As artificial intelligence clusters grow larger and more complex, the ability to move data efficiently between chips has become as critical as the processing power itself. On September 21, 2026, ams OSRAM announced a significant advancement in optical interconnect technology that could reshape how AI infrastructure handles the escalating demands of training and inference workloads. The company introduced a next-generation 850nm Thin-Film VCSEL platform specifically designed for AI scale-up networks, alongside a connectorized multi-core fiber solution developed with ecosystem partner BizLink. The news, reported via NEWMEDIAWIRE, underscores a pivotal shift toward highly parallel “wide-and-slow” optical architectures that promise drastic improvements in power efficiency, latency, and scalability.
The core of the breakthrough lies in ams OSRAM's Thin-Film microVCSEL arrays, which feature dense 25 µm pitch and addressability, enabling highly parallel optical links. In testing, the platform achieved error-free 32 Gb/s NRZ operation with an energy efficiency of approximately 0.25 pJ/bit—a level that significantly reduces the power burden of data transmission in AI clusters. Reliability validation showed no failures after more than 2,000 hours under elevated junction temperatures and current overstress conditions. These results suggest the technology is robust enough for deployment in demanding data center environments. By distributing data across many lower-speed optical lanes rather than pushing ever-higher speeds through fewer channels, the wide-and-slow approach reduces system complexity and enables scalable bandwidth growth while lowering overall power consumption.
This development matters because AI scale-up networks—the interconnects that link thousands of processors within a single cluster—are increasingly becoming a bottleneck. Traditional optical interconnects struggle to keep pace with the bandwidth and energy demands of modern AI training. ams OSRAM's platform, which builds on the same core technology used in its EVIYOS™ adaptive headlamp solutions, demonstrates how photonics expertise from automotive lighting can be repurposed for data movement. The company has successfully integrated microVCSELs on silicon TSV substrates and is now advancing fully integrated 3D photonics stacks that combine microVCSEL arrays, micro-photodiode arrays, and mixed-signal CMOS electronics. This integration path could lead to ultra-high-bandwidth optical engines that are manufacturable at scale.
The partnership with BizLink is equally important. BizLink's expertise in fiber-array integration and optical packaging has enabled a connectorized multi-core fiber demonstrator that will be shown privately at ECOC 2026 in Malaga. The collaboration highlights the need for a complete ecosystem—spanning testing, fiber infrastructure, connectors, packaging, and system integration—to accelerate industry adoption. As Ashkan Seyedi, Vice President and General Manager of Optical Interconnects at ams OSRAM, noted, “AI infrastructure requires a fundamentally different approach to moving data and wide-and-slow optical architectures deliver a step-change in power efficiency, latency and scalability.” Hendrik Coldenstrodt, CTO of BizLink Group's Computing and Transportation Business Group, added that the demonstrator validates the potential of highly parallel optical interconnects for future AI infrastructure.
The implications extend beyond performance metrics. As AI clusters consume ever more energy, the industry faces mounting pressure to reduce carbon footprints and operational costs. Optical interconnects that operate at 0.25 pJ/bit could dramatically cut the power needed for data transmission, which currently accounts for a substantial portion of data center energy use. Moreover, the compatibility of ams OSRAM's VCSEL platform with standard multimode fiber infrastructure means existing data centers could potentially upgrade without entirely replacing their cabling. For more details on the technology, ams OSRAM directs interested parties to its dedicated technology page. The company also maintains a presence on social media, including LinkedIn and YouTube, where it shares updates on its photonics developments.
For an industry racing to build larger AI models, ams OSRAM's announcement signals that the optical interconnect layer is evolving rapidly to meet the challenge. By leveraging its proven Digital Photonics platform and fostering partnerships like the one with BizLink, the company is positioning itself at the center of a critical transition. If wide-and-slow architectures gain traction, they could become the backbone of sustainable, high-performance AI infrastructure for years to come. The full release is available on www.newmediawire.com.
