PicoJool has introduced 200G VCSELs aimed at scale-up AI data center interconnects, with chip-level sampling planned for next quarter. The company’s 200G VCSEL bandwidth exceeds 37 GHz, and it’s targeting use in pluggable, near-packaged optics (NPO), and co-packaged optics (CPO) designs for AI systems.

The initial sampling lineup includes quad 100G, quad 200G, and 32x50G NRZ microVCSEL (uVCSEL) devices, which PicoJool describes as intended for “short and wide” applications. PicoJool also says it integrates its optical chips into massively parallel pluggable modules for large-scale AI systems.

For data center engineers, the practical question is less “can optical go faster?” and more “can it ship in volume at predictable cost?” If PicoJool can deliver high-volume VCSEL parts that fit into the packaging approaches now being explored for AI fabrics, it could widen the set of places operators consider optics as an alternative to copper for scale-up links. But the difference between a compelling lab device and an operator-ready supply chain is manufacturing execution.

PicoJool’s manufacturing plan centers on gallium arsenide (GaAs) foundries, including WIN Semiconductor. PicoJool says its 200G designs and process recipes have been transferred to WIN and other GaAs-focused foundries. The company argues that GaAs-based VCSELs are “unconstrained in production capacity,” positioning that supply chain as a way to avoid bottlenecks seen in other laser approaches. WIN Semiconductor is described as a high-volume VCSEL producer for 3D sensing applications, with more than a billion chips shipped over the past decade.

“Our partnership with WIN Semiconductor has been very fruitful as we get ready to release a series of VCSEL products for high volume manufacturing,” said Al Yuen, founder and CEO of PicoJool.

PicoJool expects sampling next quarter and a high-volume ramp in early 2027.

Source: PicoJool.