Key Points

• Faster AI chips are overwhelming traditional copper cable connections.
• Driven by the demand for optical networks, Coherent's revenue soared by 50%.
• Light-based connectivity may reshape the infrastructure economics of AI.

For many years, the artificial intelligence industry has been working to make processors faster. Today, some of the largest infrastructure investments in this field are being directed towards a relatively less “glamorous” direction: using light to replace electrical connections.

This issue has emerged in the vast AI data center, where thousands of GPU units must continuously exchange information. Even the most powerful processors can be forced to wait if the network connecting them to other chips cannot transfer data fast enough.

Financial impacts have begun to emerge. Optical technology manufacturer Coherent announced quarterly revenue of $2.37 billion on November 4th, a 50% year-on-year increase, due to the accelerated growth in demand for high-speed network equipment.

Why Faster NVIDIA Chips Are Creating New Bottlenecks

Traditional data centers are designed around servers that perform tasks relatively independently. Copper cables perform well because the connection distances are relatively short, and the network requirements are also more controllable.

AI training changed this situation.

Modern AI clusters connect thousands of GPU units together to train massive models. These processors must exchange data, synchronize computations, and transfer intermediate results between racks.

As processing speeds increase, communication becomes increasingly important. Delays in information transmission can prevent expensive computing hardware from being utilized to its full potential.

Copper cables are still efficient over short distances, but higher transmission speeds and longer connection distances present challenges in terms of signal loss and power consumption.

Optical technology solves some of these limitations by converting electrical signals into light and transmitting information through optical fibers.

Nvidia once explained that as the network distance increases, larger AI clusters increasingly require optical connections between servers and network switches.

This challenge also adds a new dimension to the existing AI power shortage, as network devices also contribute to the overall energy consumption.

Coherent $2.37 billion in quarterly earnings reveals where the funds are flowing

The latest performance report of Coherent indicates that this infrastructure transformation is creating opportunities for companies beyond Nvidia.

Its data center and communications business department generated revenue of approximately $1.95 billion in that quarter, a significant increase from $1.09 billion in the same period last year.

This represents a growth of approximately 79%, which also accounts for about 82% of the company's total sales for that quarter.

However, the products of this department include not only optical connections for the AI data center; therefore, this figure should not be interpreted solely as the network revenue of AI.

Nevertheless, this growth still helps to explain why semiconductor companies are pursuing mergers and acquisitions in the optical networking sector in order to acquire the technologies needed for future computing systems.

Coherent also expects that the revenue growth rate for the fiscal year 2027, calculated using the preparation criteria, will exceed 60%, but this remains a management guideline and not a guaranteed performance achievement.

Why a mere connection may reshape the economics of AI

This shift brings new considerations for investment: if the network cannot keep up, simply improving the performance of GPU is not enough.

Companies building the infrastructure for AI must strike a balance between processor costs, network, cooling, and power expenditures.

This also explains why AI's capital expenditures are increasingly benefiting equipment suppliers other than chip manufacturers.

In certain deployment scenarios, optical technology may also improve energy efficiency and reduce the power required for transmitting information within large computing clusters.

However, a connection alone is not necessarily superior in all scenarios. For short-distance connections, copper cables remain attractive due to their lower cost, simpler structure, and proven performance.

The new optical system also brings considerations regarding manufacturing complexity, component costs, and reliability.

For companies that provide funding for the AI infrastructure boom, these trade-offs are very important, as every additional investment in the network ultimately must generate sufficient economic value.