🧠 Food for thought

Implications, context, and why it matters.

This funding backs manufacturing, not early research

• Ayar Labs’ work began in a DARPA-funded project that led to a proof-of-concept microprocessor with optical input/output (I/O) in a 2015 Nature paper 1.
• The company now works within mainstream chip development, moving from lab results to deployable parts.
• ASIC (application-specific integrated circuit) design firms including Alchip and GUC are building reference platforms with Ayar’s optical chiplets, which are small modular pieces of a processor package 23.
• Ayar Labs is also shifting its optical-engine roadmap to TSMC’s advanced COUPE platform, going beyond earlier work with GlobalFoundries 4.

Optical interconnects can extend scale-up past copper limits

• Ayar Labs aims to swap copper interconnects for optical links to boost bandwidth and cut energy use in AI scale-up, changing how systems connect inside data centers.
• High-speed copper interconnects such as NVLink reach about two meters, which keeps scale-up within one or two racks 4.
• Co-packaged optics, which integrate optical links alongside compute in the same package, could let thousands of GPUs across several racks run as one system.
• New layouts become possible, including memory pooling across a full data hall, where simulations estimate two to three times higher AI training throughput 2.
• This could support larger AI models that current data movement bottlenecks make impractical.