Chip startup Retym emerges from stealth with $180 million to boost AI infrastructure with high-speed DSPs

Retym, a chip startup focused on solving some of the messier challenges inside AI data centers, has emerged from stealth with more than $180 million raised across multiple funding rounds. The company, which has been operating quietly for the past four years, plans to launch its first product later this year.

The Series D round was led by Spark Capital, with backing from returning investors Kleiner Perkins, Mayfield, and Fidelity Investments. Spark Capital’s James Kuklinski is also joining Retym’s board.

Retym builds digital signal processing (DSP) chips—the type that helps move huge amounts of data quickly across large distances in data centers. That kind of movement has become critical as demand for AI continues to spike. Training massive AI models like ChatGPT takes thousands of chips working together, and those chips need to talk to each other fast and reliably. That’s where DSPs come in.

Right now, Marvell Technology dominates this part of the chip market. Retym wants to change that.

“The focus is on building coherent DSP chips for the next generation of AI infrastructure and cloud,” said co-founder and CEO Sachin Gandhi.

According to Gandhi, today’s data centers are running into a serious bottleneck. AI workloads aren’t just about raw compute anymore—they’re about connecting all that compute efficiently. As chips collaborate on machine learning tasks, any lag in communication slows everything down.

Retym’s chips are aimed at solving this. The first version is designed to move data over distances ranging from 10km to 120km, but it’s specially tuned for the 30km to 40km sweet spot. To avoid data loss, the chip uses a modulation technique that keeps signals clean over long distances.

“They went straight for the harder problem: longer distances,” said Navin Chaddha, Managing Partner at Mayfield.

Retym is building its chips using TSMC’s 5nm process and is currently testing early samples. The tech is tailored for high-bandwidth optical networks inside and between data centers—something AI infrastructure is hungry for.

Retym’s programmable coherent DSPs aim to become a foundational part of future AI and cloud infrastructure. The Series D funding will help the company move into production and continue developing new products. The goal is a multi-generational roadmap that keeps pace with growing network demand from AI systems.

“As AI workloads continue to grow, they’re pushing infrastructure to its limits,” Kuklinski said. “Retym’s team is uniquely positioned to address this, and we’re excited to support them as they build out the future of AI networking.”

Mayfield’s Chaddha called the team “exceptional,” and said their work could define the next wave of AI infrastructure. Kleiner Perkins partner Mamoon Hamid echoed that sentiment, saying Retym’s approach to coherent DSPs could fix some of the biggest limitations holding AI back.

The timing might be right. Analysts at Dell’Oro Group estimate global spending on data center compute and networking could top $1 trillion annually within the next decade. Retym is stepping into that market with technology meant to help move data faster, more efficiently, and with less power.

“The old boundaries between inside-the-datacenter and datacenter interconnect are starting to blur,” said Vlad Kozlov, CEO of research firm LightCounting. “Retym is entering just as the need for smarter DSPs is becoming unavoidable.”

Gandhi says the company is just getting started. “We’re working closely with customers and partners to bring our DSPs into next-gen transceiver designs. There’s more coming soon.”

Founded by Sachin Gandhi (CEO) and Dr. Roni El-Bahar (CTO), Retym is a semiconductor startup focused on solving some of the toughest challenges in AI infrastructure and cloud connectivity. The company brings together top talent in analog design, DSP, VLSI, and optical communications—backed by experienced investors—to build high-performance chips aimed at reshaping how data moves through AI data centers.