iPronics Raises $125M to Scale Silicon-Photonics OCS

iPronics has raised $125 million in Series B financing to expand commercial deployment of its programmable optical circuit switching technology for AI infrastructure. Maverick Silicon and Light Street Capital co-led the round, with NVIDIA participating alongside Triatomic Capital, Bosch Ventures, Catalight Capital, the European Innovation Council Fund, The Tate Family Trust, Fine Structure Ventures, Amadeus Capital Partners, Build Collective and Criteria Venture Tech. The financing brings iPronics' total funding to $177 million. The capital raise matters because it signals investor conviction that optical circuit switching, or OCS, can move from lab demonstration to production-scale AI data center infrastructure at a moment when cluster sizes and power constraints are intensifying.

The company will direct the new capital toward scaling operations, expanding manufacturing and accelerating deployments of iPronics ONE, its rack-mounted optical circuit switch for AI data centers. ONE uses a solid-state silicon-photonic switch fabric to establish programmable optical connections within and across racks. Integrated control, telemetry, optical monitoring and APIs allow infrastructure operators to reconfigure Layer 1 connectivity for training, inference, cluster expansion and GPU failover without requiring packet processing inside the optical switch. The financing also expands iPronics' board and advisory team: Manish Muthal, senior managing director at Maverick Silicon, joins the board alongside Geoffrey Tate, an advisor to Light Street Capital and the former founding CEO of Rambus. Young Sohn of Catalight Capital, previously corporate president and chief strategy officer at Samsung and CEO of Inphi, becomes an advisor. iPronics has also opened an office in Santa Clara, California, where it is hiring for product strategy, customer deployment and AI infrastructure partnership positions.

The evidence for OCS value comes primarily from Google's production deployment, which the source cites as the best-documented example. Google's internally developed Apollo OCS uses 3D MEMS mirror arrays, WDM transceivers and a software-defined control system to reconfigure the Jupiter data-center network. In its 2022 disclosure, Google reported 30% higher throughput, 40% lower power consumption, 30% lower cost, 10% lower flow-completion time and 50-times less downtime than its cited alternatives. The source is careful to note these are system-level results from Google's complete network architecture rather than performance guarantees for every OCS implementation. iPronics is pursuing a different design point from large 300×300 MEMS systems aimed at spine replacement: ONE currently emphasizes a lower-radix, rack-mounted platform with solid-state photonic switching, active optical-performance control and faster reconfiguration. That could place the technology within or between racks, where density, topology adaptation and failure recovery are priorities.

The sector implication is that OCS is becoming a recognized architectural layer in AI infrastructure, but the commercial test now shifts from demonstrating a photonic switching engine to proving repeatable manufacturing, packaging yield, optical loss, crosstalk, thermal stability, control-plane integration and long-term reliability at volume. In March 2026, iPronics announced a dedicated Fabrinet line for packaging, assembly, testing and qualification and said it had initial wins with ten hyperscaler and AI-cluster OEM customers. The new financing gives it considerably more capital to execute that transition. NVIDIA's participation adds the AI platform company to iPronics' investor base, but the announcement does not disclose a supply agreement, product integration or commercial commitment.

Several limitations bound this analysis. The dossier relies on a single secondary source, Converge Digest, read in full; no primary filings or independent corroboration are available. The source text is truncated at the end, so any additional context beyond the disclosed facts is unknown. Valuation, ownership stakes, revenue, and specific customer names are not disclosed. The claim of ten initial hyperscaler and AI-cluster OEM customer wins comes from iPronics' own March 2026 announcement and has not been independently verified. What to watch: whether iPronics can convert its Fabrinet manufacturing line into volume shipments, whether NVIDIA's investor role evolves into a commercial relationship, and whether the company publishes optical performance and reliability data comparable to Google's Apollo disclosures.