Manufacturing & Quality
4.0 factories, automation, validation
Partner Portal
Access tools, pricing, ESG data
Hybrid cloud and AI connectivity in 2026 breaks into three tiers: dense intra-cluster links carried on copper and short-reach optics, campus and metro links between owned facilities and colocation sites, and provider handoff links into public cloud. Each tier has a different dominant constraint. Intra-cluster is power-limited, campus is fibre-limited, and handoff is interoperability-limited.
Hybrid estates have stopped being a single network with a few external links. AI workloads have pulled the intra-cluster tier into a category of its own, with density and power characteristics that have little in common with the campus and interconnect tiers around it.
Treating all three the same way produces predictable problems. Optics chosen for their reach get used where copper would do, wasting power in the densest racks. Cable plant designed for enterprise link counts gets overwhelmed by AI fabric density. And handoff links get specified without checking what the far end will accept.
AI racks run close to the facility limit for power and cooling before the network is considered. Every watt in a transceiver is a watt not available to compute, and it also adds to the cooling load.
The efficient pattern is to use copper wherever reach permits and optics only where it does not. Passive DAC covers the shortest intra-rack links with negligible power. ACC reaches around 5 m and AEC around 7 to 10 m at 400G and 800G. Beyond that, SR8 multimode covers up to around 100 m for in-row leaf to spine links.
Where optics are unavoidable, module architecture matters. Linear pluggable optics and silicon photonics designs reduce module power by simplifying the electrical path, which is why they have moved from novelty to serious option in dense fabrics.
Links between halls, buildings and nearby colocation facilities are limited by the fibre already installed and how it is terminated.
DR-class single-mode modules reach 500 m and cover hall-to-hall spans. FR-class modules, including 800G 2xFR4, reach 2 km and cover most campus distances. Beyond that, longer-reach and coherent options apply.
The practical question is usually not which module but whether the installed plant supports it. Single-mode is required for DR and FR. Multimode does not carry them at any distance. Connector type, polarity and loss budget all have to be checked before ordering, because discovering a mismatch during commissioning is expensive.
Links into a cloud provider or a carrier terminate on equipment you do not control. Both ends have to be built to the same optical standard, and each end has to accept the module installed in it.
Host coding is the recurring problem. Many switch platforms check module EEPROM content before enabling a port, and that check sits outside the MSA and IEEE standards. Meeting the standard does not guarantee the port comes up.
ATOP owns its firmware and performs vendor-specific coding across major switching platforms in house, which means modules can be supplied coded for the platform they will be installed in rather than requiring a workaround at commissioning.
Regulatory expectation has shifted from vendor attestation to product evidence. NIS2 raises security obligations for operators of important entities. The Cyber Resilience Act requires patchable, documented firmware. CSRD requires environmental reporting that reaches down the supply chain.
For network hardware, that means procurement now asks what is inside a module, where it came from, who can patch it and where it was made. ATOP maintains complete bills of materials with component-level documentation, transparent country of origin declarations, and firmware cryptographically linked to hardware.
This matters most in hybrid estates, because the same evidence has to satisfy an internal audit, a colocation provider and sometimes a customer contract.
Most estates are not moving from 400G to 800G in one step. They are running both, often in the same fabric, with 800G in the AI cluster and 400G everywhere else.
Breakout links carry a lot of that traffic. An 800G port split into 2x400G or 8x100G connects new switching to existing endpoints without replacing them. Where the fibre plant is single-mode and correctly terminated, that transition can often be made by changing modules and switches rather than pulling new cable.
ATOP's portfolio runs from 1G to 1.6T across optical and copper, which allows an estate to be supplied consistently through a transition rather than being split across sources by generation.