Opens in a new tab

Celero Raises $275M and Validates Industry's First 2nm Coherent DSP Silicon ›

Building the Optical Foundation for the Future AI Bandwidth Scaling

Company Blog

4 min read

Celero was founded around a fundamental shift taking place in AI infrastructure. As compute becomes more powerful, connecting that compute becomes increasingly difficult. Today’s largest AI systems combine thousands of GPUs, TPUs and custom accelerators into enormous distributed computing platforms. These accelerators must continuously exchange massive amounts of data across high-speed networks. In effect, the data center is evolving into one giant distributed supercomputer. And that means networking is no longer supporting the AI system from the outside. It is becoming part of the system itself.

Celero was created to address that transition. The company was founded in 2024 by Nariman Yousefi and Oscar Agazzi, two semiconductor leaders who have worked together across multiple generations of networking and DSP technology at Broadcom, ClariPhy, Inphi and Marvell. Celero represents their fifth company working together and builds on decades of experience developing networking, analog and DSP technologies.

But Celero was not created simply to build another generation of existing optical DSPs. The company’s starting point was the architecture required as AI networks move toward significantly higher bandwidth, greater distances and dramatically larger accelerator clusters. That led Celero toward coherent connectivity.

Traditional coherent DSPs were primarily designed for metro and long-haul networks. Those systems solved an enormously difficult problem: transmitting data reliably across hundreds or thousands of kilometers of fiber. But the requirements of AI infrastructure are different. AI demands extremely high bandwidth, but also low power and low latency. It must operate economically at data center volumes. And it must scale across networking environments ranging from accelerator systems to racks, clusters, buildings and campuses.

Rather than adapting a traditional coherent architecture, Celero started again from the application requirements. The company developed a low-power, low-latency DSP architecture specifically for AI infrastructure along with a new class of advanced analog technology designed to scale performance and bandwidth across multiple generations. These are the two fundamental technology innovations behind Celero: a ground-up DSP architecture optimized for low power and latency, and advanced analog technology built to scale with the increasing bandwidth requirements of AI.

That architectural approach is important because AI networking is evolving extraordinarily quickly. Today’s emerging 1.6T networks are only one step in the transition. As accelerator density increases and AI clusters become larger, networking is already moving toward next-generation 3.2T connectivity. Celero has designed its architecture around that progression rather than a single product generation. The company has now validated the industry’s first 2nm coherent DSP silicon with advanced analog technology, establishing a scalable foundation spanning 1.6T solutions through next-generation 3.2T AI optical interconnects.

The significance is not simply that the chip uses an advanced process node. It is that Celero has validated the critical technologies behind the architecture in silicon. This major technology validation represents the most important technical checkpoint to date: the difficult, fundamental elements of the architecture have now been demonstrated, while the company moves through the remaining commercialization and production-readiness steps.

As AI infrastructure scales, optical networking must solve two physical challenges simultaneously. The first is capacity. More accelerators generate more traffic, requiring substantially greater amounts of data to move over the available fiber. The second is distance. AI systems are expanding beyond individual racks and rows into increasingly large clusters, buildings and campuses. As bandwidth increases, the ability of conventional optical technologies to maintain those speeds over distance becomes more difficult.

Coherent technology addresses both. It enables substantially greater information density over optical fiber while compensating for transmission impairments that otherwise limit reach. That makes coherent increasingly relevant not only to traditional data center interconnect applications but across the AI networking fabric itself—from scale-up and scale-out to scale-across.

The company is using its new $275 million Series C financing to accelerate R&D, manufacturing readiness and the broader product roadmap. The round brings total capital raised to $415 million and values Celero at more than $3 billion.

Celero’s goal is to remove that constraint. By combining a new low-power, low-latency coherent DSP architecture, advanced analog technology and leading-edge 2nm silicon, Celero is building an optical platform intended to scale as AI networking moves from 1.6T to 3.2T and beyond.