Why we invested in Mixx Technologies
AI infrastructure is running into physical constraints that incremental engineering of traditional architectures can’t solve. The raw performance of GPUs has grown exponentially, but the ability to network them into harmonious clusters has not kept pace. As models grow to trillions of parameters, operating them efficiently across hundreds of XPUs (a catch-all term for any processing unit — GPUs, TPUs, custom ASICs, and other accelerators) becomes unavoidable. Today, those XPUs are typically connected with copper links (or interconnects) that push the limits of what physics allows in terms of speed, reach, heat, and power. The computing industry has understood this tension for years, but stuck with copper because it is cheap, familiar, and deeply baked into existing supply chains – eking out whatever remaining performance from copper that could be done.
The technical solution has also been understood for a decade: replace short‑reach copper with optical links inside the rack, not just long-reach links between racks and data centres, which already operate in the optical domain. Conceptually, this means moving information between components within the rack using light (photons) rather than electrical signals (electrons). Practically speaking, that means bringing silicon‑integrated optics right into chip packages, tightly coupling the sending and receiving of photons with the processors’ computing capacity, which operate in the electrical domain. This allows for better speed, capacity, and energy efficiency; in computing terms, unlocking much higher bandwidth, longer reach, and far lower energy per bit.
Many startups and incumbents have tried to adapt conventional optical transceivers to these dense environments and run into limits on cost, manufacturability, and reliability at the scale AI now demands. The hard problems now are system-level, rather than just about designing better components. That is:
Designing optical connectivity that can reach ultra‑high radix (connect to many different XPUs/Switches simultaneously);
Fit within the space and thermal constraints of a rack;
Meet the reliability demands for the highest compute efficiency, and
Be cost-competitive and producible at scale.
As ever with semiconductor advances, heterogeneous pieces of the value chain (from ASICs to optical engines and packaging) need to evolve together. This was the backdrop and our understanding of the space that got us excited about Mixx Technologies.
They are focused on one of the most challenging parts of the puzzle, the optical engine and associated fibre connectivity. These pieces are crucial to enabling high radix, low latency cluster topologies that existing approaches have struggled to deliver. This is the key to scaling infra efficiency for frontier models without blowing up cost, power, or complexity.
We invested in Mixx Technologies because they've taken the vision of cluster-scale optical connectivity and built a system design that works, performs on the metrics that matter, and has a credible path to cost competitiveness. Mixx’s product takes a system-level approach that spans the optical engine, interoperability, fibre attach solution, and thermal management inside the server.
Vivek Raghuraman and Dr. Rebecca Schaevitz bring an exceptional track record in silicon photonics. Vivek cut his teeth in Intel’s silicon photonics team, commercialising the first silicon photonics-based optical transceiver. Rebecca managed optical technology strategy for hyperscale datacentres at Corning, learning the supply chain from the inside. At Rockley Photonics, they both contributed to the world's first CPO prototype. In 2019, within Broadcom, Vivek and Rebecca were part of the founding team that persuaded Hock Tan to fund the first successful CPO program. They pioneered two generations of scale-out CPO productions in tandem with Meta, demonstrating tangible commercial value to a nascent market. What they uniquely bring to the market is several years of successfully translating advanced optics into reliable products.
Critically, Vivek & Rebecca understand that a disciplined approach to system design, manufacturing readiness, and supply chain resilience is vital. Our technical thesis is that the transition to CPO inside AI clusters is not a question of if, but when. Ultimately, the physical limits of copper and upside in performance for hyperscalers are unavoidable. The economics of AI scale and power consumption demand it.
This has the potential to have a very significant commercial outcome if Mixx Technologies executes efficiently on its remaining technical milestones and GTM. Mixx is not selling a single component into a box; it is designing an optical engine at the level of the entire rack, with the package integration, ASIC interface, fibre attach, thermal behaviour, and cable management all treated as one system. That system approach matters because hyperscalers now buy and qualify full reference architectures rather than loose parts. This means a CPO platform that drops into an Nvidia or Hyperscaler rack design can capture a much larger share of value than a standalone transceiver, and can follow that design across multiple generations of XPU clusters. In a world where CPO scaleup networking becomes a multi-billion dollar market, even a conservative market share of the optical engine in leading AI racks would translate into a very significant annual revenue potential for a company in Mixx Technologies’ position.
Our Investment Thesis at Systemiq Capital
Our fundamental thesis is straightforward: you invest in truly exceptional teams that you believe can uniquely execute on an ambitious vision. Over the past 24 months, we have relentlessly sharpened our view of the future of compute. Moore’s law‑style reliable and predictable efficiency gains are fading just as AI compute demand is exploding, turning data centres into the fastest‑growing load on the grid. Compute efficiency has therefore shifted from an internal engineering metric to a hard constraint on how quickly we can electrify, build resilient energy systems, and simultaneously allow AI to flourish. Our thesis is that the next wave of significant companies will be the teams building the enabling infrastructure to deliver more useful compute per watt, per rack, and per dollar of capex – and that is the lens through which we approached Mixx Technologies.
Mixx Technologies has demonstrated they can execute across all three crucial dimensions simultaneously; they combine the crucial attributes of a working product, excellent performance, and a path to cost competitiveness. That combination is rare. The team has the experience and network to navigate industry conservatism and turn vision into volume production. They understand that shipping to hyperscalers requires reliability, manufacturability, and relentless focus on cost reduction, not research papers or benchmarks.
We are backing Vivek and Rebecca’s vision to execute one of the most significant transitions in history for the networking of computation. We could not be more excited to partner with them as they build the definitive architecture for the world's leading AI clusters.