Tower Semiconductor (NASDAQ: TSEM) and NewPhotonics began high-volume shipments of laser-integrated optical chips built on Tower’s PH18DA silicon photonics platform for AI data center interconnects.
Current production covers data rates of 800G to 1.6T, with 6.4T chipsets (NPC505) targeted for volume shipments in the first half of 2027.
The deal carries execution risk: NewPhotonics is the first company scaling production on this platform, and any delay or hyperscaler spending pullback could hit Tower’s revenue timeline.
AI data centers are running out of ways to move information fast enough, and Tower Semiconductor just signed up to be part of the fix. The company announced that its collaboration with NewPhotonics has moved from development into high-volume shipments of laser-integrated, serviceable optical engine photonic integrated circuits (PICs) built for next-generation AI infrastructure. These are real, implemented shipments — not a future promise — covering data rates from 800G to 1.6T today, with more advanced 6.4T solutions planned for volume manufacturing in the first half of 2027, according to the companies.
Inside the Chip That Could Power the Next AI Wave
The products are built on Tower’s PH18DA platform, which uses heterogeneously integrated indium phosphide components to pack semiconductor optical amplifiers, modulators, lasers and photodetectors onto a single chip. That integration is designed to cut manufacturing complexity and variability while improving yield, reliability and time to market. The chips are meant to support a range of designs including Extra-Dense Packaged Optics modules and various pluggable transceiver formats, serving both scale-out and scale-up data center architectures. NewPhotonics is the first company to scale production on this specific platform, which is both the opportunity and the risk: Tower is relying on a single, first-of-its-kind partner to hit its 2027 production targets for the 6.4T NPC505 chipset.
What This Means for Your Wallet
You don’t buy optical chips directly, but you do pay for the AI services, cloud storage and streaming that run on the data centers these chips power. Faster, more energy-efficient interconnects help data center operators handle more AI workload per dollar of electricity and hardware spent, which is one of the quieter forces that can keep a lid on the computing costs eventually passed through to subscription prices for AI tools and cloud services you use.
Why It Works This Way
AI workloads require moving enormous amounts of data between chips and servers at high speed, and the connections between those components can become just as important a bottleneck as the processors themselves — which is why specialized optical interconnect technology has become a competitive battleground. Companies that reduce the cost or complexity of that connectivity can become indispensable suppliers to an entire industry. If you want to track this yourself, a simple step is to look up a company’s quarterly shipment or revenue-recognition guidance versus its stated production targets, since gaps between the two are usually where execution risk shows up first. Disclaimer: This analysis is for educational and informational purposes only and should not be construed as financial or investment advice.
What to Watch Next
Watch for updates on the first NPC505 6.4T chipset shipments expected in the first half of 2027, and any signs of hyperscaler spending pullback that could affect demand for AI infrastructure. Check the original reporting from Insider Monkey for further details on Tower Semiconductor’s roadmap.
