We’re thrilled to announce our lead investment in Celero and our partnership with CEO Nariman Yousefi, CTO Oscar Agazzi, and the rest of Celero’s exceptional team. Celero is building a new class of coherent digital signal processing (DSP) platforms to tackle one of modern computing’s most pressing challenges: the networking bottleneck that throttles AI-scale infrastructure.

The AI Revolution's Networking Bottleneck

The rise of AI has completely redefined the scale of modern computing. Training and serving today’s frontier models and AI applications requires linking thousands of GPUs / XPUs spread across racks, rows, and even entire buildings, all operating as a single, synchronized brain. At this scale, networking has quickly become a defining constraint: It’s no longer just about building faster compute chips; it’s about moving massive volumes of data across thousands (and in the future, millions) of those chips with minimal latency and near-perfect coordination. As individual data centers scale from isolated servers and racks to campus or even region-wide compute fabrics, the network has become the critical backbone and bottleneck that determines how efficiently AI infrastructure can perform.

There’s just one problem: Yesterday’s networking technology wasn’t designed to operate at this new level of scale. As clusters grow and data flows intensify, existing interconnects are hitting their physical limits. At the center of this challenge lies the digital signal processor (DSP), the tiny chip inside every optical transceiver that converts electrical signals from a GPU / XPU into pulses of light for transmission over fiber. This translation process dictates the speed, efficiency, and cost of every connection. Currently, data center architects are caught between two imperfect options:

• PAM4: The long-time workhorse of the data center. While it can be cost-effective, it hits a wall at the bandwidths and data rates tomorrow’s AI compute fabrics require (1.6T or 3.2T+), resulting in greater noise, more signal degradation, and significantly shorter reach.

• Legacy Coherent: Technology borrowed from the telecom world. While it can handle high data rates over long distances, it is over-engineered for the data center, loaded with features that make it far too costly and power-hungry to deploy at scale.

The Celero Approach: Coherence for the Data Center

Celero is bridging this gap by developing a new class of coherent DSPs designed specifically for the demands of AI-scale networking. Celero delivers the performance of telecom-grade systems with the power efficiency and cost profile required for dense data centers. Its purpose-built architecture enables high-speed optical links without the usual energy or economic tradeoffs, making coherent interconnects practical for use within and across data centers.

Delivering on this vision requires a uniquely qualified team - and Celero is one of the few in the world with the expertise to do it. The company is led by co-founder and CEO Nariman Yousefi, who previously was an SVP at Broadcom, the CEO of DSP startup ClariPhy (acquired by Inphi), and the leader of the coherent DSP businesses at both Inphi and Marvell. He’s joined by co-founder and CTO Oscar Agazzi, one of the industry’s most accomplished DSP architects, who led advanced design efforts at ClariPhy, Inphi, and Marvell.

Together, they’ve built a 100-person (and quickly growing) team with exceptional depth in DSP design, drawing top talent from Marvell, Inphi, and Broadcom. Many of these engineers helped develop the most widely deployed DSPs in use today and have a proven record of taping out complex optical technologies. This deep expertise, combined with Celero’s strong foundry relationships and innovative designs, gives the company a rare ability to bring a new generation of coherent DSPs into production.

CapitalG’s Investment in Celero

We’re proud to lead Celero’s $100 million Series B alongside Sutter Hill Ventures, Valor Equity Partners, Atreides Management, and Maverick Capital. We’re excited to partner with Nariman, Oscar, and the entire Celero team as they tackle one of the most pressing challenges in modern infrastructure and, in doing so, help unlock AI’s full potential.