You can’t buy enough DCI gear right now. The AI build-out has turned the long-haul optical equipment that stitches data centers together into one of the hardest-to-get items in the whole infrastructure stack.
Lead times that used to run 1 to 1.5 years have stretched to 2 to 2.5. Ciena, the long-time leader, runs about $3.5 billion a year of capacity — and 2027 is already booked.
Google, Microsoft and Meta have all raised budgets and tendered early to lock supply. When demand runs that far ahead of supply, the question stops being whose roadmap is prettier — it becomes who can actually deliver.
↓edge 1 — idle capacity nobody else has
two data centers · one fiber · coherent everywhere
idle US lines ~$2.5B
1.2 Tb per port
delivery, not specs
10× — own lasers
When Nokia bought Infinera, it inherited Infinera’s North American lines — centered in San Jose — sitting largely unused. Roughly $2.5 billion a year of capacity dark, up to $4 billion with overtime.
That’s the reason Google handed Nokia the biggest slice of its DCI tender — on the order of 50–60%. Not because the product won a feature shootout, but because the boxes would actually arrive.
delivery, not specs.
⚡edge 2 — it owns its own supply chain
Nokia designs its own DSP — the chip that encodes data onto the light — now at 1.2 terabits per wavelength on a 5-nanometer process. Most rivals buy this chip from Broadcom or Marvell.
Nokia also runs its own indium-phosphide (InP) wafer fab — the semiconductor lasers are built from — plus packaging and test. The single scarcest part in DCI today is the DSP; a vendor that makes its own is not exposed to the shortage.
Building in North America runs 70–150% more expensive than Asia. In this window, Nokia trades margin for speed — and is expanding InP capacity tenfold.
take share. optimize later.
⌖edge 3 — no tech gap, share map flipped
There is no meaningful technology gap between Nokia and Ciena. The standards that matter — 400G and 800G per wavelength — are settled, and Nokia has already shown a full 1.6-terabit DCI solution.
Ciena was ~90% of DCI in 2025 — a near-monopoly. In 2026 it is capacity-capped and ceding the increment. Nokia jumped from minor player to >55% of Google’s tender, the year’s main growth story.
※where the capacity actually sits
San Jose (Infinera core) is the main DCI assembly line — the $2.5B idle, up to $4B with overtime. Plus an InP fab and DSP R&D nearby — the vertical-integration core, capacity planned to grow 10×.
A Mexico line for companion boards was once slated to close, kept alive by data-center demand. And a Chinese partner (德科立) co-developed and supplied early OEM orders.
↓the risks
The capacity dividend is finite — once the backlog is consumed, the scarcity premium fades. North American cost is high. And absorbing Infinera is a multi-year integration job.
The case isn’t that Nokia built a better box. It’s that, in the narrow window where DCI demand is exploding and Ciena is sold out, Nokia grabbed the one resource nobody else has — and converted it into the majority of the year’s biggest orders.