What Happened

Absolics, a South Korean semiconductor company, is launching commercial production of glass substrates designed specifically for AI chip packaging. These substrates, measuring 700 micrometers to 1.4mm thick, replace traditional organic materials like fiberglass-reinforced epoxy that currently serve as the foundation for connecting multiple silicon chips.

The technology addresses a fundamental problem in modern computing: as AI workloads increase and chip packages grow larger, the heat generated causes traditional substrates to physically warp. This warping misaligns components and reduces cooling efficiency, potentially leading to chip damage or premature failure.

“As AI workloads surge and package sizes expand, the industry is confronting very real mechanical constraints that impact the trajectory of high-performance computing,” explains Deepak Kulkarni, a senior fellow at Advanced Micro Devices (AMD). “One of the most fundamental is warpage.”

Intel has also demonstrated the viability of glass substrates, successfully booting Windows on a system built with glass packaging technology in early 2025, though their early testing process broke hundreds of glass panels daily due to manufacturing fragility.

Why It Matters

This development could significantly impact AI infrastructure costs and performance. Glass substrates offer several crucial advantages over traditional materials:

Connection Density: Glass enables 10 times more connections per millimeter compared to organic substrates, allowing engineers to pack 50% more silicon chips into the same package area.

Thermal Stability: Glass maintains its shape under high temperatures, preventing the warpage issues that plague current substrates. This stability enables better cooling and prevents component misalignment.

Signal Quality: The smooth surface of glass (5,000 times smoother than organic substrates) allows for light-based signal pathways instead of power-hungry copper connections, potentially reducing energy consumption.

Scaling Potential: Glass “unlocks the ability to keep scaling package footprints without hitting a mechanical wall,” according to Kulkarni, enabling continued performance improvements in AI systems.

Background

Modern high-performance computing relies on a packaging approach that combines multiple specialized silicon chips on a single substrate. Each chip is designed for specific functions – processing, memory, graphics, or AI acceleration – and they work together as an integrated system.

This multi-chip packaging has become essential for AI applications because it allows engineers to optimize each component for its specific task while maintaining high-speed communication between chips. However, the approach has pushed traditional organic substrates to their physical limits.

As AI models become more complex and data centers scale up their computing power, the heat generated by these chip packages has grown exponentially. Traditional substrates, primarily made from fiberglass-reinforced epoxy, expand and contract under this thermal stress, causing the precise alignments needed for optimal performance to drift over time.

The semiconductor industry has been searching for alternatives that can handle higher temperatures while maintaining the precision required for advanced chip packaging. Glass emerged as a promising solution due to its superior thermal properties and the ability to create extremely smooth surfaces for signal pathways.

What’s Next

The commercial timeline for glass substrates appears aggressive but achievable. Absolics plans to begin production in 2026, with broader industry adoption expected throughout the late 2020s for data center applications.

Market projections suggest significant growth potential, with the glass substrate market expected to expand from approximately $1 billion in 2025 to $4.4 billion by 2036.

For consumers, the benefits may take longer to materialize. While the technology could eventually improve energy efficiency in laptops and mobile devices, widespread consumer adoption depends on production costs falling to competitive levels with traditional substrates.

Key challenges remain:

Manufacturing Fragility: Early Intel testing highlighted the brittleness of glass during manufacturing, though companies are working to improve handling processes.

Production Scaling: Moving from laboratory demonstrations to mass production will require significant manufacturing infrastructure investment.

Cost Competitiveness: Glass substrates must reach cost parity with organic alternatives for consumer device adoption.

The success of this technology could reshape the semiconductor packaging industry and enable continued performance improvements in AI systems at a time when traditional scaling approaches are reaching physical limits.