Low-Latency Optical Interconnects for Distributed AI Computing Systems
Keywords:
optical interconnects, distributed AI, low-latency networking, silicon photonics, data center architecture, collective communication, sustainability, infrastructure governanceAbstract
Distributed artificial intelligence systems increasingly depend on communication fabrics that can sustain enormous bandwidth while preserving predictable end-to-end latency. As model training and inference move toward multi-node clusters, disaggregated accelerators, and geographically distributed inference services, electrical interconnects encounter growing pressure from energy density, reach limitations, and switching complexity. Optical interconnects offer a compelling system-level alternative because they can provide high bandwidth, low propagation loss, wavelength multiplexing, and architectural flexibility. However, the transition from laboratory photonics to production AI infrastructure is not merely a component substitution. It reshapes topology design, scheduling, failure recovery, resource governance, procurement, and sustainability. This paper examines low-latency optical interconnects for distributed AI computing systems through an interdisciplinary systems lens. It analyzes the latency sensitivity of collective communication, pipeline parallelism, and distributed inference; evaluates optical architectures such as silicon photonic links, wavelength-routed fabrics, and frequency-referenced multiplexing; and considers trade-offs involving switching, buffering, reach, power, and resilience. The discussion extends to deployment models, operational governance, standards, supply chains, and policy implications. The central argument is that optical interconnects can reduce communication bottlenecks and improve energy efficiency, but their benefits depend on coherent design across hardware, network control, workload scheduling, and institutional procurement. Without such integration, optical fabrics may reproduce existing inequities, lock-in, and environmental externalities while failing to deliver the expected latency guarantees.
References
1. Dean, J., & Barroso, L. A. (2013). The tail at scale. Communications of the ACM, 56(2), 74–80. https://doi.org/10.1145/2408776.2408794
2. Barroso, L. A., Hölzle, U., & Ranganathan, P. (2019). The datacenter as a computer: Designing warehouse-scale machines (3rd ed.). Springer. https://doi.org/10.1007/978-3-031-01761-2
3. Shalf, J. (2020). The future of computing beyond Moore's Law. Philosophical Transactions of the Royal Society A, 378(2166), 20190061. https://doi.org/10.1098/rsta.2019.0061
4. Jouppi, N. P., Young, C., Patil, N., Patterson, D., Agrawal, G., Bajwa, R., Bates, S., Bhatia, S., Boden, N., Borchers, A., Boyle, R., Cantin, P.-L., Chao, C., Clark, C., Coriell, J., Daley, M., Dau, M., Dean, J., Gelb, B., ... Yoon, D. H. (2017). In-datacenter performance analysis of a tensor processing unit. ACM SIGARCH Computer Architecture News, 45(2), 1–12. https://doi.org/10.1145/3140659.3080246
5. Narayanan, D., Harlap, A., Phanishayee, A., Seshadri, V., Devanur, N. R., Ganger, G. R., Gibbons, P. B., & Zaharia, M. (2019). PipeDream: Generalized pipeline parallelism for DNN training. In Proceedings of the 27th ACM Symposium on Operating Systems Principles (pp. 1–15). ACM. https://doi.org/10.1145/3341301.3359646
6. Sergeev, A., & Del Balso, M. (2018). Horovod: Fast and easy distributed deep learning in TensorFlow. arXiv. https://arxiv.org/abs/1802.05799
7. Rajbhandari, S., Rasley, J., Ruwase, O., & He, Y. (2020). ZeRO: Memory optimizations toward training trillion parameter models. In Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis (pp. 1–16). IEEE. https://doi.org/10.1109/SC41405.2020.00024
8. Li, S., Zhao, Y., Varma, R., Salpekar, O., Noordhuis, P., Li, T., Paszke, A., Smith, J., Vaughan, B., Damania, P., & Chintala, S. (2020). PyTorch Distributed: Experiences on accelerating data parallel training. Proceedings of the VLDB Endowment, 13(12), 3005–3018. https://doi.org/10.14778/3415478.3415530
9. Sapio, A., Canini, M., Ho, C. Y., Nelson, J., Kalnis, P., Kim, C., Krishnamurthy, A., Moshref, M., Ports, D. R. K., & Asanović, K. (2021). Scaling distributed machine learning with in-network aggregation. In Proceedings of the 18th USENIX Symposium on Networked Systems Design and Implementation (pp. 785–808). USENIX Association.
10. Al-Fares, M., Loukissas, A., & Vahdat, A. (2008). A scalable, commodity data center network architecture. ACM SIGCOMM Computer Communication Review, 38(4), 63–74. https://doi.org/10.1145/1402946.1402967
11. Miller, D. A. B. (2009). Device requirements for optical interconnects to silicon chips. Proceedings of the IEEE, 97(7), 1166–1185. https://doi.org/10.1109/JPROC.2009.2014298
12. Benner, A. F., Ignatowski, M., Kash, J. A., Kuchta, D. M., & Ritter, M. B. (2005). Exploitation of optical interconnects in future server architectures. IBM Journal of Research and Development, 49(4/5), 755–775. https://doi.org/10.1147/rd.494.0755
13. Taubenblatt, M. A. (2012). Optical interconnects for high-performance computing. Journal of Lightwave Technology, 30(4), 448–457. https://doi.org/10.1109/JLT.2011.2167952
14. Zhou, Z., Wei, J., Luo, Y., Clark, K. A., Sillekens, E., Deakin, C., ... & Liu, Z. (2023). Communications with guaranteed bandwidth and low latency using frequency-referenced multiplexing. Nature Electronics, 6(9), 694-702.
15. Kachris, C., & Tomkos, I. (2012). A survey on optical interconnects for data centers. IEEE Communications Surveys & Tutorials, 14(4), 1021–1036. https://doi.org/10.1109/SURV.2011.122111.00069
16. Biberman, A., & Bergman, K. (2012). Optical interconnection networks for high-performance computing systems. Reports on Progress in Physics, 75(4), 046402. https://doi.org/10.1088/0034-4885/75/4/046402
17. Baliga, J., Ayre, R. W. A., Hinton, K., & Tucker, R. S. (2011). Green cloud computing: Balancing energy in processing, storage, and transport. Proceedings of the IEEE, 99(1), 149–167. https://doi.org/10.1109/JPROC.2010.2060451
18. Tucker, R. S. (2011). Green optical communications—Part I: Energy limitations in transport. IEEE Journal of Selected Topics in Quantum Electronics, 17(2), 245–260. https://doi.org/10.1109/JSTQE.2010.2051216
19. Rickman, A. (2014). The commercialization of silicon photonics. Nature Photonics, 8(8), 579–582. https://doi.org/10.1038/nphoton.2014.175
20. Thomson, D., Zilkie, A., Bowers, J. E., Komljenovic, T., Reed, G. T., Vivien, L., Marris-Morini, D., Cassan, E., Virot, L., Fédéli, J.-M., Hartmann, J.-M., Schmid, J. H., Xu, D.-X., Boeuf, F., O'Brien, P., Mashanovich, G. Z., & Nedeljkovic, M. (2016). Roadmap on silicon photonics. Journal of Optics, 18(7), 073003. https://doi.org/10.1088/2040-8978/18/7/073003
21. Crawford, K. (2021). Atlas of AI: Power, politics, and the planetary costs of artificial intelligence. Yale University Press.
22. International Energy Agency. (2021). Data centres and data transmission networks. IEA. https://www.iea.org/reports/data-centres-and-data-transmission-networks
23. Cisco. (2020). Cisco annual internet report (2018–2023) white paper. Cisco Systems. https://www.cisco.com/c/en/us/solutions/collateral/executive-perspectives/annual-internet-report/white-paper-c11-741490.html
24. Reed, G. T., Mashanovich, G., Gardes, F. Y., & Thomson, D. J. (2010). Silicon optical modulators. Nature Photonics, 4(8), 518–526. https://doi.org/10.1038/nphoton.2010.179
25. Smit, M., van der Tol, J., & Hill, M. (2012). Moore's law in photonics. Laser & Photonics Reviews, 6(1), 1–13. https://doi.org/10.1002/lpor.201100001
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Journal of Data Intelligence and AI Systems
This work is licensed under a Creative Commons Attribution 4.0 International License.
This article is published under the Creative Commons Attribution 4.0 International License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
