Tower Semiconductor ships 5 million coherent photonic circuits to Marvell for AI data centers

Tower Semiconductor ships 5 million coherent photonic circuits to Marvell for AI data centers

The milestone shipment highlights silicon photonics as the backbone technology for next-generation AI infrastructure connectivity

Tower Semiconductor has crossed a significant production threshold, shipping over 5 million coherent photonic integrated circuits to Marvell Technology’s customers. The components are designed for AI data center interconnect networks, the high-speed optical highways that link massive GPU clusters together.

What silicon photonics actually does here

Tower’s coherent PICs are manufactured on what the company calls its high-volume silicon photonics platform. The “coherent” part matters because it refers to a modulation technique that squeezes more data into each optical signal, making it particularly suited for the long-reach, high-bandwidth connections that AI data center interconnect networks demand.

The chips also incorporate advanced optical packaging methods and integration of non-silicon materials, which is industry speak for combining different semiconductor materials on a single chip to optimize both the electronic and photonic functions.

Advertisement

Tower’s broader photonic push

In February 2026, Tower announced work on 1.6T silicon photonics with NVIDIA, targeting the next tier of optical throughput for AI networking. The “1.6T” refers to 1.6 terabits per second, a speed class that the industry views as essential for keeping pace with the scaling demands of large language models and other AI workloads.

Tower has also been collaborating with Scintil Photonics on validating DWDM lasers, which stands for dense wavelength-division multiplexing. This technology crams multiple data streams onto different wavelengths of light traveling through a single fiber, dramatically increasing total bandwidth without laying more cable.

And then there’s Coherent, another partner in Tower’s growing photonics ecosystem.

Marvell, for its part, has been investing heavily in co-packaged optics, or CPO. This approach integrates optical components directly alongside the switching silicon inside data center equipment, rather than treating optics as a separate pluggable module. It’s a design philosophy that reduces power consumption and latency while boosting bandwidth density. Tower’s manufacturing capabilities feed directly into that strategy.

What this means for investors

Tower didn’t disclose specific financial details tied to this achievement, so it’s impossible to quantify exactly what these shipments mean for the company’s revenue mix.

The competitive landscape is worth watching closely. Companies like GlobalFoundries and TSMC have also been investing in photonics capabilities. But Tower’s head start in high-volume production, combined with its growing roster of design partners, gives it a meaningful advantage in a market that rewards reliability and manufacturing maturity over raw scale.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.

Tower Semiconductor ships 5 million coherent photonic circuits to Marvell for AI data centers

Tower Semiconductor ships 5 million coherent photonic circuits to Marvell for AI data centers

The milestone shipment highlights silicon photonics as the backbone technology for next-generation AI infrastructure connectivity

Tower Semiconductor has crossed a significant production threshold, shipping over 5 million coherent photonic integrated circuits to Marvell Technology’s customers. The components are designed for AI data center interconnect networks, the high-speed optical highways that link massive GPU clusters together.

What silicon photonics actually does here

Tower’s coherent PICs are manufactured on what the company calls its high-volume silicon photonics platform. The “coherent” part matters because it refers to a modulation technique that squeezes more data into each optical signal, making it particularly suited for the long-reach, high-bandwidth connections that AI data center interconnect networks demand.

The chips also incorporate advanced optical packaging methods and integration of non-silicon materials, which is industry speak for combining different semiconductor materials on a single chip to optimize both the electronic and photonic functions.

Advertisement

Tower’s broader photonic push

In February 2026, Tower announced work on 1.6T silicon photonics with NVIDIA, targeting the next tier of optical throughput for AI networking. The “1.6T” refers to 1.6 terabits per second, a speed class that the industry views as essential for keeping pace with the scaling demands of large language models and other AI workloads.

Tower has also been collaborating with Scintil Photonics on validating DWDM lasers, which stands for dense wavelength-division multiplexing. This technology crams multiple data streams onto different wavelengths of light traveling through a single fiber, dramatically increasing total bandwidth without laying more cable.

And then there’s Coherent, another partner in Tower’s growing photonics ecosystem.

Marvell, for its part, has been investing heavily in co-packaged optics, or CPO. This approach integrates optical components directly alongside the switching silicon inside data center equipment, rather than treating optics as a separate pluggable module. It’s a design philosophy that reduces power consumption and latency while boosting bandwidth density. Tower’s manufacturing capabilities feed directly into that strategy.

What this means for investors

Tower didn’t disclose specific financial details tied to this achievement, so it’s impossible to quantify exactly what these shipments mean for the company’s revenue mix.

The competitive landscape is worth watching closely. Companies like GlobalFoundries and TSMC have also been investing in photonics capabilities. But Tower’s head start in high-volume production, combined with its growing roster of design partners, gives it a meaningful advantage in a market that rewards reliability and manufacturing maturity over raw scale.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.