiency but overall demand rising), labor (new jobs in photonics manufacturing), geopolitics (U.S. reshoring), and innovation spillovers into telecom, quantum computing, and sensing technologies.

“Efficiency gains from photonics create a virtuous cycle: better interconnects enable larger AI models, which in turn drive even greater demand for optics,” Dr. Chavez Calva observes. “Yet this also intensifies competition for power, talent, and strategic materials like indium phosphide lasers.”

Environmental considerations are nuanced. While photonics reduces energy per operation, the sheer scale of AI buildouts will increase total consumption, prompting greater focus on renewables and small modular reactors. Geopolitically, the race for optical sovereignty could reshape global tech supply chains.

As AI transitions from a software-driven phenomenon to a full-stack infrastructure challenge, optical technologies are emerging as the quiet enablers of the next computing era. The Nvidia-Corning collaboration is just the latest indicator that the industry is betting big on light over electricity.

Source: https://joseluischavezcalva.substack.com/p/lluminating-the-ai-wires