Mixx Technologies raises $33M to advance optical infrastructure for AI systems

Mixx Technologies, a startup building optical connectivity solutions for artificial intelligence (AI) infrastructure, has raised $33 million in a Series A round. The investment was led by ICM HPQC Fund, with participation from TDK Ventures, Systemiq Capital, Banpu Innovation & Ventures, G Vision Capital, Ajinomoto Group Ventures, AVITIC Innovation Fund, and other strategic backers.

The round comes as AI model sizes and data movement requirements increase sharply, placing new performance and energy demands on data-center networks.

Mixx said the new capital will support product development, expand its global presence, and scale research centers in the United States, India, and Taiwan. The company also plans to deepen partnerships across the AI infrastructure ecosystem.

Founded by Vivek Raghuraman and Rebecca K Schaevitz, engineers who previously worked on silicon-photonics products at Intel and optical-interconnect systems at Broadcom, Mixx Technologies is developing a platform that integrates photonics, advanced packaging, and system-level architecture. The company aims to address interconnect bottlenecks that limit the performance and scalability of large AI workloads.

Headquartered in San Jose, Mixx Technologies operates in the US, India, and Taiwan, and focuses on optical-based high-bandwidth solutions for large-scale AI inference models.

Chief executive and Co-founder Vivek Raghuraman said the shift toward large-scale AI models requires a rethink of how data moves within data centers. “What once centered on link speeds and component efficiency must now account for system-wide power, latency, and reliability,” he said in a statement.

Mixx, he added, is focusing on optimising end-to-end data movement for next-generation computing environments.

A core part of Mixx’s technology is HBxIO, a silicon-integrated optical engine that serves as a communication layer for AI infrastructure. Co-founder and chief product officer Rebecca Schaevitz said the approach “flattens the network” by integrating optics directly with silicon, helping eliminate bottlenecks as AI compute scales.

According to the company, its platform enables several architectural improvements:

• Switchless clusters: Increase the number of connections and improve efficiency for running large AI models.
• 3.5D integration: Places optical components closer to the chip to make data travel faster and use less power.
• Disaggregated fabric: Lets compute, memory, and accelerator units connect and scale more flexibly.
• Open standards: Ensures the technology can work easily with existing data-center systems.

Optical interconnects and silicon photonics have quickly become a focal point in AI infrastructure as data-center operators search for faster, more energy-efficient ways to move data between chips. As AI models swell in size, traditional electrical links are hitting their limits, opening the door for companies promising higher bandwidth and lower power use through light-based systems.

A wave of startups is chasing the opportunity. Lightmatter is building photonic interposers that use light to shuttle data across chips. Ranovus and Nubis Communications are developing optical-engine chiplets that sit closer to AI processors to speed up data flow. Dust Photonics focuses on silicon-photonics transceivers that cut power draw, while Oriole Networks is working on photonic switching that could reduce dependence on conventional electrical networking.

Their approaches vary, but the goal is the same: replace or supplement copper with optical links that can handle the growing demands of modern AI. It’s a fast-moving segment that is pushing data-center design away from incremental networking upgrades toward a broader rethink of how chips communicate at scale.