Key takeaways
- Faster AI chips are overwhelming traditional copper connections
- Coherent revenue jumps 50% amid optical networking demand
- Light-based connections could reshape AI infrastructure economics
The artificial intelligence industry has spent years making processors faster. Now, some of its biggest infrastructure investments are going toward something considerably less glamorous: replacing electrical connections with light.
The problem is becoming apparent inside enormous AI data centers, where thousands of GPUs must exchange information continuously. Even the most powerful processor can spend time waiting if the network connecting it to other chips cannot move data quickly enough.
The financial consequences are already visible. Optical technology manufacturer Coherent reported $2.37 billion in quarterly revenue on November 4, up 50% year over year, as demand for high-speed networking equipment accelerated.
Why Faster Nvidia Chips Are Creating a New Bottleneck
Traditional data centers were designed around servers performing relatively independent tasks. Copper cables worked well because connections were comparatively short and networking requirements were manageable.
AI training changes that equation.
Modern AI clusters connect thousands of GPUs working simultaneously on enormous models. Those processors must exchange data, synchronize calculations and transfer intermediate results across racks.
As processing speeds increase, communication becomes increasingly important. A delay in transferring information can leave expensive computing hardware underutilized.
Copper remains efficient over short distances, but higher transmission speeds and longer connections increase signal-loss and power-consumption challenges.
Optical technology addresses some of those limitations by converting electrical signals into light and transmitting information through fiber.
Nvidia has explained that larger AI clusters increasingly require optical connections between servers and network switches, especially as network distances expand.
The challenge adds another dimension to the existing AI electricity shortage, because networking equipment also contributes to overall energy consumption.
Coherent's $2.37 Billion Quarter Shows Where the Money Is Going
Coherent's latest results illustrate how the infrastructure transition is creating opportunities beyond Nvidia.
Its Data Center & Communications segment generated approximately $1.95 billion in quarterly revenue, up from $1.09 billion a year earlier.
That represents roughly 79% growth and accounts for approximately 82% of the company's total quarterly sales.
The segment includes products beyond optical connections for AI data centers, so the figure should not be interpreted entirely as AI networking revenue.
Nevertheless, the growth helps explain why semiconductor companies are pursuing optical networking acquisitions to secure technology needed for future computing systems.
Coherent also expects fiscal 2027 pro forma revenue growth to exceed 60%, although this remains management guidance rather than guaranteed performance.
Why Optical Connections Could Reshape AI Economics
The transition creates a new investment consideration: improving GPU performance is not enough if networking cannot keep pace.
Companies building AI infrastructure must balance processor costs against spending on networking, cooling and electricity.
This helps explain why AI capital expenditure increasingly benefits equipment suppliers beyond chip manufacturers.
Optical technology could also improve power efficiency in certain deployments, reducing the electricity required to move information across large computing clusters.
However, optical links are not universally superior. Copper remains attractive for short connections because of its cost, simplicity and established performance.
New optical systems also introduce manufacturing complexity, component costs and reliability considerations.
For companies financing the AI infrastructure boom, those trade-offs matter because every additional networking investment must eventually generate sufficient economic value.
