Artificial intelligence is often described as a software revolution. Behind every new model, however, sits a rapidly expanding physical infrastructure of processors, data centers, networks, power systems and specialized electronic components.
As AI adoption accelerates, that physical infrastructure is becoming increasingly difficult to expand efficiently.
The challenge is no longer simply building more computing capacity. Data centers must find ways to move enormous quantities of information, manage rising electricity consumption and deploy increasingly sophisticated hardware without allowing costs and complexity to grow at the same pace.
That is creating opportunities for technologies that might once have appeared peripheral to the AI story.
Advanced materials, photonics and new approaches to electronics manufacturing could become increasingly important as the industry searches for ways to make AI infrastructure faster, more energy efficient and easier to scale.
Bruce Berkoff, Chief Marketing Officer of iBeam Materials, Inc., believes this shift could create an opportunity for a different kind of technology company: one focused not on building the AI systems themselves, but on improving some of the underlying materials and manufacturing processes those systems may eventually depend on.
AI is becoming a physical infrastructure challenge
The extraordinary growth of AI has placed enormous demands on data centers.
Training and running increasingly capable models requires large clusters of advanced processors operating continuously. Those processors need power, cooling and high-speed connections capable of moving vast amounts of data between computing and memory systems.
The International Energy Agency expects electricity consumption from data centers worldwide to roughly double, rising from approximately 485 terawatt-hours in 2025 to around 950 TWh by 2030.
That increase highlights an important point about the AI economy. More efficient software and more powerful processors alone may not be enough to accommodate the next phase of growth.
The infrastructure connecting those processors matters too.
As computing systems become increasingly interconnected, moving data efficiently can become a significant source of energy consumption and latency. This is one reason optical technologies are attracting greater attention.
The case for moving more data with light
Traditional electrical connections have powered computing infrastructure for decades. But as the volume of data being transferred increases, the energy and performance characteristics of those connections become increasingly important.
Optical communication offers another approach.
Instead of relying entirely on electrical signals moving through conductive connections, optical systems transmit information using light. In the right applications, that can provide advantages in bandwidth and energy efficiency.
The technology is not new. Fiber-optic networks already carry enormous quantities of information around the world. The emerging question is how optical communication can move closer to the computing hardware itself.
That creates a need for compact, high-performance optical components that can operate alongside increasingly sophisticated electronic systems.
MicroLED technology is one area that could play a role.
MicroLEDs are generally associated with next-generation displays, but their ability to generate light at extremely small scales also makes them interesting for certain photonic applications. Berkoff believes they could eventually serve as components in optical data communication and switching technologies.
The opportunity, therefore, extends beyond the display market.
Why materials could become the next bottleneck
Developing an optical system is one challenge. Producing the underlying components in sufficient quantities and at an acceptable cost is another.
This is where advanced materials enter the equation.
iBeam Materials is developing a platform based on Ion Beam Assisted Deposition, or IBAD. The process is designed to create semiconductor-quality crystalline templates on flexible metal substrates.
That foundation is important because advanced electronic materials often depend on highly controlled crystalline structures. Producing those structures on flexible material opens the possibility of using continuous manufacturing techniques rather than relying exclusively on conventional rigid-wafer processes.
Berkoff believes that distinction could become increasingly valuable as demand for advanced electronic components grows.
The company’s approach combines IBAD with roll-to-roll manufacturing, in which flexible material continuously moves through a production system.
Rather than treating every component as an individual item that must pass through a series of discrete manufacturing stages, the objective is to create a more continuous production model.
That does not automatically make a technology cheaper. The economics still need to be demonstrated at commercial scale. But it introduces another way of thinking about how advanced electronic materials could be produced.
The manufacturing economics behind AI hardware
The semiconductor industry is already preparing for substantial expansion.
PwC forecasts semiconductor fabrication capacity to grow at approximately 7% annually through 2030, alongside continued growth in semiconductor equipment investment.
The challenge is that conventional semiconductor manufacturing can require enormous amounts of capital.
Large fabrication facilities can cost billions of dollars to construct and require sophisticated equipment, highly controlled environments and extensive supporting infrastructure.
That model has been extraordinarily successful. It has also contributed to the geographic concentration of semiconductor manufacturing.
PwC expects more than 70% of semiconductor equipment investment through 2030 to remain concentrated in Asia.
For the AI industry, geographic concentration introduces another consideration. As governments and companies seek more resilient technology supply chains, there is increasing interest in manufacturing approaches that could potentially operate with different capital requirements and production footprints.
This is where continuous manufacturing could become strategically interesting.
If advanced electronic materials can eventually be manufactured using wider and faster flexible substrates, production capacity could potentially be expanded without simply replicating the same traditional factory model.
From displays to data centers
iBeam’s technology was not developed exclusively around AI.
Its potential applications span areas including advanced displays, photonics, power electronics and other electronic materials. But the rapid growth of AI provides an example of why those markets may increasingly intersect.
Consider the evolution of a data center.
Processors become more powerful. More processors are installed. More data moves between them. More electricity is consumed. More cooling capacity is required. And eventually, the physical connections between different components become an increasingly important part of the system’s overall performance.
Improving one component can therefore have implications throughout the infrastructure.
For example, if optical technologies become more widely integrated into computing systems, manufacturers will need reliable ways to produce the materials and light-generating components required by those systems.
That is the kind of market Berkoff believes iBeam could eventually serve.
The company does not need to build the complete data center or develop the AI software running inside it. Its potential role is further upstream: supplying materials and components that other companies could incorporate into larger technologies.
A manufacturing platform rather than a single product
This distinction is central to iBeam’s strategy.
Rather than focusing on a single finished consumer product, the company is developing a manufacturing platform that it believes could support several different categories of electronics.
The same underlying capabilities could potentially be adapted to different materials and applications.
MicroLEDs are one example. Photonic components are another. GaN-based power electronics represent another potential area.
That approach carries both opportunity and risk.
A platform that can address multiple markets has a broader potential addressable market than a single-purpose technology. At the same time, each application has its own technical requirements, qualification processes and commercial hurdles.
The technology must ultimately prove itself not only in a laboratory environment but within the demanding economics of real manufacturing.
The question investors and manufacturers may increasingly ask
The AI hardware race has largely focused public attention on processors.
Companies developing increasingly powerful chips have become some of the most valuable businesses in the world. But the broader AI infrastructure ecosystem is much larger than the processors themselves.
Power equipment matters. Cooling matters. Networking matters. Optical communication matters. Materials matter. And the factories capable of producing all those components matter.
As AI infrastructure continues expanding, the limiting factor may increasingly shift from what engineers can design to what manufacturers can produce economically and at sufficient volume.
That could change the way the industry evaluates emerging technologies.
A breakthrough material may be impressive. A faster optical component may be valuable. A new display architecture may offer better performance.
But if producing any of them requires prohibitively expensive manufacturing infrastructure, their commercial potential can remain limited.
Conversely, a technology that can be produced continuously, efficiently and at scale may unlock applications that previously appeared economically unrealistic.
The next AI breakthrough may happen outside the chip
For Berkoff, this is ultimately a question of industrial economics rather than technological hype.
His experience spans periods of major change in display and photovoltaic manufacturing, giving him a perspective shaped by the practical challenges involved in taking technologies from development into large-scale production.
His view is that manufacturing deserves to be considered part of the innovation itself.
That perspective may become increasingly relevant as AI pushes the world’s digital infrastructure toward unprecedented scale.
The next generation of AI systems will require more than better algorithms and faster processors. They will require an enormous supporting ecosystem capable of moving, powering and connecting those systems efficiently.
That ecosystem will depend on materials and manufacturing methods that can keep pace.
If technologies such as iBeam’s platform can demonstrate that advanced electronic materials can be produced more continuously and economically, their significance could extend well beyond the markets for which they were originally developed.
The AI revolution may have begun with software and semiconductors. Its next chapter could increasingly be shaped by the technologies that make the physical infrastructure behind them possible.
