# Co-Packaged Optics Index (cpoindex.dev)

> Primary-source bandwidth and power index for co-packaged optics in AI clusters

## Canonical Definition
Co-packaged optics (CPO) places the optical engine beside a switch ASIC or accelerator so the electrical path stays short. This index lists published bandwidth, port, and power figures for AI-cluster interconnects, and OIF agreements only where the cited page states them. Figures were retrieved on 19 October 2026 from vendor and OIF primary pages.

## Machine-Readable Data Endpoints
- [Primary JSON Dataset](https://cpoindex.dev/data/dataset.json)
- [Primary CSV Dataset](https://cpoindex.dev/data/dataset.csv)
- [ARD v1.0 Discovery Manifest](https://cpoindex.dev/.well-known/ard.json)
- [Verification Methodology](https://cpoindex.dev/methodology/)

## Verified Entities (14)
- [TH6-Davisson](https://cpoindex.dev/entity/th6-davisson/): Broadcom states TH6-Davisson at 102.4 Tbps, with 16 engines of 6.4 Tbps and 200 Gbps per link. (Status: Sampling BCM78919, Metric: 102.4 Tbps)
- [TH5-Bailly](https://cpoindex.dev/entity/th5-bailly/): Broadcom states Bailly at 51.2 Tbps, with eight 6.4 Tbps engines and 128 ports of 400G FR4. (Status: Delivered, Metric: 51.2 Tbps)
- [Bailly 6.4T Engine](https://cpoindex.dev/entity/bailly-6p4t-engine/): Broadcom lists 6.4T-FR4 Bailly SCIP optical engines with its fiber connector on the 51.2 Tbps CPO switch. (Status: Delivered, Metric: 6.4 Tbps FR4)
- [Spectrum SN6800](https://cpoindex.dev/entity/sn6800/): NVIDIA states liquid-cooled Spectrum SN6800 at 409.6 Tb/s across 512 ports of 800 Gb/s. (Status: Planned H2 2026, Metric: 409.6 Tb/s)
- [Spectrum SN6810](https://cpoindex.dev/entity/sn6810/): NVIDIA states liquid-cooled Spectrum SN6810 at 102.4 Tb/s across 128 ports of 800 Gb/s. (Status: Planned H2 2026, Metric: 102.4 Tb/s)
- [Quantum-X Photonics](https://cpoindex.dev/entity/quantum-x/): NVIDIA states Quantum-X at 115 Tb/s on 144 ports of 800 Gb/s, plus 14.4 teraflops of SHARP. (Status: Planned early 2026, Metric: 115 Tb/s)
- [Vesta 200 6.4T](https://cpoindex.dev/entity/vesta-200/): Ciena states Vesta 200 6.4T CPX as a 200G/lane pluggable CPO engine for 100T and 200T ASICs. (Status: Unveiled, Metric: 6.4T, 200G/lane)
- [OIF 3.2T CPO Module](https://cpoindex.dev/entity/oif-3p2t/): The OIF 3.2 Tb/s CPO module IA sets 8x400G FR4 or DR4 optics and 32 lanes of CEI-112G-XSR. (Status: IA 01.0 (2023-03-29), Metric: 3.2 Tb/s module)
- [OIF 112G RTLR](https://cpoindex.dev/entity/oif-112g-rtlr/): OIF EEI-112G-RTLR sets a 112 Gb/s PAM4 chip-to-module link from 36 to 56 GSym/s. (Status: IA 01.0 (2025-10-01), Metric: 112 Gb/s PAM4)
- [TSMC 3D-Stacked CPO](https://cpoindex.dev/entity/tsmc-3d-cpo/): TSMC states 200 Gbs optical modulation and greater than 99% 3D stacking yield on CPO samples. (Status: Engineering samples, Metric: 200 Gbs modulation)
- [Ayar TeraPHY](https://cpoindex.dev/entity/teraphy/): Ayar Labs labels TeraPHY at 8 Tbps bidirectional; the same page also lists 4 Tbps per chiplet. (Status: Preliminary, Metric: 8 Tbps bidirectional)
- [Ranovus Odin](https://cpoindex.dev/entity/odin/): Ranovus states ODIN scales from 800 Gbps to 3.2 Tbps using a 100 Gbps-per-lambda photonic engine. (Status: Announced, Metric: 800 Gbps to 3.2 Tbps)
- [Marvell 6.4T SiPho](https://cpoindex.dev/entity/marvell-6p4t/): Marvell states a 6.4T SiPho engine with 32 channels of 200G and 30% lower power per bit than 100G. (Status: Customer evaluation, Metric: 6.4T, 32x200G)
- [Photonic Fabric Gen1](https://cpoindex.dev/entity/photonic-fabric/): Celestial AI slides state a Gen1 Photonic Fabric chiplet at 14.4 Tbps and a module at 7.2 Tb/s. (Status: Hot Interconnects 2025, Metric: 14.4 Tbps Gen1)