{"schema_version":"2026-09-05.topic-graph-v1","canonical_url":"https://www.manufacturing.ai/topics/copper-interconnect","topic":{"slug":"copper-interconnect","name":"Copper, and Where It Runs Out","description":"A sourced reference collection on the least glamorous component in an AI rack — why a rack-scale accelerator contains miles of copper cable, why that choice saved twenty kilowatts, why copper is now being replaced by light, and why the power busbar is a materials problem measured in kilograms.","coverage_notes":"Eighth packet, and the first written around a component that became a market narrative before anyone wrote a sourced account of it. Ten sources were consulted: three were retrieved and read, one was retrieved and found to contain no citable figures, and six were surfaced and set aside with a stated reason. The concentration is the main weakness and is stated plainly: nearly every figure here originates with NVIDIA, either from its own engineering posts or from reporting of its executives' statements. That is mitigated slightly by one of the sources being a contribution of actual mechanical designs to the Open Compute Project, which is more checkable than a datasheet, and not at all by the rest. The company's efficiency, maintenance and total-cost-of-ownership projections are recorded as claims and asserted nowhere; only its statements about its own hardware are used. Named gaps: the Open Compute Project's 800 VDC whitepaper and solid-state transformer specification, which would replace one vendor's account of a multi-vendor standard with the standard, and for which no stable URL was established; an independent technical analysis of the copper-versus-optics decision that sits behind a paid newsletter; and wider context on cable backplanes, which are not an NVIDIA invention and whose engineering trend predates AI racks. Two vendor blogs from optical-component suppliers were set aside for conflict of interest rather than quality, and an anonymous social-media post describing a cableless Rubin midplane is listed as a lead with its standing made unambiguous. Not yet editor-reviewed; every assertion reads as reported.","primer":"An AI rack contains an unreasonable amount of copper, and the reason is a good one. NVIDIA's GB200 NVL72 carries four NVLink cartridges at the back of the rack holding more than 5,000 active copper cables, meshing 72 GPUs at 1.8 TB/s each for 260 TB/s of AllReduce bandwidth across the domain. The designs were contributed to the Open Compute Project, so this is a published mechanical design rather than a specification sheet.\n\nCopper was chosen over light for a reason that can be stated in watts. Reaching 1.8 TB/s optically would have needed eighteen 800 Gbps pluggable transceivers per GPU — nine on the accelerator, nine on the switch — at roughly 10 to 15 watts each. NVIDIA's chief executive said in 2024 that pluggable optics for the NVL72 would have required an additional 20,000 watts. In a rack that already needs 120 kW of cooling, twenty kilowatts spent turning electricity into light and back is twenty kilowatts not spent on arithmetic. The cheapest interconnect is the one you do not have to power.\n\nAnd then copper runs out. NVIDIA's own networking lead puts it as copper being the best connectivity if you can use it — the qualification doing the work. At 1.8 TB/s a copper cable degrades within a few feet, which is why the switches sit in the middle of the rack and why the design crams as many GPUs into one rack as it physically can. Both are workarounds for reach, and neither survives contact with a domain larger than a rack. So scale-up goes optical: Vera Rubin NVL576, then Rosa Feynman NVL1152, with co-packaged optics integrated at Feynman. NVIDIA has committed $2 billion each to Coherent and Lumentum for lasers and $2 billion to Marvell for optical I/O.\n\nTwo opposite decisions four years apart, both defensible. Copper won while the machine fitted in one rack, and stops winning the moment it does not. Anyone reading this as a reversal has the wrong frame; it is the same trade evaluated at a different size.\n\nThe second copper problem is supply rather than signalling, and it is measured in kilograms. At the 54 VDC used inside current racks, NVIDIA states a 1 MW rack needs up to 200 kg of copper busbar, and the racks alone in a 1 GW facility could need up to 200,000 kg. Lower voltage at fixed power means more current, and more current means more metal. The proposed answer is to raise the voltage — 13.8 kV AC converted directly to 800 VDC at the perimeter — with a claimed 85% more power through the same conductor and 45% less copper than a 415 VAC system, in production alongside Kyber racks in 2027.\n\nWhat is missing is named. Nearly everything here comes from one company: the engineering reasoning, the power figures and the roadmap all originate with NVIDIA or with reporting of NVIDIA statements. The independent technical analysis of the copper-versus-optics decision sits behind a paid newsletter and was not read, and the Open Compute Project whitepaper and solid-state transformer specification behind the 800 VDC material were not retrieved — so a multi-vendor specification is described here through one vendor's account of it. A social-media post describing a cableless midplane redesign for Rubin is listed as a lead and asserted nowhere.","primer_note":"Manufacturing.ai's own editorial synthesis, not evidence. Every factual statement in it is separately asserted and sourced in this response's concepts and events.","reviewed_through_date":null,"source_count":3,"inspected_source_count":3,"consulted_reference_count":13,"blocked_reference_count":4,"set_aside_reference_count":9},"figures":[{"kind":"time-series","title":"What the interconnect choice costs in power","caption":"Additional rack power that pluggable optics would have added to an NVL72, against copper. This is the whole argument for copper, in one comparison.","sourceNote":"NVIDIA's chief executive, quoted in reporting of a 2024 GTC keynote: pluggable optics for the NVL72 would have required an additional 20,000 watts. The copper figure is zero additional interconnect power by construction, not a measurement. Shown beside the rack's stated 120 kW cooling capacity for scale.","unit":"Additional rack power for the interconnect, kilowatts","points":[{"label":"Copper NVLink spine","value":0,"display":"0 kW"},{"label":"Pluggable optics equivalent","value":20,"display":"20 kW"}]},{"kind":"time-series","title":"The scale-up domain outgrows the rack","caption":"GPUs in a single scale-up domain by generation. Copper served the first; the next two are stated to go optical.","sourceNote":"NVIDIA's Open Compute Project contribution for the 72-GPU domain, and reporting of April 2026 for the NVL576 and NVL1152 generations. The latter two are announced future systems, not shipped ones.","unit":"GPUs in one scale-up domain","points":[{"label":"GB200 NVL72","value":72,"display":"72"},{"label":"Vera Rubin NVL576","value":576,"display":"576"},{"label":"Rosa Feynman NVL1152","value":1152,"display":"1,152"}]}],"blocked_references":[{"title":"800 VDC Architecture for AI Data Centers","publisher":"NVIDIA","url":"https://www.nvidia.com/en-us/data-center/technologies/800-vdc-architecture/","source_type":"company_website","retrieval_status":"retrieved_no_content","content_inspected":false,"published_at":null},{"title":"All AI Data Center Interconnects Will Be Optical Within 5 Years","publisher":"Semiconductor Engineering","url":"https://semiengineering.com/all-ai-data-center-interconnects-will-be-optical-within-5-years/","source_type":"journalism","retrieval_status":"blocked_403","content_inspected":false,"published_at":null},{"title":"Broadcom Announces Tomahawk 6 – Davisson (investor relations original)","publisher":"Broadcom Inc.","url":"https://investors.broadcom.com/news-releases/news-release-details/broadcom-announces-tomahawkr-6-davisson-industrys-first-1024","source_type":"press_release","retrieval_status":"timeout","content_inspected":false,"published_at":null},{"title":"Where co-packaged optics (CPO) technology stands in 2026","publisher":"EDN","url":"https://www.edn.com/where-co-packaged-optics-cpo-technology-stands-in-2026/","source_type":"journalism","retrieval_status":"timeout","content_inspected":false,"published_at":null}],"set_aside_references":[{"title":"Cable Backplanes Come of Age","publisher":"Connector Supplier","url":"https://connectorsupplier.com/cable-backplanes-come-of-age/","source_type":"journalism","retrieval_status":"search_result_only","content_inspected":false},{"title":"NVIDIA Blackwell Architecture Explained: B200, GB200 & PCB Design Impact","publisher":"NextPCB","url":"https://www.nextpcb.com/blog/nvidia-blackwell-architecture-b200-gb200-pcb-design","source_type":"other","retrieval_status":"search_result_only","content_inspected":false},{"title":"NVIDIA GB200 Interconnect Architecture Analysis","publisher":"NADDOD","url":"https://www.naddod.com/blog/nvidia-gb200-interconnect-architecture-analysis-nvlink-infiniband-and-future-trends","source_type":"other","retrieval_status":"search_result_only","content_inspected":false},{"title":"NVIDIA GPU History: GeForce 256 to Vera Rubin","publisher":"IOH Tech","url":"https://iohtechco.com/resources/nvidia-gpu-history-evolution","source_type":"other","retrieval_status":"search_result_only","content_inspected":false},{"title":"NVIDIA NVL72: Defining the New Benchmark for Rack-Scale AI Computing","publisher":"FiberMall","url":"https://www.fibermall.com/blog/nvidia-nvl72-for-rack-sacle-ai.htm","source_type":"other","retrieval_status":"search_result_only","content_inspected":false},{"title":"Nvidia's Optical Boogeyman — NVL72, InfiniBand Scale Out, 800G & 1.6T Ramp","publisher":"SemiAnalysis","url":"https://newsletter.semianalysis.com/p/nvidias-optical-boogeyman-nvl72-infiniband","source_type":"journalism","retrieval_status":"search_result_only","content_inspected":false},{"title":"OCP 800 VDC whitepaper and LVDC Solid-State Transformer Specification","publisher":"Open Compute Project","url":"https://www.opencompute.org/","source_type":"other","retrieval_status":"search_result_only","content_inspected":false},{"title":"Post describing a cableless Rubin midplane redesign","publisher":"X (formerly Twitter)","url":"https://x.com/TheValueist/status/1983276631348195593","source_type":"other","retrieval_status":"search_result_only","content_inspected":false},{"title":"What is CUDA? Parallel programming for GPUs","publisher":"InfoWorld","url":"https://www.infoworld.com/article/2256401/what-is-cuda-parallel-programming-for-gpus.html","source_type":"journalism","retrieval_status":"search_result_only","content_inspected":false}],"concepts":[{"slug":"ai-rack-power-density","name":"AI Rack Power Density","category":"component","canonical_url":"https://www.manufacturing.ai/topics/copper-interconnect#ai-rack-power-density","assertions":[]},{"slug":"co-packaged-optics","name":"Co-Packaged Optics (CPO)","category":"packaging_technology","canonical_url":"https://www.manufacturing.ai/topics/copper-interconnect#co-packaged-optics","assertions":[]},{"slug":"copper-at-rack-scale","name":"Copper at Rack Scale","category":"interconnect","canonical_url":"https://www.manufacturing.ai/topics/copper-interconnect#copper-at-rack-scale","assertions":[{"id":"329d558d-792e-4387-a48b-2953e10bd1a7","predicate":"concept.description","statement":"Reaching 1.8 TB/s per GPU optically would have required eighteen 800 Gbps pluggables per GPU at 10-15 W each; NVIDIA's chief executive stated in 2024 that pluggable optics for the NVL72 would have needed an additional 20,000 watts — the case for copper, stated in watts.","assessment":"reported","editor_reviewed":false,"reference_only_source_count":0,"evidence":[{"id":"ede06664-6121-42e1-b301-7ce398d953e8","stance":"supports","is_primary_for_assertion":false,"origin_independence":"unknown","note":null,"quote":null,"source":{"title":"Nvidia embraces optical scale-up as copper reaches limits","publisher":"The Register","url":"https://www.theregister.com/2026/04/05/nvidia_optical_scale_up/","source_type":"journalism","retrieval_status":"fetched","content_inspected":true,"published_at":"2026-04-05","documents":[]}}]},{"id":"7bf8a95d-961b-4292-a260-68a0d1b4fde5","predicate":"concept.description","statement":"The GB200 NVL72 uses four rear-mounted NVLink cartridges holding over 5,000 active copper cables to mesh 72 GPUs at 1.8 TB/s each and 260 TB/s AllReduce, on a 1,400-amp busbar with 120 kW of liquid cooling — designs NVIDIA contributed to the Open Compute Project.","assessment":"reported","editor_reviewed":false,"reference_only_source_count":0,"evidence":[{"id":"1a11db0b-d18b-4b17-b0e9-771f01037d33","stance":"supports","is_primary_for_assertion":true,"origin_independence":"unknown","note":null,"quote":null,"source":{"title":"NVIDIA Contributes NVIDIA GB200 NVL72 Designs to Open Compute Project","publisher":"NVIDIA","url":"https://developer.nvidia.com/blog/nvidia-contributes-nvidia-gb200-nvl72-designs-to-open-compute-project/","source_type":"company_website","retrieval_status":"fetched","content_inspected":true,"published_at":null,"documents":[]}}]}]},{"slug":"gpu","name":"Graphics Processing Unit (GPU)","category":"component","canonical_url":"https://www.manufacturing.ai/topics/copper-interconnect#gpu","assertions":[]},{"slug":"nvlink-scale-up-fabric","name":"NVLink and the Scale-Up Fabric","category":"interconnect","canonical_url":"https://www.manufacturing.ai/topics/copper-interconnect#nvlink-scale-up-fabric","assertions":[]},{"slug":"pluggable-optical-transceiver","name":"Pluggable Optical Transceiver","category":"component","canonical_url":"https://www.manufacturing.ai/topics/copper-interconnect#pluggable-optical-transceiver","assertions":[]},{"slug":"rack-power-distribution","name":"Rack Power Distribution","category":"component","canonical_url":"https://www.manufacturing.ai/topics/copper-interconnect#rack-power-distribution","assertions":[{"id":"85a88056-ebea-42e6-94b0-022e28555ac4","predicate":"concept.description","statement":"NVIDIA states a 1 MW rack needs up to 200 kg of copper busbar at 54 VDC, and up to 200,000 kg for the racks in a 1 GW facility; its proposed 800 VDC architecture converts 13.8 kV AC directly at the perimeter and claims 85% more power through the same conductor with 45% less copper than 415 VAC, in production alongside Kyber systems in 2027.","assessment":"reported","editor_reviewed":false,"reference_only_source_count":0,"evidence":[{"id":"bf23005f-9d95-4db4-8dfb-3bd83a018d95","stance":"supports","is_primary_for_assertion":true,"origin_independence":"unknown","note":null,"quote":null,"source":{"title":"NVIDIA 800 VDC Architecture Will Power the Next Generation of AI Factories","publisher":"NVIDIA","url":"https://developer.nvidia.com/blog/nvidia-800-v-hvdc-architecture-will-power-the-next-generation-of-ai-factories/","source_type":"company_website","retrieval_status":"fetched","content_inspected":true,"published_at":null,"documents":[]}}]}]},{"slug":"scale-up-versus-scale-out","name":"Scale-Up Versus Scale-Out","category":"interconnect","canonical_url":"https://www.manufacturing.ai/topics/copper-interconnect#scale-up-versus-scale-out","assertions":[]},{"slug":"copper-reach-limit","name":"The Copper Reach Limit","category":"interconnect","canonical_url":"https://www.manufacturing.ai/topics/copper-interconnect#copper-reach-limit","assertions":[{"id":"8bf99149-f123-4f32-8e94-00940faf5679","predicate":"concept.description","statement":"Copper is bounded at 1.8 TB/s over a few feet, forcing centre-of-rack switches and single-rack domains; NVIDIA is moving scale-up to optics from Vera Rubin NVL576 through Rosa Feynman NVL1152 with co-packaged optics at Feynman, and has committed $2bn each to Coherent, Lumentum and Marvell.","assessment":"reported","editor_reviewed":false,"reference_only_source_count":0,"evidence":[{"id":"b45525f5-7bd4-486d-abfc-828dd25dc2b6","stance":"supports","is_primary_for_assertion":false,"origin_independence":"unknown","note":null,"quote":null,"source":{"title":"Nvidia embraces optical scale-up as copper reaches limits","publisher":"The Register","url":"https://www.theregister.com/2026/04/05/nvidia_optical_scale_up/","source_type":"journalism","retrieval_status":"fetched","content_inspected":true,"published_at":"2026-04-05","documents":[]}}]}]}],"events":[{"slug":"eight-hundred-vdc-production-2027","title":"Rack Power Distribution Moves to 800 Volts","category":"technology_generation_milestone","event_date":"2027-01-01","date_precision":"year","country":"United States","region":null,"canonical_url":"https://www.manufacturing.ai/topics/copper-interconnect#eight-hundred-vdc-production-2027","assertions":[{"id":"2a946e1c-43f8-4213-9f46-47148a54550f","predicate":"event.description","statement":"NVIDIA states full-scale production of its 800 VDC architecture will coincide with Kyber rack-scale systems in 2027, converting 13.8 kV AC directly to 800 VDC at the perimeter, with claimed 85% more power through the same conductor and 45% less copper than 415 VAC.","assessment":"reported","editor_reviewed":false,"reference_only_source_count":0,"evidence":[{"id":"086a89df-9778-4cc3-8538-13840c9d5c46","stance":"supports","is_primary_for_assertion":true,"origin_independence":"unknown","note":null,"quote":null,"source":{"title":"NVIDIA 800 VDC Architecture Will Power the Next Generation of AI Factories","publisher":"NVIDIA","url":"https://developer.nvidia.com/blog/nvidia-800-v-hvdc-architecture-will-power-the-next-generation-of-ai-factories/","source_type":"company_website","retrieval_status":"fetched","content_inspected":true,"published_at":null,"documents":[]}}]}]},{"slug":"optical-scale-up-announced-2026","title":"Scale-Up Leaves Copper for Light","category":"technology_generation_milestone","event_date":"2026-04-05","date_precision":"day","country":"United States","region":null,"canonical_url":"https://www.manufacturing.ai/topics/copper-interconnect#optical-scale-up-announced-2026","assertions":[{"id":"85e9223f-2862-460f-85dc-8eb20e459854","predicate":"event.description","statement":"Reporting of 5 April 2026 records NVIDIA moving scale-up from copper to optics at Vera Rubin NVL576 and Rosa Feynman NVL1152, with co-packaged optics at Feynman, citing copper's few-feet reach at 1.8 TB/s, alongside $2bn commitments each to Coherent, Lumentum and Marvell.","assessment":"reported","editor_reviewed":false,"reference_only_source_count":0,"evidence":[{"id":"d1c2084f-48be-449f-8cc0-0f1a49d70c90","stance":"supports","is_primary_for_assertion":false,"origin_independence":"unknown","note":null,"quote":null,"source":{"title":"Nvidia embraces optical scale-up as copper reaches limits","publisher":"The Register","url":"https://www.theregister.com/2026/04/05/nvidia_optical_scale_up/","source_type":"journalism","retrieval_status":"fetched","content_inspected":true,"published_at":"2026-04-05","documents":[]}}]}]},{"slug":"nvl72-designs-contributed-to-ocp","title":"The Copper Rack Design Is Published","category":"standards_milestone","event_date":null,"date_precision":"unknown","country":"United States","region":null,"canonical_url":"https://www.manufacturing.ai/topics/copper-interconnect#nvl72-designs-contributed-to-ocp","assertions":[{"id":"b685a830-847c-42d2-ba42-71c362edbd14","predicate":"event.description","statement":"NVIDIA contributed the GB200 NVL72 rack, liquid-cooled tray designs, busbar, cartridge mounting and 1RU form factors to the Open Compute Project — a published mechanical design for a rack carrying over 5,000 copper cables, a 1,400 amp busbar and 120 kW of cooling.","assessment":"reported","editor_reviewed":false,"reference_only_source_count":0,"evidence":[{"id":"1f97b8ee-1ffa-49c2-992f-ee0c113eaf53","stance":"supports","is_primary_for_assertion":true,"origin_independence":"unknown","note":null,"quote":null,"source":{"title":"NVIDIA Contributes NVIDIA GB200 NVL72 Designs to Open Compute Project","publisher":"NVIDIA","url":"https://developer.nvidia.com/blog/nvidia-contributes-nvidia-gb200-nvl72-designs-to-open-compute-project/","source_type":"company_website","retrieval_status":"fetched","content_inspected":true,"published_at":null,"documents":[]}}]}]}],"policy":{"publication_model":"transparency","evidence_entries_may_share_a_source":true,"linked_documents_do_not_add_verification":true,"correction_intake_available":false,"assessment_meanings":{"primary-source confirmed":"Editor-reviewed against a first-party source.","cross-source corroborated":"Editor-confirmed across two independent origins.","reported":"Supported by the attached sources; not yet editor-reviewed.","disputed":"Sources conflict; competing accounts are both attached.","superseded":"Replaced by a newer assertion.","unverified":"No supporting evidence attached."},"evidence_eligibility":"`evidence[]` contains only sources whose content was actually retrieved and read (`content_inspected: true`). Sources that could not be retrieved are counted per-assertion in `reference_only_source_count`, listed for the whole collection in `consulted_references`, and never presented as evidence. `source_count` counts only sources an assertion actually rests on — it deliberately excludes `consulted_references`, so it is never inflated by material nobody could read.","topic_scope":"Assertions and source counts are limited to the topic's current packet lineage. Reusing a concept in another topic does not silently import that other topic's assertions.","rights":"Statements are Manufacturing.ai's own prose. Source excerpts are not redistributed; `quote` is present only where explicitly cleared.","attribution":"Cite as Manufacturing.ai, https://www.manufacturing.ai/topics/copper-interconnect","correction_url":"https://www.manufacturing.ai/methodology#corrections"}}