PicoJool Secures $27.5M Series A to Tackle AI Data Center Bandwidth Bottlenecks with Optical Interconnects

Optical connectivity startup PicoJool announced it has raised $27.5 million in a Series A funding round led by Socratic Partners, with participation from Hudson River Trading. This brings the startup's total funding to $39.5 million, following a seed round backed by Playground Global. The capital is earmarked to commercialize chip-level VCSELs (Vertical Cavity Surface Emitting Lasers) and MicroVCSELs, as well as expand the company's research, development, and engineering footprint across the United States and Taiwan. PicoJool is developing optical chips and transceiver modules designed to replace traditional copper interconnects inside hyperscale AI clusters. As distributed training and inference workloads scale across thousands of accelerators, conventional copper cabling struggles with severe signal degradation, latency penalties, and high power consumption. PicoJool's optical approach provides a direct path toward 200 Gbps per lane, enabling aggregate switch and cluster connectivity reaching up to 3.2 Tbps. This development reflects the broader trend where AI infrastructure bottlenecks have migrated from raw compute to inter-chip and inter-rack communication fabrics. As modern foundation models require synchronizing parameters across massive multi-node GPU and TPU clusters, the network fabric itself often dictates overall throughput. Startups focusing on co-packaged optics, near-packaged optics (NPO), and VCSEL lasers are becoming pivotal suppliers in the modern data center stack alongside established silicon providers. For infrastructure engineers and platform architects, the commercialization of 200G-per-lane optical interconnects means planning for significantly denser, more power-efficient server layouts without hitting the physical distance limitations of high-speed copper. Practitioners evaluating high-scale training infrastructure should closely monitor qualifications of Active Optical Cables (AOC) and NPO modules to optimize cluster power envelopes and mitigate communication overhead in distributed AI pipelines.