As AI supercomputing races toward exascale performance, the backbone of that intelligence—the data center interconnect—is under strain. Traditional optical transceivers built on externally modulated lasers (EMLs) are fast approaching their physical limits. To keep up, the industry is pivoting toward co-packaged optics (CPO) and external light sources (ELS) that can feed tomorrow’s GPU clusters with the bandwidth they crave.
Enter Sivers Semiconductors and POET Technologies, two photonics specialists joining forces to redefine how AI data centers move light. Their new partnership, announced today, combines Sivers’ distributed feedback (DFB) laser technology with POET’s Optical Interposer™ platform, promising a leap forward in scalable, power-efficient light-engine design for next-generation connectivity—think 800 Gbps to 6.4 Tbps and beyond.
Building the Optical Backbone of the AI Era
“AI is stretching the limits of existing data infrastructure,” said Vickram Vathulya, CEO of Sivers Semiconductors. “The resources needed to enable future high-speed connectivity have shifted the balance toward innovative technologies like CW lasers. With our differentiated laser portfolio, we’re positioned to capture a vastly expanded market.”
That market is massive. As hyperscale operators like Microsoft, Amazon, and Google scramble to build GPU-driven AI superclusters, internal data rates are expected to quadruple within three years, according to industry analysts. CPO architectures—where optics are integrated directly alongside chips—are seen as the next frontier for efficiency and density.
By integrating Sivers’ DFB lasers into POET’s Optical Interposer, the two companies aim to offer “plug-and-play” light engines that simplify deployment and reduce latency, heat, and cost—three critical pain points in today’s data centers.
Why This Partnership Matters
While many optics players chase incremental improvements in pluggables, Sivers and POET are targeting the transition to fully integrated optical subsystems. Their collaboration bridges both sides of the industry’s evolution:
- Upgrading pluggables with higher-speed, energy-efficient components.
- Paving the way for co-packaged optics, where ELS modules distribute light to multiple chips within a server.
“This partnership combines two complementary technologies that directly address the connectivity demands of AI and hyperscale data centers,” said Dr. Suresh Venkatesan, Chairman and CEO of POET Technologies. “We’re delivering light sources that are not just faster, but also more scalable and energy-efficient.”
The result could be a crucial piece in the race to make AI networks greener and more sustainable—a growing concern as model training consumes record amounts of power.
Looking Ahead
Sivers and POET plan to demonstrate prototype light-engine modules in the first half of 2026, with production readiness slated for the end of that year. If successful, the partnership could help redefine how hyperscale players build optical connectivity into next-gen GPU and AI accelerator architectures.
The collaboration also aligns with broader industry trends: Nvidia, Intel, and Broadcom are all investing heavily in co-packaged optics as they chase lower latency, higher throughput, and improved thermal efficiency. By combining their respective strengths in lasers and integration, Sivers and POET are positioning themselves as early movers in the AI optical connectivity market—a segment projected to exceed $10 billion by 2030.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
