TROs in action: Lumentum shows two 1.6T Ara optical DSPs operating in TRO mode at ECOC. “People are always looking for ways to save power,” said Matt Sysak, CTO of Lumentum.

While transmit-receive optical DSPs remain essential for longer links inside data centers, TRO optical DSPs like Ara T provide an optimized balance of power, performance, and signal integrity for shorter rack- or row-level links.

Interconnects, Infrastructure and Power

Do six watts matter? At the scale of AI, absolutely. Interconnect revenue is projected to more than triple to over $60 billion by 2030, according to research firm LightCounting,¹ as hyperscalers and others expand infrastructures to meet the escalating demand for increasingly sophisticated AI services.

Optical module revenue (blue) will grow by ~2x by 2030 and constitute over 50% of the interconnect market, says LightCounting.

Bandwidth transitions are also accelerating, with Marvell enabling new networking generations every two years instead of every three. The ratio of interconnects to XPUs continues to rise, increasing the number of optical modules inside systems—some high-end XPUs now rely on up to 18 modules to operate at full bandwidth.²

More modules operating at higher bandwidths, however, can also translate into higher power consumption. Electricity consumption for data center networking in the U.S. alone is projected to nearly triple to 23 terawatt-hours by 2028,³ or enough to power close to 2 million homes.⁴ TRO-based designs can help curb this rise, providing a path to lower operating costs and higher ROI. Reduced power can also alleviate strain on local power grids and simplify site approval processes.

Efficiency gains as measured by picojoules per bit (pj/bit) will play a critical role. Since 2000, pj/bit has improved by 100x while total bandwidth has increased by 1000x.⁵

Interconnect Diversity

Optimizing interconnects for AI requires a diverse and evolving technology portfolio. This includes advancing existing solutions such as ZR/ZR+ coherent modules for long-distance connections, as well as developing new product categories like TRO optical DSPs, co-packaged copper, coherent-lite DSPs, active electrical cables and active copper cables.

Interconnects will also become more intelligent. Recently, Marvell unveiled RELIANT, a telemetry suite providing real-time visibility into bit error rates, signal performance, power consumption and other critical metrics to enhance uptime and network health.

Collectively, these technologies, in conjunction with a broad ecosystem of partners, give data center operators the ability to manage costs, fine-tune their networks and build an infrastructure foundation capable of delivering on the promise of AI.

1. LightCounting
2. Vlad Kozlov, LightCounting, October 2025.
3. Lawrence Berkeley National Laboratory and the U.S. Department of Energy 2024 Report on Data Center Use.
4. U.S. Energy Information Agency.
5. An Extreme Makeover for Data Centers, Marvell, 2023.

# # # 

This blog contains forward-looking statements within the meaning of the federal securities laws that involve risks and uncertainties. Forward-looking statements include, without limitation, any statement that may predict, forecast, indicate or imply future events or achievements. Actual events or results may differ materially from those contemplated in this blog. Forward-looking statements are only predictions and are subject to risks, uncertainties and assumptions that are difficult to predict, including those described in the “Risk Factors” section of our Annual Reports on Form 10-K, Quarterly Reports on Form 10-Q and other documents filed by us from time to time with the SEC. Forward-looking statements speak only as of the date they are made. Readers are cautioned not to put undue reliance on forward-looking statements, and no person assumes any obligation to update or revise any such forward-looking statements, whether as a result of new information, future events or otherwise.

Tags: AI infrastructure, Optical Interconnect, Optical DSPs, DSP, data center interconnect, AI