Michael Murray Looks to AI’s Next Infrastructure Challenge

As artificial intelligence (AI) computing demands increase, discussions are often centered around what AI software can do, not the infrastructure that supports it. Electricity and cooling are often at the forefront of such concerns, but moving massive amounts of data is becoming an engineering issue of its own. 

The underlying infrastructure of AI is what interests Michael Murray, the chairman, CEO, and president of Kopin Corporation. He is focused on bandwidth, power consumption, and the possibility of using optical technology to move large amounts of data using light. 

“AI is just starting to become part of our way of life,” Murray says. 

A career built around solving technical problems

Murray grew up in Northern Ontario, Canada, the child of working-class parents. His interest in electronics was born out of necessity. If something broke, there wasn’t always the option to buy a replacement. Murray would figure out how something worked and then attempt to fix what was broken. He learned how to repair old radios, record players, and other electric equipment. Later, this knowledge helped him branch out to working on cars and doing larger household repairs.

When Murray entered the technology industry during the late-1990s telecom boom, he experienced the industry’s collapse. His experience with the rapid expansion and decline of the telecommunications industry helped shape his views on resilience and disciplined leadership. He understood that instead of focusing only on current conditions, he had to consider how the decisions he made would affect a longer-term strategy. 

In his current leadership role, he continues to apply his ability to turn around situations through strategic direction, leading organizations through cultural change and operational improvements. 

Why AI infrastructure matters

Graphics processing units (GPUs) don’t handle tasks only in sequence; instead, they can perform many calculations at the same time. This ability makes them useful for the intensive computational demands of AI systems. Larger AI systems require many processors working together, which requires them to exchange enormous amounts of data.

That coordination becomes more difficult as systems expand. Each processor needs access to data quickly enough to avoid leaving computing capacity idle.

Data movement between processors can become a bottleneck when there isn’t enough bandwidth to handle the amount of information being exchanged, Murray says. As AI systems grow in scale, this challenge can become more significant. And since data movement consumes energy, the bandwidth isn’t the only cause for concern. Power consumption is also part of the engineering problem Murray is interested in addressing. 

Using light to move AI data

Traditionally, computers use copper connections to move electrical signals between components. Since AI systems require moving more data, engineers are exploring optical interconnects that transmit information via light rather than electrical signals. The appeal is not simply speed. The amount of energy required to move each bit of data also matters as systems scale.

Kopin and Fabric.AI are developing a technology called Neural I/o that uses MicroLEDs (microscopic light-emitting diodes) as part of an optical interconnect system. The programmable MicroLED pixels are designed to function as optical transceivers, with the goal being to send and receive signals carried by photons (light).

The system is being developed for GPUs and other high-performance processors. According to the developers, the goal is to increase bandwidth and reduce the amount of energy needed to move data.  

Leadership through three lenses

Organizations, much like electronics, are complicated systems, and Murray says his leadership approach is based on the Three Lenses framework taught at MIT Sloan. The framework includes strategic design, political, and cultural lenses, with the political lens referring to power and competing interests within an organization rather than partisan politics. 

The three lenses have helped him look at how an organization is structured, recognize the individuals and groups with varying interests, and look for the shared beliefs that influence how people operate.

Murray believes that by looking at an organization from more than one perspective, he can identify whether a problem comes from its formal structure, competing priorities among the people inside it, or the culture that shapes how everyone works together. 

“Filter the noise from the signal and stay focused on driving your plan for the long run,” he says. 

Looking beyond software

Murray’s view of AI is shaped by his perception that it is in the early stages of its development and that the physical infrastructure supporting AI systems will need to develop alongside software advancements. Bandwidth and energy use are challenges that need to be addressed as AI becomes more computationally demanding. 

Optical interconnects are just one of the many potential approaches currently being explored. Neural I/o is under development in hopes of making improvements to data transfer. That work remains at an early stage. Though it has yet to be demonstrated as a commercially viable product, it is a part of Murray’s approach to understanding a system, identifying a problem, and finding solutions. 

Entrepreneur Media was not involved in the creation of this content.