iPronics Raises $125M Series B to Scale Optical Networking for AI Data Centers

iPronics has raised $125 million in Series B funding to accelerate the commercial deployment of its programmable optical networking technology for artificial intelligence (AI) data centers. The round was co-led by Maverick Silicon and Light Street Capital, with NVIDIA joining as an investor alongside Triatomic Capital, Bosch Ventures, Catalight Capital, the European Innovation Council Fund and several existing backers. The financing brings iPronics’ total funding to $177 million.

The unusually large Series B reflects a growing infrastructure problem accompanying the expansion of AI computing: adding more GPUs only helps if data can move between those processors quickly and efficiently. As clusters grow from racks of accelerators into increasingly large systems, networking is becoming as important to performance, power consumption and utilization as the underlying compute hardware.

Networking is becoming part of the AI compute problem

Modern AI training depends on enormous amounts of communication between GPUs. The larger the model and cluster, the more traffic has to move between accelerators, racks and potentially multiple facilities.

Traditional data center networks have primarily relied on copper interconnects and electronic packet switches. But as bandwidth requirements rise, these architectures require increasing numbers of high-speed electrical and optical components, adding power consumption, heat, cost and complexity.

This is one reason the AI infrastructure industry is increasingly moving toward optical connectivity. Rather than continually increasing the capacity of traditional networking equipment, companies are exploring architectures where more of the communication fabric itself operates optically.

iPronics is approaching the problem using silicon photonics-based optical circuit switching (OCS). Instead of examining and electronically forwarding individual packets in the same way as a conventional network switch, an optical circuit switch can establish direct optical paths between endpoints and dynamically change those connections as requirements shift.

How iPronics’ optical switching works

At the center of the company’s technology is iPronics ONE, a rack-mounted optical switching platform designed for AI infrastructure.

Its silicon photonic chips contain networks of what iPronics calls Programmable Unit Cells, which can steer light through different physical paths. Software controls those photonic elements, effectively making the underlying optical topology programmable rather than permanently wiring servers and accelerators into a fixed configuration.

The ONE platform combines this photonic layer with integrated monitoring, telemetry and application programming interfaces (APIs), allowing network orchestration software to change optical connections as workloads move or infrastructure conditions change.

That programmability could be particularly relevant to AI clusters because their networking requirements are not necessarily static. Different training jobs can benefit from different accelerator configurations, failed GPUs may need to be bypassed, and inference workloads can produce very different traffic patterns from large-scale training.

The current ONE product family is designed to scale from 32 to 256 ports and supports sub-millisecond reconfiguration alongside real-time optical telemetry and integration with software-defined networking environments.

From photonics research to commercial AI infrastructure

iPronics was founded in 2019 as a spin-off from the Universitat Politècnica de València, building on research into programmable silicon photonics and large-scale photonic systems. The company has since shifted its focus toward turning that research into deployable networking infrastructure for AI clusters.

Its commercialization efforts accelerated significantly over the past two years.

In January 2025, iPronics raised a €20 million Series A to advance its optical networking technology. Two months later, the company introduced the ONE-32, a 32-port silicon photonics optical circuit switch targeting scale-up and scale-out AI networking. iPronics said the system could reconfigure connections in under 300 microseconds while introducing less than 30 nanoseconds of latency.

The company has also been building the manufacturing capacity necessary to move beyond relatively small deployments. In March 2026, iPronics announced a dedicated manufacturing line with Fabrinet for packaging, assembling, testing and qualifying its ONE Series switches. At the time, the company said it was working with its first ten hyperscaler and AI-cluster original equipment manufacturer customers.

That manufacturing expansion is important because optical networking technologies have historically faced a difficult transition from promising laboratory demonstrations to reliable hardware that can be manufactured economically at data center scale.

What the $125 million Series B changes

The new capital gives iPronics substantially more resources to address that commercialization challenge.

The company plans to expand production, accelerate customer deployments and continue developing the ONE platform. It is also increasing its presence in the United States with a new office in Santa Clara, placing the company close to NVIDIA and many of the semiconductor, networking and AI infrastructure companies designing the next generation of data centers.

The participation of NVIDIA is particularly notable. NVIDIA’s GPUs increasingly operate as components within enormous computing systems rather than as standalone processors, making the interconnect between accelerators a critical component of overall system performance.

Maverick Silicon senior managing director Manish Muthal will also join the iPronics board, while former Rambus CEO Geoffrey Tate, now an adviser to Light Street Capital, will take a board seat. Young Sohn of Catalight Capital, previously Corporate President and Chief Strategy Officer at Samsung, will become an adviser to the company.

Optical networking could become a bigger part of the AI infrastructure stack

The broader significance of iPronics’ funding extends beyond a single optical switch.

The AI infrastructure race has largely been defined by GPUs and other accelerators, but simply deploying more processors does not guarantee that they will be used efficiently. Memory bandwidth, networking, power delivery and cooling are increasingly determining how much useful computing performance a data center can extract from its hardware.

Optical circuit switching offers one possible way of attacking the networking portion of that equation. A programmable optical layer could allow clusters to change their physical connectivity based on workloads, route around hardware failures and expand GPU domains without repeatedly redesigning the underlying network.

There are still questions around how widely OCS architectures will be adopted and where they will sit alongside increasingly powerful Ethernet and proprietary AI networking technologies. But the scale of iPronics’ Series B, combined with NVIDIA’s participation and the company’s move toward volume manufacturing, suggests programmable photonics is moving from an experimental data center technology toward a serious contender for the next generation of AI infrastructure.