NewPhotonics and Tower Semiconductor ship 800G-1.6T photonic ICs for AI
NewPhotonics and Tower Semiconductor have moved their silicon photonics collaboration into high-volume production, shipping laser-integrated photonic integrated circuits (PICs) designed for the optical interconnect layer of AI infrastructure. The announcement marks a step-change from engineering samples to commercial-scale supply, with current products supporting data rates of 800G and 1.6T and a roadmap extending to 6.4T.
The chips are built on Tower's PH18DA silicon photonics platform, which allows heterogeneous integration of indium phosphide (InP) components directly onto a silicon photonic substrate. This means lasers, semiconductor optical amplifiers (SOAs), modulators, and photodetectors are combined on a single monolithic PIC rather than assembled from separate die. The companies say the approach improves manufacturing yield, reduces assembly variability, and shortens the path from wafer to qualified module.
What is shipping and what is coming
The production devices are the NewPhotonics NPG102 PICs, which use a 200G-per-lane architecture to reach 800G and 1.6T aggregate. They are compatible with a range of external form factors: FRO, LRO, LPO, and Extra-Dense Packaged Optics (XPO) pluggable transceivers, as well as Near-Packaged Optics (NPO) socket pluggable designs. The second generation, the NPC505 chipset targeting 6.4T under the CPX-MSA standard, is on track for volume shipment in the first half of 2027.
Doron Tal, SVP and GM of NewPhotonics, said the partnership was translating "laser-integrated photonics into a practical choice and volume availability," citing power efficiency and bandwidth density as the primary customer requirements the platform addresses. Tower's Ed Preisler, VP and GM of its RF business unit, described NewPhotonics as "the first to bring heterogeneous laser-integrated optical engines on the PH18DA platform into high-volume production."
Both companies will exhibit at ECOC 2026 in Malaga, Spain, from 21 to 23 September, where NewPhotonics will demonstrate the NPG102 and NPC505 products.
Market context and competitive landscape
The shift to optical interconnect inside AI clusters is accelerating as GPU-to-GPU bandwidth requirements outpace the reach of copper cables at scale. Hyperscalers and neo-cloud operators building out scale-out (rack-to-rack) and scale-up (GPU-to-GPU within a system) fabrics are increasingly specifying pluggable optics and, in newer architectures, co-packaged optics (CPO) to cut power consumption and latency.
The silicon photonics foundry landscape is competitive but concentrated. Intel and Cisco (via its Acacia acquisition) operate at volume in pluggable transceiver PICs, while GlobalFoundries and imec offer SiPho process platforms to fabless customers. Tower's PH18DA differentiates on heterogeneous laser integration, a capability that many platforms still route around by sourcing external laser chips. If the yield and reliability claims hold at volume, that is a meaningful process advantage in a market where packaging complexity is a persistent cost driver.
The CPX-MSA standard that the NPC505 targets is a multi-source agreement aimed at enabling socket-pluggable near-packaged optics across multiple board and chassis suppliers. Industry adoption of such standards tends to be the gating factor for volume ramp; the fact that NewPhotonics is citing CPX-MSA compliance is a signal that it expects the standard to gain traction with OEMs ahead of the 1H27 ramp.
Regulatory and supply-chain read-across
Tower Semiconductor's manufacturing footprint spans Israel, the United States, and Japan, with a shared 300mm facility in Agrate, Italy. That geographic spread gives some insulation against single-country supply disruption, though Israel-based fabs remain subject to geopolitical risk disclosures that Tower flags in its SEC filings. US export-control rules around advanced semiconductor equipment (administered by the Bureau of Industry and Security) have not to date restricted silicon photonics production at Tower's process nodes, but customers building AI infrastructure at hyperscale will monitor BIS guidance as restrictions evolve.
The 6.4T roadmap, if delivered on schedule, would place NewPhotonics and Tower at the front of the pluggable optics curve at a time when AI cluster bandwidth requirements are doubling roughly every 18 months.