Since the start of 2026, the global computing power supply chain has been undergoing a fundamental transformation. As AI training clusters scale from tens of thousands to hundreds of thousands of chips, traditional electrical interconnect technologies—copper cabling and pluggable optical modules—are nearing their physical limits. Soaring power consumption, signal degradation, and latency accumulation are becoming critical bottlenecks that constrain the full release of computing power. Facing this challenge, chip giant NVIDIA (NVDA.US) is placing a substantial financial wager on an emerging solution: silicon photonics and co-packaged optics (CPO) technology.
**A $6.5 Billion Sprint: NVIDIA's "Speed of Light" Strategy**
NVIDIA's investment pace in the optical interconnect sector during the spring of 2026 has been dizzying. On March 2nd, NVIDIA announced investments of $2 billion each in U.S. optical technology firms Lumentum (LITE.US) and Coherent (COHR.US), totaling $4 billion. The non-exclusive agreement with Coherent includes multi-billion dollar procurement commitments and future access to advanced laser component manufacturing capacity. Concurrently, NVIDIA is supporting Lumentum's R&D and capacity build-out, aiding its establishment of a new wafer fabrication plant in the United States.
Three weeks later, NVIDIA invested another $2 billion in Marvell Technology (MRVL.US) for a joint development effort on silicon photonics technology and custom XPU solutions. In early May, NVIDIA announced a $500 million investment in fiber optics leader Corning (GLW.US), securing priority access to its advanced optical connectivity production capacity. Corning plans to increase its U.S. production capacity for optical connectivity products to ten times current levels and expand fiber capacity by over 50%. Furthermore, NVIDIA participated in a $500 million Series E funding round for Ayar Labs, a photonic chip startup whose valuation has reached unicorn status.
This series of investments is interpreted by the market as a highly systematic, large-scale strategic move. Through strategic investments rather than acquisitions, NVIDIA is simultaneously securing capabilities across three layers: optical components (Coherent and Lumentum), optical interconnect architecture (Marvell), and optical I/O chips (Ayar Labs), covering the complete chain from photonic chips to system-level interconnects. Analysts note that NVIDIA is leveraging its massive profits to build a robust ecosystem supporting advanced AI technology, boosting demand for its computing infrastructure by directly investing in key optical component suppliers.
**Hitting the Power Wall: The Imperative Shift from "Electrical" to "Optical"**
These substantial investments are driven by the imminent power consumption and bandwidth crisis in AI data centers. The physical limits of copper are becoming apparent. Industry analysis indicates that as AI computing scale continues to expand, data centers face severe power and transmission bandwidth bottlenecks. The physical properties of copper dictate that signal loss increases exponentially with frequency. As cluster sizes grow from thousands to tens of thousands or more GPUs, the interconnects between chips consume an increasingly large share of the system's total power budget, becoming a bottleneck for overall performance.
In contrast, optical interconnects offer significant advantages. CPO integrates high-speed optical engines with switch or AI compute chips on the same package substrate using advanced packaging, limiting high-speed electrical signal transmission to millimeter-scale distances, while longer distances are handled by fiber optics. Compared to traditional pluggable optical module solutions, CPO can reduce power consumption by over 40%, increase bandwidth by 3x, and cut latency by 50%. For hyperscale data centers, this translates to potential annual electricity cost savings in the hundreds of millions of dollars alongside higher training throughput.
Brian Colello, Senior Equity Analyst at Morningstar, points out that while copper remains the primary connection method due to its low cost and high reliability, photonics will become increasingly vital in AI infrastructure. "The roadmap for NVIDIA's next-generation AI rack-scale solutions requires more and more optical interconnect to handle the exponential bandwidth growth from new models and higher usage." At the recent ISSCC 2026 conference, NVIDIA again highlighted the fundamental difference in interconnect needs between AI data centers and traditional cloud data centers, noting that the inherent limitations of electrical interconnects are one of three core bottlenecks for computing power release.
"Photonics enables NVIDIA to scale its AI infrastructure while avoiding the high energy costs associated with continued reliance on electricity and copper," said Alvin Nguyen, Senior Analyst at Forrester, in an interview. "By investing in photonics companies, NVIDIA ensures continued progress in the field, avoiding scalability and performance bottlenecks that would come from sticking with electricity and copper technology."
**2026: The "Industrialization Year" for CPO's Move from Concept to Orders**
Silicon photonics and CPO are not entirely new concepts. Intel (INTC.US) began its efforts in this area over two decades ago and has shipped more than 8 million silicon photonic optical transceivers. Broadcom (AVGO.US) has also been an early proponent of CPO standards. However, what is truly pushing this technology to an inflection point is the quantum leap in computing power demand driven by AI.
The market widely anticipates 2026 will be the critical starting point for CPO's move toward commercial deployment. Guojin Securities stated in a recent report that "2026 is the industrialization year for CPO." According to TrendForce research, the penetration rate of CPO in optical communication modules for AI data centers is expected to be only about 0.5% in 2026, but is projected to rise to 35% by 2030. From an industry timeline perspective, scale-out scenarios are expected to materialize first in 2026, with potential extension to scale-up scenarios from 2027–2028.
The industrialization expectations for CPO have triggered chain reactions in both primary and secondary markets. Supply chain tightness is intensifying. Lumentum disclosed that its indium phosphide laser capacity is fully booked with lead times extending to 32 months, and the supply-demand gap continues to widen. Israeli silicon photonics foundry Tower Semiconductor (TSEM.US) reported a 300% year-over-year increase in its silicon photonics business revenue for Q1 2026 and signed long-term supply contracts worth $1.3 billion. Over 70% of its total silicon photonics capacity through 2028 is already booked or in the booking process. Indium phosphide substrates are monopolized by Japan's Sumitomo, China's Beijing Tongmei, and Japan's JX, which control over 90% of the market share, while delivery lead times for high-end EBL equipment commonly exceed one year—these are practical bottlenecks constraining downstream capacity expansion.
Capital markets have followed with fervor. Lumentum's stock surged 339% throughout 2025 and has risen approximately 140% year-to-date in 2026, with its market capitalization approaching $70 billion, leading to its inclusion in both the S&P 500 and Nasdaq 100 indices. Coherent's stock is up about 110% year-to-date, hitting all-time highs. Marvell has gained 140% since the start of the year, and Corning is up 110%. In the Taiwan and A-share markets, silicon photonics concept stocks are also among the most sought-after sectors in the current AI infrastructure rally.
However, the "industrialization year" does not imply that mass deployment is imminent. "The technology itself is sound; the challenge is production scale. The manufacturing yield for complex co-packaged optical components remains a significant hurdle, as the precision alignment tolerance between optical and silicon-based components is extremely low. If issues arise during the packaging process, rework is often not possible," said Nick Patience, Head of AI Business at Futurum Group.
"The industry transformation has begun, but it is still in the early stages. I expect mass adoption to begin from 2028 onwards," Patience added.