🔔 Subscribe to The Trend Catcher / The Value Anchor — unlock the full historical archive and never miss a weekly pick again.

The AI data center network is approaching a critical inflection point. NVIDIA Corporation(NVDA.US) has officially commenced mass production of its Spectrum-X Ethernet Photonics switches based on Co-Packaged Optics (CPO) technology. This milestone signals that CPO is shifting from a conceptual technology to commercial scale.

Compared to traditional pluggable optical modules, Nvidia’s new architecture delivers a 5x improvement in optical power efficiency, 5x longer AI application runtime, and a 10x increase in reliability.

A formidable supply chain has already formed around this rollout: Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR(TSM.US) is manufacturing the silicon photonics chips, ASE Technology Holding Co., Ltd. Sponsored ADR(ASX.US) (via SPIL) is handling chip-level packaging and testing, Lumentum Holdings, Inc.(LITE.US) and TFC (Tianfu Communications) are supplying laser components, and Foxconn is responsible for system assembly. Nvidia conducts final testing before shipment.

Why CPO is the Future of AI Networks

Traditional data centers rely on independent optical modules plugged into switch panels. While mature, this architecture is hitting physical bottlenecks as AI clusters scale from thousands to hundreds of thousands of GPUs. The long electrical transmission paths lead to high power consumption, heat dissipation issues, and elevated latency, while the sheer number of independent lasers introduces additional failure points.

Nvidia’s Spectrum-X CPO solution integrates the optical engine directly adjacent to the switch ASIC. This drastically shortens the electrical transmission distance, reducing signal degradation and latency. Furthermore, by using External Laser Sources (ELS), the system reduces the total number of lasers required by about 75%, slashing both power consumption and potential failure points.

Nvidia is currently offering two main configurations:

• SN6810: A single-ASIC switch delivering 102.4Tb/s total bandwidth.
• SN6800: A four-ASIC switch delivering 409.6Tb/s with an internal fiber shuffle mechanism, allowing GPU clusters to scale horizontally without adding extra switching layers.

Supply Chain Dynamics: Where is the Value?

According to Bank of America (BofA) supply chain research, while short-term shipments of CPO switches will be relatively low, the true volume inflection point is expected around 2028–2029. BofA estimates Nvidia's CPO switch shipments will reach 14,000 units in 2026, scaling to 190,000 units by 2028.

Breaking down the Bill of Materials (BOM) reveals where the profit pools lie:

1. Optical Engines (OE): The most valuable component, making up 40%–55% of the BOM.
2. External Laser Sources (ELS): Accounting for 10%–15% of the BOM.
3. Fiber Array Units (FAU): Making up 5%–10% of the BOM.

BofA highlights FAUs as the segment with the highest performance elasticity. Connecting the optical engine to the fiber requires extreme micron-level precision and automated testing. Because of these high barriers to entry, BofA projects the FAU market within Nvidia's CPO ecosystem could reach an $8 billion TAM (Total Addressable Market) by 2030, with only a few dominant players surviving.

An Investment Paradigm Shift

The most significant takeaway from the Spectrum-X mass production is the shift in investment logic. The market is no longer trading on the theme of "Can CPO technology work?" but rather focusing on tangible metrics: "Who has entered the mass-production supply chain, who has stable yields, and what is the dollar content per unit?"

While the true inflection point has arrived, the divergence between winners and losers in the CPO supply chain is just beginning.

📈 US Stock Investment Opportunities to Watch

For US investors looking to play the CPO and AI networking revolution, here are the key publicly traded companies positioned across the value chain:

• NVIDIA Corporation(NVDA.US): The primary orchestrator and system designer driving the CPO ecosystem via its Spectrum-X platforms.
• Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR(TSM.US): The indispensable foundry handling the complex silicon photonics chip manufacturing for Nvidia and others.
• Broadcom Limited(AVGO.US) & Marvell Technology(MRVL.US): While Nvidia is pushing its own tech, Broadcom and Marvell are the dominant players in custom networking ASICs and optical engines. Both are heavily investing in proprietary CPO technologies for hyperscalers.
• Lumentum Holdings, Inc.(LITE.US) & Coherent Corp.(COHR.US): Leading suppliers of the External Laser Sources (ELS) and specialized photonic chips required to power CPO systems.
• ASE Technology Holding Co., Ltd. Sponsored ADR(ASX.US): A critical player in the advanced packaging and testing required to integrate optical engines with ASICs (via its subsidiary SPIL, which is part of Nvidia's confirmed chain).
• Fabrinet(FN.US): A highly specialized contract manufacturer dominating the optical communications assembly space. They are uniquely positioned to benefit from the complex automated assembly and testing required for CPO components.
• Corning Inc(GLW.US): A potential beneficiary in the optical fiber and "shuffle box" segment, providing the high-density glass and fiber routing needed inside these next-gen switches.

(Disclaimer: This article is for informational purposes only and does not constitute financial advice. The semiconductor and optical networking sectors can be highly volatile; always conduct your own due diligence before investing.)