By Nikolajus Gavrilinas, Co-founder and CEO of LITILIT

Artificial intelligence has a brain. Giving it a body is proving considerably harder.

Companies including OpenAI, Meta and Anthropic are pouring billions into AI models that can write code, reason through problems and hold increasingly sophisticated conversations. Every month brings new advances in what these systems can accomplish.

But behind every AI breakthrough lies an enormous physical infrastructure: data centres, processors, memory chips, cooling systems and the optical connections that allow these components to communicate.

As we’ve explored in our recent coverage of physical AI and autonomous systems, bringing artificial intelligence into the real world requires far more than advances in software. It also depends on the hardware, manufacturing processes and industrial systems that make those capabilities possible.

And building that infrastructure is becoming one of the industry’s most pressing challenges.

The problem isn’t simply the scale of investment required. It’s the extraordinary precision needed to manufacture the hardware on which modern AI depends.

AI’s trillion-dollar infrastructure challenge

According to Gartner, AI providers are expected to spend nearly $1 trillion on infrastructure this year, with a further $6 trillion projected between 2027 and 2030.

Much of that investment will go towards the hardware inside data centres.

AI chips and the high-bandwidth memory stacked alongside them are only part of the equation. Glass substrates, advanced packaging technologies, optical and photonic components, and sophisticated cooling systems are increasingly essential to keeping thousands of interconnected processors operating efficiently.

Yet manufacturing these components presents challenges that conventional tools are struggling to overcome.

As AI systems become more powerful, the components supporting them must become smaller, more complex and more precisely engineered.

In many cases, we’re approaching the physical limitations of established manufacturing techniques.

When conventional manufacturing reaches its limits

Consider through-glass vias, or TGVs.

These microscopic holes are drilled through glass and filled with metal, creating electrical connections that allow signals and power to pass vertically through advanced semiconductor packaging.

Chipmakers are increasingly interested in using these connections, but producing them reliably at commercial scale remains difficult.

Conventional drilling methods can leave microscopic cracks, uneven holes and other defects that compromise the integrity of the glass.

According to TrendForce, challenges associated with these manufacturing processes could delay commercial-scale production of certain glass-core substrates until after 2030.

And glass is only one example.

High-bandwidth memory is constructed by stacking multiple chips, often 12 or 16 layers high. Each layer must be manufactured and assembled with extraordinary precision.

Meanwhile, advanced circuit boards increasingly rely on hard, low-loss materials that can quickly wear down conventional drilling tools.

As components become thinner and more densely packed, the margin for error continues to shrink.

These are not problems that can be solved simply by increasing computing power or developing better software.

They require new approaches to manufacturing.

Why femtosecond lasers matter

One promising technology is the femtosecond laser.

Unlike conventional laser systems, femtosecond lasers emit pulses lasting just a quadrillionth of a second.

These extraordinarily short pulses can remove material before significant heat spreads into the surrounding area.

The result is an exceptionally precise manufacturing process that can create microscopic features while minimising cracking, melting and other forms of thermal damage.

For advanced glass substrates, this means drilling extremely small holes without compromising the surrounding material.

And because the process can be repeated thousands of times across a single panel, it offers a potential route towards producing complex components at industrial scale.

At LITILIT, we believe this combination of precision and scalability will become increasingly important as AI hardware evolves.

From glass substrates to optical connections

The potential applications extend well beyond semiconductor packaging.

As AI processors become more powerful, moving information efficiently between chips is becoming a growing challenge.

Co-packaged optics, which bring optical communication components closer to processors, are one approach to addressing this problem.

These systems require microscopic channels and other precisely manufactured features that allow optical connections to be integrated into complex hardware assemblies.

Femtosecond lasers can help create those structures within glass and other materials.

Cooling presents another challenge.

Advanced processors generate substantial amounts of heat, and engineers are exploring increasingly sophisticated methods of bringing cooling fluids closer to the components that need them most.

Creating microscopic channels in glass and silicon could help support these approaches.

Across these applications, the underlying requirement is remarkably similar: manufacturing extremely small, intricate structures without damaging the surrounding material.

That’s precisely where ultrafast laser technology has an advantage.

The challenge of scaling precision manufacturing

However, developing a technology capable of producing microscopic features is only part of the challenge.

The next step is making that technology practical for industrial production.

Femtosecond lasers originated largely as scientific instruments, designed for research environments where precision often matters more than cost, complexity or manufacturing throughput.

Industrial production demands something different.

Manufacturers need equipment that is reliable, repeatable, automated and available in sufficient quantities to support large-scale operations.

At LITILIT, our objective has been to address that gap.

Our laser architecture is protected by two patents, including one developed in collaboration with Lithuania’s Center for Physical Sciences and Technology.

Rather than adapting laboratory equipment for industrial use, we designed the technology around manufacturing requirements from the outset.

That means a modular architecture, reduced component complexity and a high degree of automation.

Our new factory in Vilnius is designed to produce up to 3,000 lasers annually at full capacity.

We also intend to replicate this manufacturing model in other countries as demand grows.

The broader lesson is that precision alone is not enough.

For a technology to transform an industry, it must also be possible to manufacture and deploy it at scale.

Beyond data centres: Precision manufacturing everywhere

Although AI infrastructure is an important driver of demand, the applications for femtosecond lasers extend into many other industries.

Consumer electronics manufacturers increasingly require precision cutting and shaping techniques for glass, camera lenses and display components.

Smartphones, laptops and smart glasses all depend on components that must be manufactured within increasingly demanding tolerances.

Medical technology offers further opportunities.

Next-generation laser systems are being developed for applications including corneal reshaping in eye surgery, where precision and control are essential.

These industries may appear unrelated, but they share a common manufacturing challenge.

As products become more sophisticated, the materials used to build them become harder to process using conventional techniques.

The ability to work at microscopic scales without introducing cracks, heat damage or excessive material waste is becoming increasingly valuable.

AI’s future will be built on factory floors

The extraordinary progress of artificial intelligence has understandably focused attention on software.

We measure advances through increasingly capable models, more sophisticated reasoning and new applications that would have seemed impossible only a few years ago.

But none of these developments can continue indefinitely without corresponding advances in the physical systems supporting them.

The next generation of AI infrastructure will depend on technologies capable of manufacturing smaller, denser and more complex components at industrial scale.

That means solving challenges involving glass, silicon, advanced packaging, optical communications and thermal management.

It also means developing manufacturing tools capable of delivering microscopic precision consistently and economically.

The brain of AI was built in software laboratories.

Its body is being built on factory floors, one microscopic hole and one crack-free cut at a time.

And how quickly AI advances from here may depend as much on our ability to manufacture that body as on our ability to improve the intelligence running inside it.

About the author: Nikolajus Gavrilinas is co-founder and CEO of LITILIT, a Lithuanian manufacturer of femtosecond laser technology. His work focuses on developing ultrafast laser systems for industrial applications.

The views expressed in this article are those of the author and do not necessarily reflect the editorial position of Tech Trends.

The post AI has a brain. Who will build its body? appeared first on Tech Trends.