NewPhotonics and Tower Semiconductor Shipped Optical Engines
The collaboration brings laser-integrated circuits to high-volume AI infrastructure production.
Updated on Sept. 28, 2026 in Semiconductors
Live Poll
Do you believe rapid expansion in AI hardware infrastructure will improve the nation's economic productivity?
NewPhotonics and Tower Semiconductor have initiated high-volume shipments of laser-integrated photonic circuits for AI infrastructure. The currently shipping lineup includes 800G and 1.6T devices.
Why it matters
The integration of optical components directly onto silicon platforms scales AI networking infrastructure to meet the increasing data throughput demands of large-scale computational clusters.
The 800G and 1.6T photonic integrated circuits utilize Tower Semiconductor’s PH18DA silicon-photonics platform. This process supports speeds reaching 1.6T, with 6.4T capacity expected in future iterations.
The players
NewPhotonics
A developer of optical engines and photonic integrated circuits designed for high-performance AI networking.
Tower Semiconductor
A specialty foundry providing analog and mixed-signal semiconductor manufacturing services, including their proprietary silicon-photonics platforms.
The details
The technology relies on heterogeneously integrated indium phosphide — a semiconductor material used for high-speed laser and light-detection applications — directly onto a silicon-photonics platform. By combining lasers, optical amplifiers, modulators, and photodetectors on a single chip, the system reduces the footprint and power required for data transmission. This approach replaces discrete optical components with a monolithic-like architecture to improve efficiency for AI workloads.
Timeline
Volume shipments of 6.4T near-packaged optics chipsets are planned for H1 2027.
The Tech Race
This development represents a critical milestone in the transition toward near-packaged optics to mitigate the bandwidth bottleneck in data centers. It competes with traditional pluggable transceiver models by embedding high-speed optical components directly into the silicon fabric.
These engines serve as the foundational infrastructure for high-bandwidth AI hardware, enabling faster data processing in large-scale server clusters. While currently shipping to infrastructure providers, these chipsets will progressively influence the latency and throughput available for future AI cloud services.
The takeaway
The industry is shifting toward heterogeneously integrated photonics to overcome the physical limits of traditional electrical copper interconnects. Watch for the 6.4T chipset production start in early 2027 as a litmus test for the scalability of the PH18DA platform in mass-market AI deployment.
What happens next
Volume shipments of 6.4T near-packaged optics chipsets are scheduled for the first half of 2027.
Further reading
For broader trends in optical chip manufacturing, see our latest coverage in Semiconductors.
Source note: This article includes information reported by Metal.
Live Poll
Do you believe rapid expansion in AI hardware infrastructure will improve the nation's economic productivity?
