LITILIT CEO: Femtosecond lasers build AI infrastructure
An op-ed by Nikolajus Gavrilinas, CEO and co-founder of LITILIT, examines manufacturing obstacles that hinder construction of physical infrastructure for advanced AI. He focuses on precision shortfalls of conventional tools when producing critical hardware elements.
Companies such as OpenAI, Meta and Anthropic advance AI models that write code, reason through problems and sustain conversations. These systems require supporting physical infrastructure including data centers, chips and related components. Gartner forecasts nearly $1 trillion in AI infrastructure spending this year, followed by a further $6 trillion between 2027 and 2030. Funds will support AI chips, high-bandwidth memory, glass substrates, optical fiber, photonic elements and cooling systems inside data centers.
Production of through-glass vias creates persistent bottlenecks. These microscopic holes drilled through glass and filled with metal enable vertical signal and power transmission. Drilling often produces cracks and inconsistent diameters. TrendForce projects that glass core substrates will not achieve commercial-scale output until after 2030.
Femtosecond lasers address glass processing limits by emitting pulses lasting one quadrillionth of a second. Material removal occurs before heat disperses into surrounding areas, yielding clean holes free of cracks or melting. The process supports repeated precision drilling across entire panels.
High-bandwidth memory production stacks 12 to 16 chips, each thinned to tens of micrometers, a dimension thinner than human hair and prone to fracture under mechanical saws. AI circuit boards employ hard low-loss materials that rapidly degrade conventional drill bits.
Co-packaged optics require microscopic light-guiding channels formed inside glass to position optical connections adjacent to processors. Integrated cooling channels in glass and silicon allow liquid flow near high-heat zones. Femtosecond lasers also cut wafers thinner than hair and drill hard board materials to exact tolerances where standard equipment fails.
Applications extend beyond data centers to consumer electronics such as smartphones, laptops and smart glasses that run AI models locally. Lasers cut cover glass, camera lenses and display components without inducing cracks. Medical uses include corneal reshaping during eye surgery. Most current femtosecond lasers originated as scientific instruments and remain difficult to manufacture at volume.
LITILIT developed a modular laser architecture protected by two patents, one created jointly with Lithuania’s Center for Physical Sciences and Technology. The design emphasizes reduced component complexity and extensive automation suited to industrial settings. Full-capacity operation at the new Vilnius factory targets annual production of 3,000 lasers, with replication planned in additional countries.
Software laboratories created AI computational capabilities. Realization of continued scaling now depends on factory-floor advances that deliver microscopic holes and crack-free cuts at sufficient speed and volume.
