The next AI bottleneck: Networking

• Larger AI systems need faster connections, making physical networking an increasingly important part of infrastructure spending.
• Within these networks, fiber optics and copper wires carry data between racks and across data centers.
• New capacity and new ways to connect AI systems could reshape margins, even as demand continues to grow.

The performance of an AI system depends on how quickly its chips can communicate. Thousands of GPUs often work together on the same task and continuously exchange large amounts of data, so any delay in the network can leave valuable computing capacity idle.

Copper vs fiber? Copper remains an efficient and relatively inexpensive solution to transmit data over short distances, as it carries electrical signals directly. As data speeds increase and distances grow, however, optical fiber becomes more effective than copper because it transmits information as light. Fiber already connects equipment across data centers, and manufacturers are now working to bring optical connections closer to the chips.

Energy consumption is the other part of the equation. Moving data requires electricity, and the amount needed rises as AI systems become larger. More efficient networks reduce the energy spent on communication, leaving a greater share of the data center’s power available for the core task: computing.

Let’s now look at how optical connections are made and how they can be brought closer to the chip.

Optical connections combine several components:

• Lasers, which produce light;
• Optical components, which use that light to send and receive data and
• Specialized chips that help keep the signal clear.

These parts are often combined into modules called transceivers, which connect equipment to optical fiber.

We believe that laser is the part of the value chain where most of the value will be captured. That is because laser manufacturing requires specialist materials, precise production processes and yields high enough to make each batch profitable. Customers also need to test and approve the products before using them. Adding capacity therefore takes time, which can help manufacturers defend prices and margins when supply falls short of demand.

In contrast, competition is tougher further downstream. Transceiver suppliers face pricing pressure because customers can choose from several manufacturers globally. Likewise, equipment makers depend on investment in factories, leaving revenues exposed to the timing of expansion projects.

AI requires faster connections capable of moving more data while using less energy. New optical architectures are changing the mix of components used in AI infrastructure, creating opportunities for some suppliers while gradually eroding demand for older products.

The scale of this shift will depend on the pace of customer adoption and suppliers’ ability to adjust their products and production. Expanding capacity requires time and investment, while adding too much too quickly can intensify competition and put pressure on prices.

Growth across optical networking could therefore be significant, but lasting profits will depend on keeping pace with technological change, managing capacity and controlling costs. Share prices may already reflect part of this potential. Therefore, we focus on which suppliers can defend pricing during a capacity buildout phase, favoring companies that sit in the part of the value chain that captures most value, like lasers and optical chips. 

El Niño’s impacts extend beyond soft commodities, with notable examples being electricity generation, mining, and transport.

Hydropower-dependent regions are particularly sensitive because lower reservoir levels reduce cheap generation and may force corporates to purchase more expensive replacement power. This matters especially for smelters, which need a continuous, large-scale power supply. As some of China’s smelting capacity has moved to hydro-rich regions, its exposure to drought has increased.

Copper presents a more mixed picture. Heavy rainfall and flooding can disrupt mining operations in parts of Peru and northern Chile, reducing output and supporting prices. The impact on earnings, however, depends on each miner’s geographic exposure and the severity of the disruption.

Finally, transport links form another potential transmission channel, as low water levels can restrict vessel capacity through for instance the Panama Canal, similarly to what happened during summer 2026 to the Rhine river in Germany. Such constraints can raise freight costs and lengthen delivery times, although the financial impact depends on the duration of the disruption and the availability of alternative routes.