Retym is a Silicon Valley semiconductor company building the coherent digital signal processing (DSP) silicon that underpins long-distance data movement in modern AI infrastructure. As AI clusters scale beyond a single rack, single room or even single building, the dominant performance and cost constraint shifts from raw compute to interconnect: how quickly and cheaply data can travel between accelerators, servers and data centers. Coherent DSPs sit at the heart of optical networking, encoding and decoding light signals so that enormous volumes of data can be reliably transmitted across fiber. Retym's mission is to make that layer programmable, power-efficient and economical for the AI era.
The company's first chip is a programmable coherent DSP fabricated on TSMC's advanced 5nm process node. It is engineered to drive optical links spanning roughly 10 to 120 kilometers, with a sweet spot around 30 to 40 kilometers — the distances that matter most for connecting metro-scale clusters of AI data centers that hyperscalers increasingly operate as distributed compute fabrics. By making the DSP programmable, Retym aims to let operators adapt a single platform across multiple link distances and deployment scenarios rather than buying purpose-built parts for each.
Retym emerged from stealth in March 2025 having raised more than $180 million across multiple rounds, including a $75 million Series D led by Spark Capital with participation from Kleiner Perkins, Mayfield and Fidelity Investments. The backing from premier venture and crossover investors reflects strong conviction that interconnect silicon is among the most defensible and valuable layers of the AI buildout.
The broader thesis behind Retym is that AI scaling laws are now gated as much by networking as by GPUs. Training frontier models and serving inference at scale requires moving petabytes between machines with minimal latency and energy overhead. By delivering coherent DSPs purpose-built for AI data center connectivity, Retym positions itself as critical infrastructure for the companies racing to build the largest compute clusters on earth.
With its silicon in testing and validation and manufacturing partner TSMC behind it, Retym is targeting cloud providers, hyperscalers and networking equipment vendors that need to interconnect AI campuses efficiently as model and cluster sizes continue to grow.