Photo by Brett Sayles on Pexels
The AI Bandwidth Tsunami Is Here — and Optical Networks Are in the Hot Seat
Artificial intelligence is no longer just a buzzword driving software innovation — it is rapidly becoming one of the most powerful forces reshaping physical network infrastructure. According to Nokia’s Rob Shore, head of optical networks solution marketing at Nokia Network Infrastructure, AI is fundamentally altering how much bandwidth is required between locations, and more importantly, where that bandwidth needs to flow. The result is a seismic shift that is pushing optical networks into what Nokia is calling the “scale-across” era.
Unlike previous waves of network demand — which largely focused on scaling capacity upward within individual facilities — the AI era is characterized by massive lateral data movement. Training large language models, running inference workloads, and synchronizing distributed AI clusters all require enormous, low-latency data transfers across data centers, between cloud regions, and through interconnected enterprise environments. This “east-west” traffic pattern is straining optical infrastructure in ways that conventional network architectures were simply not designed to handle.
What the ‘Scale-Across’ Era Actually Means
The terminology Nokia is introducing reflects a real architectural pivot. Traditional telecom and enterprise networks were primarily built to scale “up” — adding capacity vertically within a node or facility. But AI changes the calculus entirely. Distributed GPU clusters powering AI model training can span multiple data centers, sometimes across different geographic regions. Each of those nodes must communicate with extraordinary speed and volume, generating bandwidth demands between locations that dwarf anything seen in prior compute generations.
Shore’s observations align with broader industry data. According to recent analyst estimates, data center interconnect (DCI) traffic is expected to grow at a compound annual growth rate exceeding 25% through 2028, driven primarily by AI and machine learning workloads. Hyperscalers like Microsoft, Google, and Amazon Web Services are already investing aggressively in private optical transport infrastructure to keep pace — a trend that is creating both opportunity and urgency for optical networking vendors like Nokia.
Data Center Interconnect Under Pressure
The DCI segment is perhaps where the scale-across challenge is most acute. As AI model sizes grow — with some frontier models requiring tens of thousands of GPUs to train — the data pipelines between nodes must deliver terabit-scale throughput with microsecond-level precision. Any bottleneck in the optical layer translates directly into degraded AI performance and spiraling operational costs.
Nokia has been positioning its optical portfolio — including its Photonic Service Engine (PSE) coherent technology — as purpose-built for these high-capacity, high-efficiency requirements. The company’s focus on pluggable coherent optics and open line systems is particularly relevant as network operators look to maximize spectral efficiency on existing fiber assets while rapidly deploying new capacity.
AI Demands Are Reshaping Vendor Roadmaps Industry-Wide
Nokia is far from alone in recognizing this inflection point. Competitors including Ciena, Infinera (now part of Nokia following a landmark acquisition), and Fujitsu have all recalibrated their optical roadmaps around AI-driven demand. The race is now centered on delivering higher baud rates, smarter network automation, and AI-native network management platforms that can dynamically route and optimize traffic in real time.
Ironically, AI itself is being deployed as a solution to the management complexity it creates. Vendors are embedding machine learning algorithms directly into optical network control planes to predict traffic surges, automate fault remediation, and optimize wavelength routing without human intervention. Nokia’s own network management platforms are increasingly leveraging AI-driven analytics to handle the sheer operational complexity of large-scale optical deployments.
The Role of 400G, 800G, and Beyond
From a purely technical standpoint, the industry’s response to AI bandwidth pressure is clearly visible on the standards and silicon fronts. 400G coherent optical interfaces have moved from cutting-edge to mainstream in just a few years, and 800G deployments are now beginning in earnest at major hyperscalers. Nokia, Ciena, and others are already discussing 1.6 Terabit per second (Tbps) roadmaps that could begin commercial deployment within the next two to three years.
Advanced modulation schemes, higher baud rates enabled by improved digital signal processors (DSPs), and next-generation forward error correction (FEC) techniques are all critical components of this capacity push. The challenge is delivering these speeds at an economics point that makes sense not just for hyperscalers, but also for telecom carriers, cloud-neutral interconnection facilities, and large enterprises building private AI infrastructure.
Implications for Telecom Carriers and Service Providers
For traditional telecommunications carriers, the AI-driven optical opportunity is a double-edged sword. On one hand, exploding bandwidth demand between data centers and cloud regions creates a clear market for wholesale optical transport and wavelength services. On the other hand, hyperscalers are increasingly inclined to own and operate their own optical infrastructure, potentially bypassing carrier networks altogether on key routes.
The carriers that position themselves successfully will likely be those that can offer differentiated capabilities — ultra-low latency on specific routes, geographic reach that hyperscalers cannot replicate economically, or deeply integrated service assurance backed by AI-powered network operations centers.
Industry Outlook: Optical Networks as AI’s Critical Backbone
The “scale-across” framing Nokia is promoting is more than marketing language — it captures a genuine architectural reality that network planners and infrastructure investors are grappling with right now. As AI workloads continue to distribute across hybrid and multi-cloud environments, the optical layer becomes, arguably, the single most critical piece of infrastructure underpinning the entire AI economy.
The companies — vendors, carriers, and hyperscalers alike — that invest intelligently in optical capacity, automation, and AI-native operations today will be best positioned to serve the insatiable connectivity demands of tomorrow’s AI-driven world. The scale-across era has arrived, and optical networks will never be the same.
