Nokia operates one fab and has another in development (the newly expanded six-inch indium phosphide facility in San Jose), as well as an advanced packaging center in Pennsylvania. In an era where “sovereign AI” and supply chain resilience are paramount, being the only Western vendor with this level of vertical integration should act as a strategic moat.
Moving beyond the one-size-fits-all approach
For decades, the industry chased a single goal: maximum capacity per fiber. But the AI era has bifurcated the market. On one hand, we have the “scale-across” architecture of hyperscalers, where they are deploying upwards of 13,000 fibers between data centers. With that much fiber, it’s important to simplify what is typically a complex environment to deliver the lowest cost and power per bit. On the other hand, service providers dealing with fiber-constrained long-haul routes still need to squeeze every last drop of spectral efficiency out of a single strand.
Nokia’s response is a building-block approach that breaks the monolithic tradition of optical engines. Instead of one or two engines, they’ve introduced four distinct DSP engines—Ontario, Huron, Superior, and Pacific—which can be mixed and matched into 13 different applications.
Three innovations defining the new network
While the product list from OFC was extensive, three specific announcements stand out to me as thought leadership in action:
- The double-sided pluggable: As switches move toward integrated optics (CPO/XPO), the industry struggles with how to convert that “gray” light to coherent light for transport. Nokia’s solution integrates gray and coherent optics into a single device. As Rob Shore, head of optical network marketing, noted, this can result in a 70% TCO saving by eliminating the need for separate thin transponders and extra cabling.
- The full-band transponder: For hyperscalers lighting up ten fibers at once, the manual labor of connecting 1,000+ components is a recipe for human error. Nokia’s new appliance integrates the line-side, coherent, and client optics into one unit. It turns a chaotic rack of spaghetti cabling into a single fiber out.
- The superior engine: By taking the restrictions off the power envelope of a standard pluggable, Nokia has created a 2.4 Terabit engine that delivers a 60% reduction in TCO compared to today’s embedded engines.
The Jevons paradox of bandwidth
There is a lingering fear in the industry that radical cost reductions will crater the market. Nokia is betting on the Jevons paradox: the idea that as a resource becomes more efficient and cheaper, its use actually increases overall consumption. In some ways, the Internet has been proof of this, as bandwidth and costs for consumers have fallen, usage has grown at unprecedented rates. There have been many parallels made between the Internet and AI, and this is another that’s likely to hold true.
By driving down the power and cost of 1.6T and 2.4T connections, Nokia isn’t just serving the existing market; it’s enabling agentic AI workloads that were previously cost-prohibitive to become a reality as the network costs won’t drive all the economic value from the ecosystem. This lets the AI builders change their question from “can we build it?” to “how fast can we scale it?”
