Category 8 & Beyond: 40 Gbps over Copper

Copper’s last leap or a quiet revolution?

When the IEEE released 802.3bq in 2016, many analysts called it a swan-song: “Surely 40 Gbps belongs to fibre.” Yet nine years later Category 8 cabling is humming inside colo suites from Docklands to Dublin, cutting latency between top-of-rack switches and servers while letting operators keep the beloved RJ-45 jack. In the same period multi-gigabit Wi-Fi, AI edge nodes and PoE++ lighting have kept copper firmly in the spotlight.

At ACCL we treat Cat 8 as a specialist tool rather than a default choice, but when the use-case aligns (low-latency trading meshes, pilot AI clusters, rapid retro-fits in power-dense data halls), it delivers outstanding value. This long-form guide explores the technology, the standards, the practical design constraints and the crystal-ball question: Is there a viable copper roadmap beyond 40 Gbps?

Standards in brief: ISO versus TIA and the two classes of Cat 8

International standards split Cat 8 into Class I and Class II. Class I (ISO/IEC 11801-1 channel Class I, TIA Cat 8) uses familiar 8P8C RJ-45 connectors, ensuring backward compatibility to Cat 6A. Class II, aligned with ISO’s Cat 8.2, employs the GG45 or TERA interface and supports a slightly higher alien-crosstalk head-room. In the UK marketplace Class I dominates: it fits existing patch panels and doesn’t force tool changes.

Both classes certify to 2 000 MHz, four times the 500 MHz ceiling of Cat 6A. 40 GBASE-T and 25 GBASE-T are specified for 30 metre channel length, which is short in office terms but perfect inside a data hall where rows rarely exceed 20 m.

ISO/IEC 14763-4 dictates Level VI field-tester accuracy for Cat 8; if your commissioning partner tries to use a Level V analyser, you will not obtain the 25-year warranty from cable manufacturers.

Performance profile: latency, power and PoE coexistence

Latency is where Cat 8 trumps fibre. A 40 GBASE-SR4 link converts serial data to four-lane PAM4, runs the photons and then re-serialises, which takes roughly 2 to 3 µs end-to-end. A 40 GBASE-T PHY can deliver < 2 µs, shaving valuable microseconds in algorithmic trading or real-time analytics.

Power consumption has dropped since early silicon. The first 40 GBASE-T NICs drew 8 W per port; 2025 chipsets sit around 4.5 W, barely above 10 GBASE-T. They do run warmer than short-reach fibre, so rack design must preserve front-to-back airflow.

PoE remains an edge technology; the 30 m limit suits in-row equipment that already uses AC. However, IEEE has ratified PoE++ over Cat 8 for up to 100 W, handy for direct-attach GPU trays and smart-PDUs. Our PoE Budget Calculator shows voltage drop is trivial on 22 AWG at 25 m.

Designing a Cat 8 link: five critical dimensions

Distance is king. Measure rack-to-rack runs with a laser wheel, not CAD straight lines: the service loop, vertical manager and horizontal finger together consume 3 to 4 m. If your tape reads 27 m, the installed channel will brush the 30 m ceiling once patch cords are included.

Containment must segregate power. Shield stops ingress but not differential ground potential; keep 200 mm air gap or metal divider from 415 V busbars.

Bonding the braid to the Telecommunications Grounding Busbar at both ends lowers common-mode noise and fulfils BS EN 50310. A 6 mm² earth strap per rack bay suffices; impedance should read ≤ 0.1 Ω.

Testing uses a Level VI analyser with Cat 8 adapters. ACCL scripts include TCL and ELTCTL (critical for the 2 GHz spectrum), and we sample alien-crosstalk on a full bundle, not cherry-picked pairs.

Patch leads must match shielding type. A stray Cat 6A UTP cord will pass continuity but flunk NEXT spectacularly and may radiate above EN 55032 class B limits. Colour-code or physically key leads to eliminate swaps.

Cost comparison: cabling, electronics, lifecycle

A Cat 8 S/FTP channel lands about 40 % above Cat 6A per installed metre but 20 % below OM4 parallel-fibre once you fold in eight-lane MPO cassettes and SR4 optics. Electronics flip the equation (40 G fibre transceivers cost less per port than the still-niche 40 GBASE-T NICs), yet the enclosure count often drops with copper because one ToR switch can power-share PoE gear and provide 10 G downlinks simultaneously. In a recent Canary Wharf upgrade ACCL found the copper solution fell 8 % cheaper overall, thanks to saved fan-tray licences and integrated management.

Lifecycle matters too. If you foresee a hop to 100 G within five years, fibre’s upgrade path is smoother; Cat 8 tops out at 40 G by standard. But for estates locked to 25 G/40 G lanes for the foreseeable horizon, copper’s OpEx advantage (no optic inventory, zero cleaning routine) wins hands down.

Installation pitfalls we still encounter

The most common failing is untwisting pairs beyond 6 mm when punch-down caps close. At 2 GHz that extra centimetre slashes TCL. Second is improper drain-wire trim; a whisker touching the IDC creates intermittent shorts. Third is forgotten re-certification after MAC works: moving a rack by just one tile can lengthen the channel past 30 m. ACCL’s data-centre contracts bake in post-move Level VI tests to keep warranties intact.

Decision matrix: should you specify Cat 8?

Opt for Category 8 when:

• Distance is under 30 m and you need 25 G or 40 G with the lowest possible latency. 
• Rack airflow allows more than 5 W per port and you wish to avoid optics maintenance. 
• PoE++ distribution for in-row edge servers or camera clusters simplifies LV works. 
• Change windows are tight and engineers already trained on RJ-45 toolsets must execute the cut-over overnight. 

Choose fibre when:

• Runs exceed 30 m or may need 100 G upgrades. 
• EMC environment is unpredictable and adding more bonding bars is unrealistic. 
• Future tenant churn would prefer de-facto standard LC or MPO connectivity. 

For a cross-technology view, revisit our article Copper vs Fibre: Choosing the Right Backbone.

Conclusion: copper’s headroom is narrower, but still valuable

Category 8 will never blanket open-plan offices, yet in its design envelope it brings a compelling mix of backward-compatible connectors, microsecond latency and zero-optic maintenance. Future ultra-short-reach copper, whether in shielded pairs or waveguide tubes, suggests the medium’s demise has again been exaggerated.

If your data-centre refresh, AI pilot or low-latency trading floor could benefit from 25 to 40 Gbps over copper, talk to ACCL. Our BICSI-certified designers will model distance, EMI, PoE and containment, then deliver a certified, warranty-backed installation that keeps every nanosecond, and every packet, in your favour.