For most of the last decade, the optical layer connecting data centers was an IT plumbing decision, delegated to network engineers and reviewed by executives only when something broke. That era is ending.
As AI training workloads outgrow the power available at any single site, hyperscalers are splitting mega-clusters across multiple campuses, cities, and regions, and stitching them back together with high-capacity optical links so they behave as one logical data center.
The connectivity choices being made right now, largely invisible to the CFO and the board, will determine which companies can scale AI infrastructure economically over the next decade and which will find themselves locked into constrained, expensive architectures.
Optical Networking Market Development
ABI Research forecasts that Open Line System, or OLS, scale-across revenue in the United States will climb from under $1 billion in 2026 to approximately $5 billion by 2035, a trajectory that reflects how quickly distributed AI infrastructure is becoming the default rather than the exception.
Hyperscalers will remain firmly in control of this build-out domestically, accounting for nearly 70 percent of OLS links deployed, which signals that the largest cloud platforms are treating optical interconnect as core infrastructure rather than an outsourced utility.
Europe tells a more instructive story about market maturity: hyperscalers' share of activated links there is expected to fall from 59 percent in 2026 to 44 percent in 2035 as Neocloud operators gain ground, with European revenue peaking around 2033 before the market settles.
That divergence matters strategically, because it shows a technology segment that starts hyperscaler-led does not necessarily stay that way once capacity and standards mature. On the vendor side, incumbents such as Cisco, Ciena, and Nokia continue to hold ground through scale and integration depth.
However, ABI Research identifies OLS specialists -- including Edgecore, Ekinops, PacketLight, and Smartoptics -- as well positioned to gain share as demand grows for disaggregated optical infrastructure built on open standards like OpenROADM, OpenConfig, OIF, and TIP.
Outlook for Optical Networking Investment Growth
The strategic lesson here has less to do with optical transport itself and more to do with how Applied-AI investment is quietly forcing a rethink of what counts as core IT infrastructure.
When Dimitris Mavrakis of ABI Research describes connectivity as a strategic differentiator rather than a supporting layer, he is describing a pattern that experienced technology buyers have seen before: a capability that starts as a back-office concern becomes a source of competitive advantage or constraint once scale and complexity cross a threshold.
Power availability, not chip supply, is now the binding constraint on where AI compute gets built, and that reality is pushing infrastructure decisions into geographies and architectures that were not designed with this workload in mind.
Any enterprise or hyperscaler betting its AI roadmap on a single-site model is making an implicit bet that power, grid connection timelines, and real estate will cooperate. The data suggests otherwise.
For CIOs and infrastructure leaders evaluating cloud and colocation partners, the open standards trend deserves particular attention. A market moving toward OpenROADM, OpenConfig, and multi-vendor interoperability is a market that is trying to avoid repeating the vendor lock-in mistakes of earlier networking cycles.
That should factor directly into how buyers negotiate long-term capacity commitments and how they assess a provider's architectural flexibility, not just its current price per gigabit.
The point about orchestration software becoming the next competitive battleground, rather than photonic hardware alone, is worth considering. It suggests the tech vendors and cloud providers who win this decade will be the ones who can manage low-latency control across geographically distributed infrastructure, not simply the ones who can lay the most fiber.
That being said, I believe the broader takeaway for senior leadership is this: distributed does not have to mean fragmented, but only if the connectivity strategy is treated as a first-order design decision rather than an afterthought.
Chief executives approving multi-year AI infrastructure investments should be asking their CIO how power constraints are shaping site selection today, and whether the optical architecture underneath those decisions is flexible enough to adapt toward the more distributed, multi-vendor landscape later in the decade.