The conversation around AI data center constraints is shifting away from raw computing power and electricity supply toward a less obvious but equally critical problem: how data moves between thousands of chips. As artificial intelligence systems grow larger and more complex, traditional copper-based connections are hitting hard limits in speed, distance, and energy efficiency, forcing the industry to embrace photonics, a technology that uses light instead of electrical signals to transmit data.

Why Is Data Movement Becoming the New Bottleneck?

When GPU clusters scale to tens of thousands of chips for training massive foundation models, the sheer volume of data flowing between processors becomes a limiting factor. Copper connections, which have powered data centers for decades, cannot keep pace with the bandwidth demands of modern AI systems. Optical connectivity solves this by enabling multi-terabit bandwidth with ultra-low latency and significantly greater energy efficiency than traditional wiring.

Goldman Sachs projects that the total addressable market opportunity for optical networking will grow more than nine times over through 2028, expanding from an estimated $15 billion to an estimated $154 billion. This explosive growth reflects how central photonics has become to the future of AI infrastructure.

What Technologies Are Driving the Photonics Shift?

One of the most important emerging technologies in optical connectivity is Co-Packaged Optics, or CPO, which integrates optical engines directly with switching or computing chips. This represents a fundamental redesign of how data centers are built, moving away from separate copper cables toward integrated optical systems. However, industry experts expect traditional pluggable optical modules to remain dominant through 2026 due to their lower costs, flexibility, and easier maintenance.

CPO adoption is likely to accelerate from 2027 onward, with experts expecting penetration of roughly 20 to 30 percent by 2028. Near-Packaged Optics could serve as an intermediate solution as the industry addresses CPO's reliability, manufacturing yield, and packaging challenges.

How Are Tech Giants Positioning Themselves in the Photonics Race?

NVIDIA has committed at least $6.5 billion since March to companies developing photonics technology, including investments in Coherent and Lumentum. The company's Feynman architecture, expected in 2028, represents a major shift toward photonics and away from traditional copper-based interconnects. NVIDIA has already moved CPO technology into production with its Spectrum-X networking platform, while Broadcom remains another major player in the space.

Industry experts expect NVIDIA and Broadcom to collectively control about 70 percent of the global CPO market by 2028, with NVIDIA potentially accounting for 50 percent and Broadcom 20 percent. This concentration reflects how critical these two companies have become to the future of AI infrastructure.

What Investment Opportunities Are Emerging?

The photonics boom has attracted significant investor attention, with new exchange-traded funds (ETFs) launching to capture exposure to the trend. Roundhill Investments launched the Roundhill Photonics and Optics ETF, ticker LYTE, on August 6, 2026, as a pure-play actively managed fund targeting global photonics and optics leaders. The fund holds 19 stocks and charges 35 basis points in fees, with top holdings including Coherent at 15.99 percent weight, Lumentum at 14.82 percent, and Eoptolink Technology at 13.73 percent.

Another actively managed fund in the space holds 28 stocks total, with Lumentum at 15.22 percent weight, AXT at 11.4 percent, and Aixtron SE at 9.09 percent. The fund charges 75 basis points in fees and maintains geographic diversification, with the United States representing about 50 percent of holdings, followed by China at 10.51 percent and Germany at 9.09 percent.

Steps to Understanding the Photonics Investment Landscape

• Market Size Growth: The optical networking market is projected to expand from $15 billion to $154 billion by 2028, representing a nine-fold increase driven by AI data center demand.
• Technology Timeline: Traditional pluggable optical modules will remain dominant through 2026, while Co-Packaged Optics adoption accelerates from 2027 with 20 to 30 percent penetration expected by 2028.
• Competitive Landscape: NVIDIA and Broadcom are expected to control 70 percent of the global CPO market by 2028, with NVIDIA holding 50 percent and Broadcom 20 percent of the market share.
• Investment Vehicles: New photonics-focused ETFs offer exposure to companies like Coherent, Lumentum, and Eoptolink Technology, with fees ranging from 35 to 75 basis points.

Why Does This Matter Beyond Wall Street?

The shift toward photonics has profound implications for how AI systems will be built and deployed in the coming years. While much of the public debate around AI data centers has focused on electricity consumption and nuclear power, the real constraint may be how efficiently data moves through these facilities. By solving the data transmission problem, photonics could unlock the next generation of AI capabilities without requiring proportional increases in power consumption.

This technological transition also highlights how AI infrastructure is becoming increasingly specialized and complex. The companies that master photonics integration will likely dominate the next phase of AI development, making this not just a technical challenge but a strategic competitive advantage for the world's largest technology companies.