In a recent report, Bank of America (BofA) analysts estimated that the total addressable market (TAM) for AI networking could grow from $95.6 billion in 2026 to $394 billion by 2030. A significant portion of this growth is expected to come from the optical segment, which is projected to expand from $15 billion to $111 billion over the same period.

• AI Networking: Scaling from $95.6 billion in 2026 to $394 billion by 2030, representing nearly 4x growth.
• Optical Technologies: Increasing from $15 billion in 2026 to $111 billion by 2030, a massive 7.4x expansion.
• CPO & NPO Architectures: Designed to reduce reliance on traditional pluggable optical transceivers, BofA projects the Co-Packaged Optics (CPO) market to reach $14.1 billion, and Near-Packaged Optics (NPO) to touch $10.7 billion by 2030.

Note: The $0 starting values used here for CPO and NPO are illustrative and indicate that these markets are still at an early stage. They do not mean that commercial products do not yet exist.   

Meanwhile, TrendForce forecasts that the combined CPO and NPO market could exceed $39 billion by 2030. This differs significantly from the $24.8 billion combined figure implied by BofA's estimates. However, these figures should not be compared directly without considering potential differences in market definitions and forecasting methodologies. Both remain projections, not guaranteed outcomes. Even with these uncertainties, the forecasts point to potentially significant changes in data center architecture over the coming years.  

Based on these projections, companies such as Broadcom (AVGO), Marvell (MRVL), Coherent (COHR), Lumentum (LITE), Credo Technology Group (CRDO), and Astera Labs (ALAB) could benefit from the continued expansion of AI infrastructure. However, it is worth examining whether all these companies are equally well-positioned to benefit and whether the optimistic forecasts are likely to translate into attractive investment returns.  

One of the main arguments behind rising demand for optical technology is that copper connections are becoming increasingly difficult to scale to meet the requirements of modern AI clusters. As data transmission speeds increase, signal losses and power consumption become more challenging to manage.   However, copper technology should not be dismissed prematurely. Over longer distances and at higher data rates, optical transceivers become increasingly important and, in many applications, effectively indispensable. However, Active Electrical Cables (AECs), which use advanced copper connections, continue to develop and remain competitive within server racks and over relatively short distances.  

According to TrendForce analysts, traditional optical transceivers, CPO, NPO, and advanced copper solutions are likely to coexist for an extended period. The choice of architecture will depend on power consumption, cost, bandwidth requirements, and ease of maintenance. This means that companies such as Credo, which specializes in high-speed connectivity solutions, can continue competing for market share rather than automatically losing out as optical technology advances.

Co-Packaged Optics (CPO) is an architecture that places optical engines within the same package as a main chip or very close to it. This approach can significantly reduce the power required for data transmission, offering an important advantage as AI data centers continue to scale. However, CPO also introduces several engineering challenges, including thermal management, repair complexity, laser-source reliability, and other manufacturing and integration issues.  

Critics of CPO often point out that lasers can degrade or fail more frequently than silicon chips. This can create the impression that a single laser failure would require replacing an entire expensive processor. That conclusion, however, would be too broad.  

The industry is developing architectures that use External Laser Sources (ELS). In these configurations, the laser remains a separate module that can be replaced without removing the entire chip assembly, provided the system is designed to support such servicing. Nevertheless, significant challenges remain. Manufacturing complexity, the risk of low production yields, and the high cost of early CPO implementations could slow widespread adoption.

A projected AI networking TAM of $394 billion does not mean that the market capitalizations of all six companies will grow proportionally. It is essential to distinguish between technological growth potential and the investment attractiveness of each stock at its current price.

1. Broadcom (AVGO) & Marvell (MRVL)

• The Role: Important players in networking ASICs, custom AI chips, and high-performance networking solutions.
• The Catch: Both have strong positions, but neither is an absolute monopolist. Google, Amazon, and Meta are actively developing their own chips to reduce their dependence on external suppliers. This creates opportunities for Broadcom and Marvell, but it also introduces competitive risks as major cloud providers expand their in-house silicon capabilities.

2. Coherent (COHR) & Lumentum (LITE)

• The Role: Suppliers of lasers, optical components, and other technologies used in high-speed data transmission. Both companies could benefit from increasing demand for optical infrastructure.
• The Catch: There is a downside. High development costs and intense pricing competition could put pressure on profitability. Rising revenue does not necessarily translate into proportionally higher net income.

3. Astera Labs (ALAB) & Credo (CRDO)

• The Role: Developers of high-speed connectivity solutions, including retimers, active cables, and other components used to move data throughout AI infrastructure.
• The Catch: The future of these companies will depend substantially on the architectures chosen by major cloud providers. If certain retimer functions become integrated into other chips, or if connectivity architectures change, demand for some standalone products could decline. At the same time, the expansion of AI clusters could create demand for new generations of their solutions. These companies therefore have growth potential, but their stocks may be highly volatile.

To sum up, BofA's $394 billion estimate provides a useful reference point for understanding the potential scale of AI networking infrastructure. However, it remains a forecast, not a guaranteed financial outcome. Its realization will depend on whether major technology companies continue investing heavily in AI systems, networking equipment, and data centers.  

One of the main risks is that investors may have already priced a substantial portion of the expected growth into current stock valuations, particularly for large technology companies such as Broadcom. For this reason, investors should look beyond news headlines and evaluate each company individually:

• Analyze Free Cash Flow (FCF) and profitability rather than focusing solely on top-line revenue growth.
• Monitor quarterly reports from Microsoft, AWS, and Google, paying close attention to capital expenditures and their willingness to continue investing in AI infrastructure.
• Recognize the impact of changing technology standards, which could significantly reshape market shares and alter the competitive positions of individual players.

Thank you for reading, and have a great day!