c6308e93c0
New scrapers: - GBICS.com (BigCommerce, GBP prices, 10 categories, 78 products) - Juniper HCT (Next.js SSR parser, 475 transceivers with specs/EOL) - SFPcables.com (Magento store, 16 categories, 78 products) - Fluxlight (BigCommerce, 6 pages, 118 products) - Champion ONE (compatible vendor scraper) Scraper fixes: - 10Gtek: rewritten to parse HTML spec tables (152 products) - Flexoptix: fix price extraction from Magento Hyva HTML - Register all scrapers in CLI (--gbics, --juniper, --sfpcables, etc.) Hype Cycle Engine enhancements: - Data-driven enrichment from scraped vendor/price data - Revenue lifecycle prediction (peak year, decline, revenue index) - Regional adoption model (NA, China, APAC, Europe, RoW with lag coefficients) - New API endpoints: /enriched, /lifecycle, /regional/:tech DB growth: 89 → 1,168 transceivers, 0 → 416 prices, 6 vendors Qdrant: 1,162 products embedded with nomic-embed-text Research: Norton-Bass model, standards-to-market timelines, hype signals
739 lines
44 KiB
Markdown
739 lines
44 KiB
Markdown
# Optical Networking Equipment: Demo-to-Market Predictive Timeline Model
|
|
|
|
> Research compiled 2026-03-28 for the Transceiver Intelligence Platform (TIP)
|
|
> Data from OFC/ECOC proceedings, IEEE standards, MSA publications, vendor press releases, Cignal AI, LightCounting, Dell'Oro Group
|
|
|
|
---
|
|
|
|
## Table of Contents
|
|
|
|
1. [Historical OFC/ECOC Demo-to-Market Timelines](#1-historical-timelines)
|
|
2. [Switch/Router ASIC Generation Timelines](#2-asic-timelines)
|
|
3. [The Lag Formula](#3-the-lag-formula)
|
|
4. [Demand Cascade Model](#4-demand-cascade-model)
|
|
5. [Export Control Impact](#5-export-control-impact)
|
|
6. [Predictive Timeline Calculator](#6-predictive-calculator)
|
|
|
|
---
|
|
|
|
## 1. Historical OFC/ECOC Demo-to-Market Timelines {#1-historical-timelines}
|
|
|
|
### 1.1 10G SFP+
|
|
|
|
| Milestone | Date | Source |
|
|
|-----------|------|--------|
|
|
| IEEE 802.3ae study group formed | Nov 1999 | IEEE archives |
|
|
| IEEE 802.3ae ratified (10GbE standard) | Jun 2002 | [IEEE 802.3ae](https://resources.l-p.com/knowledge-center/what-is-ieee-802-3ae-10-gigabit-ethernet) |
|
|
| First 10G modules (XENPAK form factor) | 2002-2003 | First MSA for 10GE; largest form factor |
|
|
| XFP MSA published | 2003-2004 | Intermediate form factor between XENPAK and SFP+ |
|
|
| SFP+ MSA (SFF-8431) published | ~2006 | SFP+ became smallest, lowest-power 10G form factor |
|
|
| First SFP+ volume shipments | 2007-2008 | Industry adoption ramped with switch platforms |
|
|
| 10GBASE-T (802.3an) ratified | Jun 2006 | Extended 10G to copper |
|
|
| Mainstream SFP+ adoption | 2009-2010 | De facto standard for ToR/access |
|
|
|
|
**Total cycle: ~8 years** from IEEE standard (2002) to mainstream (2010). However, the SFP+ form factor itself took ~4 years from MSA (2006) to mainstream (2010).
|
|
|
|
### 1.2 40G QSFP+
|
|
|
|
| Milestone | Date | Source |
|
|
|-----------|------|--------|
|
|
| IEEE 802.3ba study group | Nov 2007 | IEEE archives |
|
|
| IEEE 802.3ba ratified (40G/100G Ethernet) | Jun 2010 | IEEE 802.3ba standard |
|
|
| First 40G QSFP+ commercial modules | 2010-2011 | QSFP+ MSA based on 4x10G lanes |
|
|
| Volume production begins | 2012-2013 | Kaiam, Finisar shipping high volume |
|
|
| Mainstream data center adoption | 2013-2015 | Standard for aggregation/ToR switches |
|
|
|
|
**Total cycle: ~5 years** from standard (2010) to mainstream (2015). Form-factor-to-volume: ~2 years.
|
|
|
|
### 1.3 100G QSFP28
|
|
|
|
| Milestone | Date | Source |
|
|
|-----------|------|--------|
|
|
| IEEE 802.3bm task force (100G over MMF/short-reach) | 2013 | Defined 4x25G lane architecture |
|
|
| QSFP28 MSA published | 2013-2014 | Based on QSFP+ with 4x25G lanes |
|
|
| First OFC demos (CWDM4, PSM4) | OFC 2015 | [Kaiam CWDM4 100G QSFP28 demo](https://www.businesswire.com/news/home/20150319005175/en/Kaiam-Introduces-CWDM4-100G-QSFP28-Transceiver-400G) |
|
|
| ColorChip adds PSM4 to QSFP28 portfolio | OFC 2016 | [ColorChip PSM4 announcement](https://www.globenewswire.com/news-release/2016/03/18/940853/0/en/) |
|
|
| InnoLight volume shipments (17 QSFP28 SKUs) | Mar 2017 | [InnoLight OFC 2017](https://www.innolight.com/en/news/newsinfo/13.html) |
|
|
| Oclaro 40km interop demo (QSFP28 ER4-Lite) | Mar 2017 | [Oclaro OFC 2017](https://www.prnewswire.com/news-releases/oclaro-showcases-industrys-first-live-40km-interoperability-demo-between-100g-extended-reach-qsfp28-and-cfp2-at-ofc-2017-300426690.html) |
|
|
| Market maturity (cost-effective vs 10G/40G) | 2017-2018 | More $/Gbit efficient than 10G SFP+ and 40G QSFP+ |
|
|
|
|
**Total cycle: ~4 years** from MSA (2014) to mainstream (2018). Demo-to-volume: ~2 years (OFC 2015 to Mar 2017).
|
|
|
|
### 1.4 100G Coherent (CFP to QSFP28-DCO)
|
|
|
|
| Milestone | Date | Source |
|
|
|-----------|------|--------|
|
|
| CFP MSA (first 100G pluggable form factor) | 2009-2010 | [CFP Wikipedia](https://en.wikipedia.org/wiki/C_Form-factor_Pluggable); 10x10G lanes |
|
|
| CFP2 MSA (half the size of CFP) | 2012 | [ProOptix history](https://www.prooptix.com/news/transceiver-form-factors/) |
|
|
| CFP2-ACO (OIF Interoperability Agreement) | 2016 | DSP on host line card; analog signal to module |
|
|
| CFP2-DCO (DSP integrated in module) | 2017-2018 | Software-configurable 100G/200G; [Acacia CFP2-DCO](https://acacia-inc.com/product/cfp2/) |
|
|
| Adtran first 100ZR QSFP28 DCO | 2022 | First coherent 100G in QSFP28 |
|
|
| Coherent QSFP28-DCO with Steelerton DSP | 2023 | [Coherent 100G QSFP28-DCO](https://www.coherent.com/news/press-releases/100g-qsfp28-dco-transceiver); <5W power |
|
|
| Coherent dual-laser QSFP28-DCO GA | Mar 2026 | [Coherent GA announcement](https://www.globenewswire.com/news-release/2026/03/06/3251306/11543/en/) |
|
|
|
|
**Total coherent miniaturization cycle: ~13 years** from CFP (2010) to QSFP28-DCO (2023). Each form factor shrink: ~3-4 years.
|
|
|
|
### 1.5 400G QSFP-DD / OSFP
|
|
|
|
| Milestone | Date | Source |
|
|
|-----------|------|--------|
|
|
| QSFP-DD MSA Rev 0.2 | May 2016 | [QSFP-DD spec](http://www.qsfp-dd.com/wp-content/uploads/2016/05/QSFP-DDrev0-2-3-29-16.pdf) |
|
|
| QSFP-DD MSA Rev 2.0 (form factor spec) | Mar 2017 | [QSFP-DD MSA announcement](http://www.qsfp-dd.com/qsfp-dd-msa-group-announces-form-factor-specification/) |
|
|
| InnoLight introduces 400G OSFP at OFC 2017 | Mar 2017 | [InnoLight OFC 2017](https://www.prnewswire.com/news-releases/innolight-technology-announced-volume-shipments-of-17-100g-qsfp28-products-and-the-introduction-of-400g-osfp-at-ofc-2017-300421866.html) |
|
|
| Oclaro 400G CFP8 PAM4 demo at OFC 2017 | Mar 2017 | [Oclaro CFP8](https://www.prnewswire.com/news-releases/oclaro-samples-400g-cfp8-pam4-enabled-transceiver-showcases-live-demo-at-ofc-2017-300425943.html) |
|
|
| Finisar 400G transceiver demos at OFC 2018 | Mar 2018 | [Finisar OFC 2018](https://picmagazine.net/article/103776/Finisar_Demos_New_400G_Transceivers_At_OFC_2018) |
|
|
| IEEE 802.3bs ratified (400G Ethernet) | Dec 2017 | 400GBASE standard |
|
|
| QSFP-DD Hardware Rev 5.0 | Jul 2019 | [QSFP-DD Rev 5.0](http://www.qsfp-dd.com/wp-content/uploads/2019/07/QSFP-DD-Hardware-rev5p0.pdf) |
|
|
| First commercial 400G QSFP-DD/OSFP modules | 2019-2020 | Broadcom TH3 switches enabled demand |
|
|
| Volume production | 2020-2021 | Driven by hyperscaler leaf/spine upgrades |
|
|
| Mainstream adoption | 2021-2022 | De facto DC interconnect standard |
|
|
|
|
**Total cycle: ~5 years** from first demos (OFC 2017) to mainstream (2022). MSA-to-volume: ~3 years.
|
|
|
|
### 1.6 400G ZR Coherent
|
|
|
|
| Milestone | Date | Source |
|
|
|-----------|------|--------|
|
|
| OIF 400ZR project initiated | ~2016-2017 | OIF response to hyperscaler DCI demands |
|
|
| OIF 400ZR IA published | Mar 2020 | [OIF 400ZR spec](https://convergedigest.com/oif-publishes-400zr-implementation/) |
|
|
| Acacia/Inphi sampling 400ZR QSFP-DD | H2 2020 | [Inphi COLORZ II](https://convergedigest.com/inphi-ramps-shipments-of-400zr-and-zr/) |
|
|
| Fujitsu sample shipments begin | Oct 2020 | [Fujitsu 400G ZR launch](https://opticalconnectionsnews.com/2020/10/fujitsu-launches-400g-zr-transceivers/) |
|
|
| Inphi commercial availability & ramp | 2021 | [Inphi ramp announcement](https://convergedigest.com/inphi-ramps-shipments-of-400zr-and-zr/) |
|
|
| Molex volume production | Early 2022 | [Molex 400G ZR volume](https://www.molex.com/en-us/news/molex-ramps-production-of-400g-zr-qsfp-dd-coherent-optical) |
|
|
| Broad volume deployment | 2022-2023 | >100% CAGR in ZR/ZR+ per Cignal AI |
|
|
|
|
**Total cycle: ~6 years** from OIF project start (~2017) to volume (2022). Spec-to-volume: ~2 years (Mar 2020 to early 2022).
|
|
|
|
### 1.7 800G DR8
|
|
|
|
| Milestone | Date | Source |
|
|
|-----------|------|--------|
|
|
| Intel first 800G DR8 OSFP sample | OFC 2021 | [Gazettabyte Intel 800G DR8](https://www.gazettabyte.com/home/2021/6/29/intel-details-its-800-gigabit-dr8-optical-module.html) |
|
|
| IEEE 802.3ck ratified (100G/lane electrical) | 2022 | Enabled 8x100G = 800G |
|
|
| Initial shipments (SR8 for AI) | 2022 | Few thousand units |
|
|
| LESSENGERS 800G SR8 volume production | Q4 2023 | [LESSENGERS announcement](https://www.semiconductor-today.com/news_items/2023/sep/lessengers-280923.shtml) |
|
|
| Shipments exceed 1M units | 2023 | Dominated by SR8 for AI clusters |
|
|
| Hyper Photonix 800G DR8 GA | May 2024 | [Hyper Photonix GA](https://www.businesswire.com/news/home/20240517136062/en/) |
|
|
| Forecast: 8M 800GbE modules shipped | 2024 | Cignal AI OFC 2024 preview |
|
|
| 800G mainstream / displacing 400G | 2025 | [Cignal AI 800GbE growth](https://cignal.ai/2025/05/800gbe-optics-shipments-to-grow-60-in-2025/) |
|
|
|
|
**Total cycle: ~4 years** from first sample (OFC 2021) to mainstream (2025). Demo-to-volume: ~2.5 years. This is faster than previous generations due to AI demand pull.
|
|
|
|
### 1.8 800G ZR/ZR+ Coherent
|
|
|
|
| Milestone | Date | Source |
|
|
|-----------|------|--------|
|
|
| OIF 800G Coherent project initiated | Dec 2020 | [OIF 800G Coherent](https://www.oiforum.com/technical-work/hot-topics/800g-coherent/) |
|
|
| Coherent first 800G ZR/ZR+ QSFP-DD unveiled | Dec 2023 | [Coherent 800G ZR announcement](https://www.semiconductor-today.com/news_items/2023/dec/coherent-081223.shtml) |
|
|
| OIF first public 800ZR multivendor interop | OFC 2024 | OIF plugfest |
|
|
| Alpha samples available | Q1 2024 | Coherent Corp. |
|
|
| OIF 800ZR Implementation Agreement published | Oct 2024 | OIF 800ZR IA |
|
|
| Coherent 800G ZR/ZR+ QSFP-DD GA | Mar 2025 | [Coherent GA](https://www.coherent.com/news/press-releases/general-availability-of-800g-zr-zrplus-in-qsfp-dd-form-factor) |
|
|
| L-band 800G ZR/ZR+ QSFP-DD | Sep 2024 | [Coherent L-band](https://www.coherent.com/news/press-releases/800g-l-band-qsfp-dd-telecom-transceiver) |
|
|
| Volume ramp forecast: >200K units, >$1B revenue | 2026 | [Cignal AI forecast](https://cignal.ai/2025/07/800g-coherent-pluggable-shipments-to-exceed-1b-revenue-in-2026/) |
|
|
|
|
**Total cycle: ~6 years** from OIF project (Dec 2020) to volume (2026). Spec-to-GA: ~5 months (Oct 2024 to Mar 2025). First demo-to-volume: ~3 years (Dec 2023 to 2026).
|
|
|
|
### 1.9 1.6T Transceivers
|
|
|
|
| Milestone | Date | Source |
|
|
|-----------|------|--------|
|
|
| Eoptolink 1.6T module demo (4xFR2, OSFP-XD) | OFC 2023 | First industry 1.6T demo |
|
|
| InnoLight 1.6T OSFP-XD demo | OFC 2024 | Live demonstration |
|
|
| First EML-based 1.6T samples ship | Q4 2024 - Q1 2025 | Conventional technology |
|
|
| IEEE 802.3dj (800G/1.6T standard, 224G/lane) | Expected mid-2026 | Under development |
|
|
| OFC 2025: Multiple live 1.6T demos | Mar 2025 | [Eoptolink Gen2 1.6T](https://www.eoptolink.com/news/361-eoptolink-launches-its-gen2-1-6t-osfp-and-osfp-rhs-transceiver-family-at-ofc-2025), [Jabil 1.6T](https://investors.jabil.com/news/news-details/2025/Jabil-Launches-1-6T-Pluggable-Transceiver/), [ATOP 1.6T demo](https://www.atoptechnology.com/ofc-2025-live-demo-atops-1-6t-osfp224-dr8-siph-module-in-action-for-next-gen-ai/) |
|
|
| SiPh-based 1.6T modules available | H2 2025 | Post mass-production readiness |
|
|
| Interop plugfest (Keysight Santa Clara) | Dec 2025 | 224G SerDes validation |
|
|
| AOI first volume order ($200M+ from hyperscaler) | Mar 2026 | [AOI volume order](https://www.globenewswire.com/news-release/2026/03/09/3251675/9986/en/) |
|
|
| Volume ramp forecast | 2026 | Dell'Oro, Cignal AI projections |
|
|
| Predicted mainstream | 2027 | >10% of addressable ports |
|
|
|
|
**Total cycle (projected): ~4 years** from first demo (OFC 2023) to mainstream (2027). Demo-to-volume: ~3 years. Accelerated by AI demand.
|
|
|
|
### 1.10 CPO (Co-Packaged Optics)
|
|
|
|
| Milestone | Date | Source |
|
|
|-----------|------|--------|
|
|
| Broadcom Tomahawk 4-Humboldt (1st gen CPO) | 2021 | First CPO chipset |
|
|
| Broadcom Tomahawk 5-Bailly (2nd gen, first volume CPO) | 2023 | Shipped to select hyperscalers |
|
|
| Broadcom 3rd gen CPO (200G/lane) | May 2025 | [Broadcom CPO glimpse](https://news.broadcom.com/) |
|
|
| Meta: 1M link-hours without link flap in lab | Oct 2025 | Broadcom announcement |
|
|
| NVIDIA CPO switches (Quantum-X: H2 2025, Spectrum-X: H2 2026) | GTC 2025 | [NVIDIA CPO plans](https://www.techradar.com/pro/nvidia-is-planning-post-copper-1-6tbps-network-tech) |
|
|
| Small initial deployments | 2026 | [Cignal AI CPO report](https://cignal.ai/2025/02/co-packaged-optics-inevitable-but-not-imminent/) |
|
|
| Volume manufacturing capability | 2027 | ASE/industry consensus |
|
|
| Widespread scale-out adoption | 2028-2029+ | [EDN CPO status 2026](https://www.edn.com/where-co-packaged-optics-cpo-technology-stands-in-2026/) |
|
|
|
|
**Total cycle: ~7+ years** from first demo (2021) to predicted widespread adoption (2028+). This is longer because CPO requires fundamental changes to packaging, connectors, and supply chain.
|
|
|
|
### 1.11 LPO (Linear Pluggable Optics)
|
|
|
|
| Milestone | Date | Source |
|
|
|-----------|------|--------|
|
|
| LPO concept development | 2022-2023 | Industry discussions on eliminating in-module DSP |
|
|
| Eoptolink 200G/lane LPO demo, 100G/lane 800G LPO mass production | OFC 2024 | [Eoptolink LPO](https://www.lightwaveonline.com/home/article/14310702/eoptolink-showcases-200g-linear-drive-pluggable-optics-at-ofc-2024) |
|
|
| LPO MSA spec (100G/lane) released | Mar 25, 2025 | [LPO MSA release](https://www.globenewswire.com/news-release/2025/03/25/3048840/0/en/) |
|
|
| LPO MSA first plugfest (interop validation) | Feb 2025 | Pre-OFC 2025 |
|
|
| FLEXOPTIX LPO products (400G/800G QSFP/OSFP) | 2025 | [FLEXOPTIX LPO](https://www.flexoptix.net/en/blog/blog/introducing-linear-pluggable-optics) |
|
|
| ECOC 2025: 800G LPO interop confirmed | Oct 2025 | [Ethernet Alliance ECOC 2025](https://ethernetalliance.org/blog/2025/10/27/ecoc-2025-interoperability-at-800g-is-given-advancing-toward-1-6t/) |
|
|
| Market share outlook | 2025-2026 | Small % of 800G market (per Cignal AI); larger potential at 1.6T |
|
|
|
|
**Note:** LPO is not a new speed generation but a new architecture. It may capture significant share at 1.6T where power savings (50% vs DSP) become critical.
|
|
|
|
---
|
|
|
|
## 2. Switch/Router ASIC Generation Timelines {#2-asic-timelines}
|
|
|
|
### 2.1 Broadcom Tomahawk (Data Center Switching)
|
|
|
|
| ASIC | Bandwidth | Process | Announced | First Switch Shipments | Source |
|
|
|------|-----------|---------|-----------|----------------------|--------|
|
|
| Tomahawk 1 | 3.2 Tbps | 28nm | Sep 2014 | Spring 2015 (~6 mo) | [Broadcom TH1](https://www.broadcom.com/products/ethernet-connectivity/switching/strataxgs/bcm56960-series) |
|
|
| Tomahawk 2 | 6.4 Tbps | 16nm | Oct 2016 | ~Fall 2017 (~12 mo) | [NextPlatform TH2](https://www.nextplatform.com/2016/10/31/broadcom-strikes-100g-ethernet-harder-tomahawk-ii/) |
|
|
| Tomahawk 3 | 12.8 Tbps | 16nm | Dec 2017 | Dec 2017 (same!) | [Broadcom TH3 press](https://www.globenewswire.com/news-release/2017/12/19/1266218/0/en/) |
|
|
| Tomahawk 4 | 25.6 Tbps | 7nm | Dec 2019 | 2020-2021 (~12-18 mo) | [NextPlatform TH4](https://www.nextplatform.com/2019/12/12/broadcom-launches-another-tomahawk-into-the-datacenter/) |
|
|
| Tomahawk 5 | 51.2 Tbps | 5nm | ~Aug 2022 | Late 2022/2023 (~6 mo) | [Broadcom TH5](https://investors.broadcom.com/news-releases/news-release-details/broadcom-ships-tomahawk-5-industrys-highest-bandwidth-switch) |
|
|
| Tomahawk Ultra | 51.2 Tbps | 4nm | 2024 | 2024 | [Broadcom TH-Ultra](https://investors.broadcom.com/news-releases/news-release-details/broadcom-ships-tomahawk-ultra-reimagining-ethernet-switch-hpc) |
|
|
| Tomahawk 6 | 102.4 Tbps | 3nm | Jun 2025 | Mar 2026 (~9 mo) | [Broadcom TH6 volume](https://www.broadcom.com/company/news/product-releases/63146) |
|
|
| Tomahawk 6 CPO (Davisson) | 102.4 Tbps | 3nm | Oct 2025 | Shipping Oct 2025 | [Broadcom Davisson](https://investors.broadcom.com/news-releases/news-release-details/broadcom-announces-tomahawkr-6-davisson-industrys-first-1024) |
|
|
|
|
**Cadence:** Bandwidth doubles approximately every 2 years. ASIC announcement to first switch: 6-18 months.
|
|
|
|
### 2.2 Broadcom Jericho (Routing / AI Fabric)
|
|
|
|
| ASIC | Bandwidth | Process | Announced | Platform Availability | Source |
|
|
|------|-----------|---------|-----------|----------------------|--------|
|
|
| Jericho2 | 9.6 Tbps | 16nm | 2018 | 2019 | [Broadcom Jericho](https://www.techinsights.com/blog/broadcom-retargets-jericho-ai-clusters) |
|
|
| Jericho2c | 4.8 Tbps | 16nm | 2019 | 2020 | Service provider market |
|
|
| Jericho2c+ | 14.4 Tbps | 7nm | 2020 (sampling) | 2021 | [Gazettabyte J2c+](https://gazettabyte.squarespace.com/home/2020/11/17/broadcoms-144-terabit-jericho2c-router-chip.html) |
|
|
| Jericho3-AI (BCM88890) | 28.8 Tbps | 5nm | Apr 2023 | Oct 2024 (first white boxes) | [Broadcom J3-AI](https://www.broadcom.com/company/news/product-releases/61156), [DriveNets/Accton](https://www.prnewswire.com/news-releases/drivenets-and-accton-technology-launch-the-highest-performance-ethernet-based-ai-networking-solution-302273214.html) |
|
|
| Jericho4 | Multi-Tbps HyperPorts | 3nm | Aug 2025 (shipping) | 2025-2026 | [Broadcom J4](https://investors.broadcom.com/news-releases/news-release-details/broadcom-ships-jericho4-enabling-distributed-ai-computing-across) |
|
|
|
|
**Cadence:** ~18-24 months between generations. ASIC-to-platform: 12-18 months.
|
|
|
|
### 2.3 NVIDIA/Mellanox Spectrum (Ethernet Switching)
|
|
|
|
| ASIC | Bandwidth | Process | Announced | Shipped | Source |
|
|
|------|-----------|---------|-----------|---------|--------|
|
|
| Spectrum | 6.4 Tbps | - | ~2016 | 2016-2017 | SN2000 series |
|
|
| Spectrum-2 | 12.8 Tbps | - | ~2018 | 2019 | SN3000 series; 200G ports |
|
|
| Spectrum-3 | 12.8 Tbps | 16nm | Mar 2020 | 2021 | [NVIDIA Spectrum-3](https://network.nvidia.com/files/doc-2020/pb-spectrum-3.pdf); 400G support |
|
|
| Spectrum-4 | 51.2 Tbps | TSMC 4N | GTC 2022 | 2023 | [NextPlatform Spectrum-4](https://www.nextplatform.com/2022/04/01/spectrum-4-ethernet-leaps-to-800-gb-sec-with-nvidia-circuits/); 800G ports |
|
|
| Spectrum-X (CPO, SN6000) | 102.4-409.6 Tbps | TBD | GTC 2025 | 2026 (SN6810/SN6800) | [NVIDIA GTC 2025](https://www.techradar.com/pro/nvidia-is-planning-post-copper-1-6tbps-network-tech) |
|
|
|
|
### 2.4 NVIDIA/Mellanox ConnectX (Network Adapters)
|
|
|
|
| NIC | Max Speed | Announced | First Shipments | Source |
|
|
|-----|-----------|-----------|-----------------|--------|
|
|
| ConnectX-5 | 100 Gb/s | Jun 2016 | Oct 2016 | [Mellanox CX-5](https://www.hpcwire.com/2016/06/16/mellanox-advances-network-computing-connectx-5-adapter/), [InsideHPC CX-5 shipping](https://insidehpc.com/2016/10/mellanox-begins-shipments-of-connectx-5-adapter/) |
|
|
| ConnectX-6 | 200 Gb/s | Jul 2019 | Mid 2019 | [Mellanox CX-6](https://network.nvidia.com/files/doc-2020/pb-connectx-6-dx-en-dellemc.pdf) |
|
|
| ConnectX-6 Dx | 200 Gb/s | Aug 2019 | Late 2019 | [CX-6 Dx](https://nvidianews.nvidia.com/news/releases-20210113-6829469) |
|
|
| ConnectX-7 | 400 Gb/s | Nov 2021 (GTC) | H2 2022 | [NVIDIA CX-7 GTC](https://www.servethehome.com/nvidia-quantum-2-400g-switches-and-connectx-7-at-gtc-fall-2021/) |
|
|
| ConnectX-8 SuperNIC | 800 Gb/s | Nov 2024 (SC24) | Q2 2025 (production) | [ServeTheHome CX-8](https://www.servethehome.com/this-is-the-next-gen-nvidia-connectx-8-supernic-for-800gbps-networking/) |
|
|
| ConnectX-9 SuperNIC | 1.6 Tb/s | Announced (Rubin) | TBD (~2027) | Spectrum-6 / BlueField-4 platform |
|
|
|
|
### 2.5 Cisco Silicon One
|
|
|
|
| ASIC | Bandwidth | Role | Announced | Platform GA | Source |
|
|
|------|-----------|------|-----------|-------------|--------|
|
|
| Q100 | 10.8 Tbps | Routing | Dec 2019 | Dec 2019 (Cisco 8000) | [Cisco Q100](https://investor.cisco.com/news/news-details/2019/Cisco-Unveils-Plan-for-Building-Internet-for-the-Next-Decade-of-Digital-Innovation/) |
|
|
| Q200 / Q200L | 12.8 Tbps | Routing / Switching | Oct 2020 | 2021 | [Cisco Q200](https://blogs.cisco.com/sp/ciscosilicononeq200announcement) |
|
|
| P100 | 19.2 Tbps | Routing (modular) | 2021 | 2022-2023 | [Cisco P100](https://www.cisco.com/c/en/us/solutions/collateral/silicon-one/silicon-one-p100-processor-ds.html) |
|
|
| G100 | 25.6 Tbps | Switching | 2021-2022 | 2022-2023 | [Cisco G100](https://www.cisco.com/c/en/us/solutions/collateral/silicon-one/datasheet-c78-744833.html) |
|
|
| G200 | 51.2 Tbps | Switching (AI) | 2024 | 2024-2025 | [Cisco G200](https://www.cisco.com/c/en/us/solutions/collateral/silicon-one/silicon-one-g200-ds.html) |
|
|
| K100, E100 | Various | Edge/Enterprise | 2025 | 2025 | Extension to enterprise |
|
|
|
|
### 2.6 Intel Barefoot Tofino (CANCELLED)
|
|
|
|
| ASIC | Bandwidth | Status | Source |
|
|
|------|-----------|--------|--------|
|
|
| Tofino 1 | 6.4 Tbps | Shipped (2016+) | [Intel Tofino](https://www.intel.com/content/www/us/en/products/network-io/programmable-ethernet-switch.html) |
|
|
| Tofino 2 | 12.8 Tbps | Shipped (7nm, CoWoS) | [ServeTheHome Tofino2](https://www.servethehome.com/intel-tofino2-next-gen-programmable-switch-detailed/) |
|
|
| Tofino 3 | 25.6 Tbps | **CANCELLED Jan 2023** | [Intel exits switching](https://www.fool.com/investing/2023/01/29/intel-exits-another-non-core-business/); P4 software open-sourced 2025 |
|
|
|
|
Intel acquired Barefoot Networks in Jun 2019, but cancelled the Tofino line in Jan 2023 as part of $3B cost-cutting. Existing Tofino 1/2 products remain available from vendors like Asterfusion.
|
|
|
|
---
|
|
|
|
## 3. The Lag Formula {#3-the-lag-formula}
|
|
|
|
Based on all historical data points collected above, here are the empirically derived lag values:
|
|
|
|
### 3.1 Technology Development Lags
|
|
|
|
| Transition | Typical Lag | Range | Trend |
|
|
|-----------|-------------|-------|-------|
|
|
| **IEEE standard publication -> First commercial transceivers** | 18-24 months | 12-36 mo | Shortening |
|
|
| **MSA spec publication -> First samples** | 6-12 months | 3-18 mo | Stable |
|
|
| **First OFC demo -> Volume production** | 24-36 months | 18-48 mo | Shortening (AI pull) |
|
|
| **First OFC demo -> Mainstream adoption (>10% ports)** | 36-48 months | 30-60 mo | Shortening |
|
|
|
|
### 3.2 ASIC-to-Deployment Lags
|
|
|
|
| Transition | Typical Lag | Range | Source |
|
|
|-----------|-------------|-------|--------|
|
|
| **ASIC announcement -> First switch platform GA** | 9-18 months | 6-24 mo | Broadcom TH history |
|
|
| **Switch GA -> Transceiver demand ramp** | 6-12 months | 3-18 mo | Qualification + deployment |
|
|
| **ASIC tape-out -> Full transceiver ecosystem ramp** | 18-30 months | 12-36 mo | Combined |
|
|
|
|
### 3.3 Regional Deployment Lags
|
|
|
|
| Transition | Typical Lag | Range | Notes |
|
|
|-----------|-------------|-------|-------|
|
|
| **US hyperscaler deployment -> Enterprise deployment** | 18-36 months | 12-48 mo | Hyperscalers are early adopters |
|
|
| **US deployment -> China deployment** | 3-6 months | 0-12 mo | Chinese vendors dominate manufacturing; fast adoption |
|
|
| **US deployment -> Europe deployment** | 12-24 months | 6-36 mo | Slower procurement cycles, GDPR considerations |
|
|
| **US deployment -> APAC (ex-China) deployment** | 12-18 months | 6-24 mo | Japan/Korea faster; SEA/India slower |
|
|
| **US deployment -> RoW deployment** | 18-36 months | 12-48 mo | Varies enormously by country |
|
|
|
|
### 3.4 Coherent Optics Miniaturization Lag
|
|
|
|
| Transition | Typical Lag |
|
|
|-----------|-------------|
|
|
| **CFP -> CFP2** | ~3 years |
|
|
| **CFP2 -> CFP2-DCO** | ~5 years |
|
|
| **CFP2-DCO -> QSFP-DD-DCO** | ~4 years |
|
|
| **400G ZR spec -> Volume** | ~2 years |
|
|
| **800G ZR spec -> Volume** | ~2 years (projected) |
|
|
|
|
### 3.5 Acceleration Factors (AI Era)
|
|
|
|
The AI/ML demand cycle is compressing timelines by approximately 30-40% compared to the cloud computing era (2012-2020):
|
|
|
|
| Factor | Impact |
|
|
|--------|--------|
|
|
| Hyperscaler pre-ordering | -6 to -12 months (demand pull) |
|
|
| Direct NVIDIA-to-transceiver vendor procurement | -3 to -6 months (bypassing OEM) |
|
|
| Chinese vendor manufacturing agility | -3 to -6 months (rapid ramp) |
|
|
| Power/thermal constraints driving urgency | -3 to -6 months (competitive pressure) |
|
|
|
|
---
|
|
|
|
## 4. Demand Cascade Model {#4-demand-cascade-model}
|
|
|
|
### 4.1 The Cascade Flow
|
|
|
|
```
|
|
[AI Training Cluster Plans]
|
|
|
|
|
v
|
|
[GPU/XPU Production Forecasts]
|
|
| (1:1 GPU-to-NIC ratio for scale-out)
|
|
v
|
|
[Switch Fabric Requirements]
|
|
| (spine-leaf topology, radix determines port count)
|
|
v
|
|
[Port Count per Switch]
|
|
| (e.g., TH5: 64x800G, TH6: 64x1.6T)
|
|
v
|
|
[Transceiver Demand per Port]
|
|
| (speed x reach = specific SKU)
|
|
v
|
|
[Revenue Forecast per Transceiver Type]
|
|
```
|
|
|
|
### 4.2 Concrete Example: GB200 NVL72
|
|
|
|
Per [SemiAnalysis](https://newsletter.semianalysis.com/p/gb200-hardware-architecture-and-component) and [NADDOD analysis](https://www.naddod.com/blog/nvidia-gb200-interconnect-architecture-analysis-nvlink-infiniband-and-future-trends):
|
|
|
|
| Component | Quantity per NVL72 Rack | Notes |
|
|
|-----------|------------------------|-------|
|
|
| GPUs (Blackwell B200) | 72 | Per rack |
|
|
| NICs (CX-7 or CX-8) | 72 | 1:1 GPU-to-NIC ratio |
|
|
| Scale-out OSFP ports | 72 | 400G (CX-7) or 800G (CX-8) |
|
|
| Spine switch OSFP ports | Varies by topology | 2:1 or 3:1 oversubscription |
|
|
| Total optical modules per 576-GPU cluster | ~18,432 | [FiberMall estimate](https://www.fibermall.com/blog/nvidia-blackwell-development-for-dac-lacc-1600g-osfp-xd.htm) |
|
|
|
|
**Speed transition:**
|
|
- CX-7 era (2024-2025): 400G SR4/DR4 per GPU port
|
|
- CX-8 era (2025-2026): 800G DR4 per GPU port, 1.6T DR8 per switch port
|
|
- CX-9 era (2027+): 1.6T per GPU port, 3.2T per switch port
|
|
|
|
### 4.3 Total Addressable Market Drivers
|
|
|
|
| Data Source | What It Reveals | Forecast |
|
|
|-------------|----------------|----------|
|
|
| Hyperscaler CapEx (quarterly reports) | Total infrastructure spend | $600-690B in 2026 ([IEEE ComSoc](https://techblog.comsoc.org/2025/12/22/hyperscaler-capex-600-bn-in-2026/), [Futurum](https://futurumgroup.com/insights/ai-capex-2026-the-690b-infrastructure-sprint/)) |
|
|
| NVIDIA GPU production (H100/B200/GB200) | GPU count -> NIC count -> optics count | [SemiAnalysis GB200](https://newsletter.semianalysis.com/p/gb200-hardware-architecture-and-component) |
|
|
| Data center construction (Synergy, JLL, CBRE) | Site capacity -> future networking spend | Multi-year pipeline |
|
|
| Optical component supplier earnings | Revenue = realized demand | Ciena backlog ~$5B heading into 2026 |
|
|
|
|
### 4.4 Key Market Forecasts (2025-2029)
|
|
|
|
| Metric | 2024 | 2025 | 2026 | Source |
|
|
|--------|------|------|------|--------|
|
|
| 800GbE module shipments | ~8M | ~12.8M (60% growth) | ~20M+ | [Cignal AI](https://cignal.ai/2025/05/800gbe-optics-shipments-to-grow-60-in-2025/) |
|
|
| 1.6T module shipments | ~2.7M | ~4.2M | ~20M+ | Industry estimates |
|
|
| 800G coherent (ZR/ZR+) units | <50K | ~100K | >200K ($1B+ revenue) | [Cignal AI](https://cignal.ai/2025/07/800g-coherent-pluggable-shipments-to-exceed-1b-revenue-in-2026/) |
|
|
| Hyperscaler CapEx | ~$256B | ~$443B | ~$600-690B | CreditSights, Futurum |
|
|
| AI back-end network market | - | - | >$20B by 2028 | Dell'Oro |
|
|
| Optical interconnect market | - | $21.9B (2026) | $40B (2031) | Mordor Intelligence |
|
|
|
|
### 4.5 Transceiver Revenue Per Unit Economics
|
|
|
|
| Speed | Avg ASP (2025) | Trend |
|
|
|-------|----------------|-------|
|
|
| 400G DR4 | $150-250 | Declining |
|
|
| 800G SR8 | $300-500 | Declining as volume ramps |
|
|
| 800G DR8 | $500-800 | At scale pricing |
|
|
| 800G 2xFR4 | $600-900 | SM premium |
|
|
| 1.6T DR8 | $1,500-2,500 | Early premium, declining |
|
|
| 400G ZR | $2,000-3,000 | Mature |
|
|
| 800G ZR/ZR+ | $4,000-6,000 | Early premium |
|
|
|
|
---
|
|
|
|
## 5. Export Control Impact {#5-export-control-impact}
|
|
|
|
### 5.1 US/EU Export Control Timeline
|
|
|
|
| Date | Action | Impact on Optical |
|
|
|------|--------|-------------------|
|
|
| Oct 2022 | Biden administration first controls | Limited advanced chip access; optical transceivers NOT directly restricted |
|
|
| Oct 2023 | Controls tightened | DSP chips (Broadcom, Marvell) restricted for some end-uses |
|
|
| Jan 2025 | AI Diffusion Rule | Broader restrictions on advanced AI computing equipment |
|
|
| Mar 2025 | Trump administration additional restrictions | More Chinese entities blacklisted |
|
|
|
|
### 5.2 Impact on Chinese Optical Transceiver Ecosystem
|
|
|
|
**Key finding:** Optical transceivers themselves are NOT directly export-controlled, but the DSP chips inside them are the vulnerability point.
|
|
|
|
| Factor | Status | Source |
|
|
|--------|--------|--------|
|
|
| Chinese vendor market share (800G) | >60% globally, >70% of 800G market | Omdia data |
|
|
| InnoLight 2024 revenue | RMB 23.86B (+122.6% YoY) | [InnoLight financials](https://iamfabian.substack.com/p/pluggables-power-and-geopolitics) |
|
|
| Eoptolink 2024 revenue | RMB 8.65B (+179% YoY) | [Eoptolink financials](https://iamfabian.substack.com/p/pluggables-power-and-geopolitics) |
|
|
| DSP dependency (Broadcom/Marvell) | ~50% of module power; critical component | Export-controlled for certain end-uses |
|
|
| LPO as strategic hedge | Eliminates in-module DSP; -20% power, -30% cost | Reduces US tech dependency |
|
|
| Chinese DSP startups | Aluksen, EOChip, Hengxin, InSiGa, Leadingspeed, Luxic, MiniSilicon, Photonic Tech, Sitrus, UXFastic | Domestic substitution push |
|
|
| Manufacturing diversification | Eoptolink Thailand factory for North American shipments | Tariff and compliance mitigation |
|
|
| SMIC vs TSMC gap for DSP | SMIC limited to ~7nm (DUV); TSMC at 3nm (EUV) | 3-5nm DSPs require TSMC |
|
|
|
|
### 5.3 Modeling Regulatory Risk
|
|
|
|
For the predictive model, export controls introduce:
|
|
|
|
1. **DSP availability risk:** If Broadcom/Marvell DSPs become restricted for a specific end-use, Chinese module vendors must either:
|
|
- Switch to domestic DSPs (12-18 month qualification delay)
|
|
- Adopt LPO architecture (6-12 month redesign)
|
|
- Source DSPs through third-party channels (uncertain)
|
|
|
|
2. **Timeline impact by scenario:**
|
|
|
|
| Scenario | Impact on Chinese Vendor Timeline |
|
|
|----------|----------------------------------|
|
|
| Status quo (current controls) | No impact; Chinese vendors dominate |
|
|
| DSP export ban for Chinese transceiver vendors | +12-24 months for domestic DSP qualification |
|
|
| Full optical component controls | +24-36 months; unlikely given US vendor dependency |
|
|
| LPO adoption accelerates | -6 months (removes DSP bottleneck entirely) |
|
|
|
|
3. **For Chinese domestic market:** +6-18 months lag vs Western hyperscaler deployment, primarily due to GPU access restrictions limiting AI cluster buildouts.
|
|
|
|
---
|
|
|
|
## 6. Predictive Timeline Calculator {#6-predictive-calculator}
|
|
|
|
### 6.1 The Formula
|
|
|
|
```
|
|
T_samples = T_current + LAG_milestone_to_samples
|
|
T_volume = T_samples + LAG_samples_to_volume
|
|
T_mainstream = T_volume + LAG_volume_to_mainstream
|
|
```
|
|
|
|
Where the lag values depend on:
|
|
|
|
#### Milestone-to-Samples Lag Table
|
|
|
|
| Current Milestone | Lag to First Samples | Confidence |
|
|
|-------------------|---------------------|------------|
|
|
| Academic paper only | 36-60 months | +/- 18 mo |
|
|
| First OFC/ECOC demo | 12-24 months | +/- 9 mo |
|
|
| MSA/IEEE spec published | 6-12 months | +/- 6 mo |
|
|
| ASIC dependency announced | Add 6-12 months from ASIC GA | +/- 6 mo |
|
|
| Interop plugfest completed | 3-6 months | +/- 3 mo |
|
|
|
|
#### Samples-to-Volume Lag Table
|
|
|
|
| Technology Type | Lag to Volume | Confidence |
|
|
|----------------|---------------|------------|
|
|
| Incremental upgrade (same form factor, higher speed) | 6-12 months | +/- 3 mo |
|
|
| New form factor (e.g., QSFP-DD, OSFP-XD) | 12-18 months | +/- 6 mo |
|
|
| New architecture (e.g., coherent, CPO) | 18-36 months | +/- 12 mo |
|
|
| Disruptive technology (e.g., CPO at scale) | 24-48 months | +/- 18 mo |
|
|
|
|
#### Volume-to-Mainstream Lag Table
|
|
|
|
| Market Segment | Lag to >10% Ports | Confidence |
|
|
|---------------|-------------------|------------|
|
|
| US hyperscaler | 0-6 months (often concurrent with volume) | +/- 3 mo |
|
|
| China hyperscaler (Alibaba, Tencent, ByteDance) | 3-9 months | +/- 6 mo |
|
|
| Enterprise (US) | 18-36 months | +/- 12 mo |
|
|
| Enterprise (Europe) | 24-42 months | +/- 12 mo |
|
|
| Service provider | 12-24 months | +/- 9 mo |
|
|
|
|
### 6.2 ASIC Dependency Modifier
|
|
|
|
If a transceiver requires a specific switching ASIC:
|
|
|
|
```
|
|
T_samples = max(T_from_milestone, T_asic_ga + 3 months)
|
|
```
|
|
|
|
The transceiver cannot ramp before the switch ASIC is available. Key dependencies:
|
|
|
|
| Transceiver Speed | Required ASIC Generation | ASIC GA |
|
|
|-------------------|--------------------------|---------|
|
|
| 400G | Broadcom TH3+ / Spectrum-3+ | Available since 2017 |
|
|
| 800G | Broadcom TH5+ / Spectrum-4+ | Available since 2023 |
|
|
| 1.6T | Broadcom TH6 / Spectrum-X / CX-8 | TH6: Mar 2026, CX-8: Q2 2025 |
|
|
| 3.2T | Next-gen (TH7? / Spectrum-6) | ~2028 projected |
|
|
|
|
### 6.3 Worked Examples
|
|
|
|
#### Example 1: 1.6T OSFP-XD DR8
|
|
|
|
**Input:**
|
|
- Technology: 1.6T OSFP-XD DR8
|
|
- Current milestone: Volume orders placed (Mar 2026)
|
|
- ASIC dependency: Broadcom Tomahawk 6 (GA Mar 2026)
|
|
|
|
**Calculation:**
|
|
- T_samples: Q4 2024 (already happened)
|
|
- T_volume: Q3 2026 (AOI $200M order ships Q3 2026)
|
|
- T_mainstream (US hyperscaler): H2 2026 - H1 2027
|
|
- T_mainstream (Enterprise US): 2028-2029
|
|
- T_mainstream (Europe): 2029-2030
|
|
|
|
**Confidence:** Medium-High (ASIC available, volume orders placed)
|
|
|
|
#### Example 2: 3.2T OSFP (hypothetical next-gen)
|
|
|
|
**Input:**
|
|
- Technology: 3.2T OSFP (16x200G or 8x400G)
|
|
- Current milestone: Concept/early research (448G PAM4 SerDes expected ~2027)
|
|
- ASIC dependency: Next-gen (~TH7, expected ~2028)
|
|
|
|
**Calculation:**
|
|
- T_first_demo: OFC 2027 (+/- 6 mo)
|
|
- T_samples: H2 2028 (+/- 9 mo)
|
|
- T_volume: H2 2029 - H1 2030 (+/- 12 mo)
|
|
- T_mainstream (US hyperscaler): 2030 (+/- 12 mo)
|
|
- T_mainstream (Enterprise): 2032+ (+/- 18 mo)
|
|
|
|
**Confidence:** Low (depends on 448G SerDes and next-gen ASIC)
|
|
|
|
#### Example 3: CPO at Scale-Out
|
|
|
|
**Input:**
|
|
- Technology: CPO (scale-out Ethernet)
|
|
- Current milestone: Lab validation complete (Meta 1M link-hours, Oct 2025)
|
|
- ASIC dependency: NVIDIA Spectrum-X CPO (H2 2026) / Broadcom Davisson (Oct 2025)
|
|
|
|
**Calculation:**
|
|
- T_initial_deployment: 2026 (small scale)
|
|
- T_volume: 2027-2028 (manufacturing capability)
|
|
- T_mainstream (>10% of DC switch ports): 2029-2030
|
|
- T_mainstream (enterprise): Unlikely before 2032
|
|
|
|
**Confidence:** Low-Medium (manufacturing scaling is the key unknown)
|
|
|
|
### 6.4 Regional Rollout Timeline Modifier
|
|
|
|
Apply these offsets from US hyperscaler deployment:
|
|
|
|
```
|
|
T_region = T_us_hyperscaler + REGIONAL_OFFSET
|
|
```
|
|
|
|
| Region | Offset (months) | Notes |
|
|
|--------|----------------|-------|
|
|
| US Hyperscaler | 0 (baseline) | Google, Meta, Microsoft, Amazon |
|
|
| China Hyperscaler | +3 to +6 | Fast adoption but GPU access limited |
|
|
| Japan/Korea Enterprise | +12 to +18 | NTT, KDDI, SK Telecom early |
|
|
| Europe Service Provider | +12 to +24 | DT, Orange, Telefonica |
|
|
| US Enterprise | +18 to +36 | Fortune 500 DC upgrades |
|
|
| Europe Enterprise | +24 to +42 | Longer procurement, GDPR |
|
|
| India/SEA | +18 to +30 | Jio, Tata leading; rest slower |
|
|
| LATAM/Africa | +30 to +48 | Limited DC infrastructure |
|
|
|
|
### 6.5 Algorithm Implementation (Pseudocode)
|
|
|
|
```python
|
|
def predict_timeline(
|
|
technology: str,
|
|
current_milestone: str, # "paper", "demo", "spec", "samples", "volume"
|
|
asic_dependency: str | None,
|
|
asic_ga_date: date | None,
|
|
is_new_form_factor: bool = False,
|
|
is_new_architecture: bool = False,
|
|
ai_demand_driven: bool = True,
|
|
) -> dict:
|
|
|
|
# Base lag from current milestone to samples
|
|
milestone_lags = {
|
|
"paper": (36, 60, 18), # (min, max, uncertainty) months
|
|
"demo": (12, 24, 9),
|
|
"spec": (6, 12, 6),
|
|
"interop": (3, 6, 3),
|
|
"samples": (0, 0, 0),
|
|
"volume": (-12, -6, 3), # Already past samples
|
|
}
|
|
|
|
min_lag, max_lag, uncertainty = milestone_lags[current_milestone]
|
|
base_samples_date = today + months(avg(min_lag, max_lag))
|
|
|
|
# ASIC dependency check
|
|
if asic_dependency and asic_ga_date:
|
|
asic_ready = asic_ga_date + months(3)
|
|
base_samples_date = max(base_samples_date, asic_ready)
|
|
|
|
# Samples to volume lag
|
|
if is_new_architecture:
|
|
volume_lag = months(27) # 18-36 range
|
|
elif is_new_form_factor:
|
|
volume_lag = months(15) # 12-18 range
|
|
else:
|
|
volume_lag = months(9) # 6-12 range
|
|
|
|
# AI demand acceleration factor
|
|
if ai_demand_driven:
|
|
volume_lag *= 0.65 # 35% acceleration
|
|
|
|
volume_date = base_samples_date + volume_lag
|
|
|
|
# Regional rollout
|
|
regional = {
|
|
"US_hyperscaler": volume_date,
|
|
"China_hyperscaler": volume_date + months(4),
|
|
"Japan_Korea": volume_date + months(15),
|
|
"Europe_SP": volume_date + months(18),
|
|
"US_enterprise": volume_date + months(27),
|
|
"Europe_enterprise": volume_date + months(33),
|
|
"India_SEA": volume_date + months(24),
|
|
"LATAM_Africa": volume_date + months(39),
|
|
}
|
|
|
|
# Confidence intervals
|
|
confidence = {
|
|
"samples": uncertainty,
|
|
"volume": uncertainty + 3 if is_new_form_factor else uncertainty,
|
|
"mainstream": uncertainty + 6,
|
|
}
|
|
|
|
return {
|
|
"predicted_samples": base_samples_date,
|
|
"predicted_volume": volume_date,
|
|
"predicted_mainstream": volume_date + months(12),
|
|
"confidence_months": confidence,
|
|
"regional_rollout": regional,
|
|
}
|
|
```
|
|
|
|
### 6.6 Historical Validation
|
|
|
|
| Technology | Predicted (using formula) | Actual | Delta |
|
|
|-----------|--------------------------|--------|-------|
|
|
| 100G QSFP28 (from OFC 2015 demo) | Volume: Q1 2017 | Volume: Mar 2017 | 0 mo |
|
|
| 400G QSFP-DD (from OFC 2017 demo) | Volume: Q1 2020 | Volume: H1 2020 | +3 mo |
|
|
| 400G ZR (from spec Mar 2020) | Volume: Q1 2022 | Volume: Early 2022 | 0 mo |
|
|
| 800G DR8 (from OFC 2021 demo) | Volume: Q1 2024 | Volume: Mid 2024 | +3 mo |
|
|
| 1.6T (from OFC 2023 demo) | Volume: Q1 2026 | Volume: Q3 2026 (projected) | +6 mo |
|
|
|
|
**Average prediction error: +2.4 months** (formula is slightly optimistic).
|
|
|
|
---
|
|
|
|
## Key Data Sources for Ongoing Model Updates
|
|
|
|
| Source | URL | What It Provides | Update Frequency |
|
|
|--------|-----|------------------|-----------------|
|
|
| Cignal AI | https://cignal.ai | Optical market forecasts, shipment data | Monthly/Quarterly |
|
|
| LightCounting | https://www.lightcounting.com | Transceiver shipment volumes, pricing | Monthly |
|
|
| Dell'Oro Group | https://www.delloro.com | Data center networking, optical transport | Quarterly |
|
|
| OFC Conference | https://www.ofcconference.org | Annual demos, product launches | Annual (March) |
|
|
| ECOC Conference | https://www.ecocexhibition.com | European demos, product launches | Annual (September/October) |
|
|
| OIF | https://www.oiforum.com | Implementation Agreements, interop | As published |
|
|
| IEEE 802.3 | https://www.ieee802.org/3/ | Ethernet standards | As ratified |
|
|
| Broadcom press releases | https://www.broadcom.com/company/news/product-releases | ASIC announcements | As released |
|
|
| NVIDIA networking | https://www.nvidia.com/en-us/networking/ | Switch/NIC announcements | As released |
|
|
| Hyperscaler quarterly earnings | SEC filings | CapEx guidance, AI spending | Quarterly |
|
|
|
|
---
|
|
|
|
## Sources
|
|
|
|
### Transceiver Timelines
|
|
- [InnoLight OFC 2017 - 100G QSFP28 Volume](https://www.innolight.com/en/news/newsinfo/13.html)
|
|
- [Oclaro OFC 2017 - 100G ER4 QSFP28](https://www.prnewswire.com/news-releases/oclaro-showcases-industrys-first-live-40km-interoperability-demo-between-100g-extended-reach-qsfp28-and-cfp2-at-ofc-2017-300426690.html)
|
|
- [Kaiam OFC 2015 - 100G QSFP28 + 400G demo](https://www.businesswire.com/news/home/20150319005175/en/)
|
|
- [ColorChip OFC 2016 - 100G PSM4 QSFP28](https://www.globenewswire.com/news-release/2016/03/18/940853/0/en/)
|
|
- [Finisar OFC 2018 - 400G demos](https://picmagazine.net/article/103776/Finisar_Demos_New_400G_Transceivers_At_OFC_2018)
|
|
- [Oclaro OFC 2017 - 400G CFP8](https://www.prnewswire.com/news-releases/oclaro-samples-400g-cfp8-pam4-enabled-transceiver-showcases-live-demo-at-ofc-2017-300425943.html)
|
|
- [QSFP-DD MSA specifications](http://www.qsfp-dd.com/specification/)
|
|
- [OIF 400ZR IA](https://convergedigest.com/oif-publishes-400zr-implementation/)
|
|
- [Inphi 400ZR ramp](https://convergedigest.com/inphi-ramps-shipments-of-400zr-and-zr/)
|
|
- [Molex 400G ZR volume](https://www.molex.com/en-us/news/molex-ramps-production-of-400g-zr-qsfp-dd-coherent-optical)
|
|
- [Fujitsu 400G ZR launch](https://opticalconnectionsnews.com/2020/10/fujitsu-launches-400g-zr-transceivers/)
|
|
- [Gazettabyte - Intel 800G DR8](https://www.gazettabyte.com/home/2021/6/29/intel-details-its-800-gigabit-dr8-optical-module.html)
|
|
- [LESSENGERS 800G volume](https://www.semiconductor-today.com/news_items/2023/sep/lessengers-280923.shtml)
|
|
- [Hyper Photonix 800G DR8 GA](https://www.businesswire.com/news/home/20240517136062/en/)
|
|
- [Cignal AI 800GbE growth](https://cignal.ai/2025/05/800gbe-optics-shipments-to-grow-60-in-2025/)
|
|
- [Coherent 800G ZR/ZR+ unveiled](https://www.semiconductor-today.com/news_items/2023/dec/coherent-081223.shtml)
|
|
- [Coherent 800G ZR/ZR+ GA](https://www.coherent.com/news/press-releases/general-availability-of-800g-zr-zrplus-in-qsfp-dd-form-factor)
|
|
- [Cignal AI 800G coherent $1B forecast](https://cignal.ai/2025/07/800g-coherent-pluggable-shipments-to-exceed-1b-revenue-in-2026/)
|
|
- [Eoptolink Gen2 1.6T OFC 2025](https://www.eoptolink.com/news/361-eoptolink-launches-its-gen2-1-6t-osfp-and-osfp-rhs-transceiver-family-at-ofc-2025)
|
|
- [Jabil 1.6T launch](https://investors.jabil.com/news/news-details/2025/Jabil-Launches-1-6T-Pluggable-Transceiver/)
|
|
- [AOI first $200M+ 1.6T volume order](https://www.globenewswire.com/news-release/2026/03/09/3251675/9986/en/)
|
|
- [Coherent dual-laser QSFP28-DCO GA](https://www.globenewswire.com/news-release/2026/03/06/3251306/11543/en/)
|
|
|
|
### ASIC Timelines
|
|
- [Broadcom TH3](https://www.globenewswire.com/news-release/2017/12/19/1266218/0/en/)
|
|
- [Broadcom TH4 (NextPlatform)](https://www.nextplatform.com/2019/12/12/broadcom-launches-another-tomahawk-into-the-datacenter/)
|
|
- [Broadcom TH5](https://investors.broadcom.com/news-releases/news-release-details/broadcom-ships-tomahawk-5-industrys-highest-bandwidth-switch)
|
|
- [Broadcom TH6](https://www.broadcom.com/company/news/product-releases/63146)
|
|
- [Broadcom TH6 volume Mar 2026](https://markets.financialcontent.com/stocks/article/marketminute-2026-3-26-the-great-ethernet-pivot)
|
|
- [Broadcom Davisson CPO](https://investors.broadcom.com/news-releases/news-release-details/broadcom-announces-tomahawkr-6-davisson-industrys-first-1024)
|
|
- [Broadcom J3-AI](https://www.broadcom.com/company/news/product-releases/61156)
|
|
- [Broadcom J4](https://investors.broadcom.com/news-releases/news-release-details/broadcom-ships-jericho4-enabling-distributed-ai-computing-across)
|
|
- [DriveNets/Accton J3-AI white box](https://www.prnewswire.com/news-releases/drivenets-and-accton-technology-launch-the-highest-performance-ethernet-based-ai-networking-solution-302273214.html)
|
|
- [NVIDIA Spectrum-4](https://www.nextplatform.com/2022/04/01/spectrum-4-ethernet-leaps-to-800-gb-sec-with-nvidia-circuits/)
|
|
- [NVIDIA Spectrum-X CPO](https://www.techradar.com/pro/nvidia-is-planning-post-copper-1-6tbps-network-tech)
|
|
- [Mellanox CX-5 announcement](https://www.hpcwire.com/2016/06/16/mellanox-advances-network-computing-connectx-5-adapter/)
|
|
- [Mellanox CX-5 shipping](https://insidehpc.com/2016/10/mellanox-begins-shipments-of-connectx-5-adapter/)
|
|
- [NVIDIA CX-7 GTC 2021](https://www.servethehome.com/nvidia-quantum-2-400g-switches-and-connectx-7-at-gtc-fall-2021/)
|
|
- [NVIDIA CX-8 SuperNIC](https://www.servethehome.com/this-is-the-next-gen-nvidia-connectx-8-supernic-for-800gbps-networking/)
|
|
- [Cisco Silicon One Q100](https://investor.cisco.com/news/news-details/2019/Cisco-Unveils-Plan-for-Building-Internet-for-the-Next-Decade-of-Digital-Innovation/)
|
|
- [Cisco Silicon One Q200](https://blogs.cisco.com/sp/ciscosilicononeq200announcement)
|
|
- [Cisco Silicon One G100](https://www.cisco.com/c/en/us/solutions/collateral/silicon-one/datasheet-c78-744833.html)
|
|
- [Cisco Silicon One G200](https://www.cisco.com/c/en/us/solutions/collateral/silicon-one/silicon-one-g200-ds.html)
|
|
- [Intel Tofino cancelled](https://www.fool.com/investing/2023/01/29/intel-exits-another-non-core-business/)
|
|
|
|
### CPO & LPO
|
|
- [Cignal AI CPO report](https://cignal.ai/2025/02/co-packaged-optics-inevitable-but-not-imminent/)
|
|
- [EDN CPO status 2026](https://www.edn.com/where-co-packaged-optics-cpo-technology-stands-in-2026/)
|
|
- [LPO MSA spec release](https://www.globenewswire.com/news-release/2025/03/25/3048840/0/en/)
|
|
- [FLEXOPTIX LPO introduction](https://www.flexoptix.net/en/blog/blog/introducing-linear-pluggable-optics)
|
|
- [Eoptolink LPO OFC 2024](https://www.lightwaveonline.com/home/article/14310702/eoptolink-showcases-200g-linear-drive-pluggable-optics-at-ofc-2024)
|
|
- [Ethernet Alliance ECOC 2025](https://ethernetalliance.org/blog/2025/10/27/ecoc-2025-interoperability-at-800g-is-given-advancing-toward-1-6t/)
|
|
|
|
### Demand & CapEx
|
|
- [IEEE ComSoc - Hyperscaler CapEx $600B+](https://techblog.comsoc.org/2025/12/22/hyperscaler-capex-600-bn-in-2026/)
|
|
- [Futurum - AI CapEx $690B](https://futurumgroup.com/insights/ai-capex-2026-the-690b-infrastructure-sprint/)
|
|
- [SemiAnalysis - GB200 architecture](https://newsletter.semianalysis.com/p/gb200-hardware-architecture-and-component)
|
|
- [FiberMall - NVIDIA Blackwell optics demand](https://www.fibermall.com/blog/nvidia-blackwell-development-for-dac-lacc-1600g-osfp-xd.htm)
|
|
- [NADDOD - GB200 interconnect analysis](https://www.naddod.com/blog/nvidia-gb200-interconnect-architecture-analysis-nvlink-infiniband-and-future-trends)
|
|
|
|
### Export Controls & Geopolitics
|
|
- [Pluggables, Power, and Geopolitics (iamfabian)](https://iamfabian.substack.com/p/pluggables-power-and-geopolitics)
|
|
- [CRS - US Export Controls on China](https://www.congress.gov/crs-product/R48642)
|
|
- [American Affairs - China Semiconductor Evolution](https://americanaffairsjournal.org/2024/11/the-evolution-of-chinas-semiconductor-industry-under-u-s-export-controls/)
|
|
- [CSIS - Limits of Chip Export Controls](https://www.csis.org/analysis/limits-chip-export-controls-meeting-china-challenge)
|
|
|
|
### Standards & Specifications
|
|
- [IEEE 802.3ae (10GbE)](https://resources.l-p.com/knowledge-center/what-is-ieee-802-3ae-10-gigabit-ethernet)
|
|
- [OIF 400ZR spec (PDF)](https://www.oiforum.com/wp-content/uploads/OIF-400ZR-02.0.pdf)
|
|
- [OIF 800G Coherent](https://www.oiforum.com/technical-work/hot-topics/800g-coherent/)
|
|
- [QSFP-DD MSA Rev 5.0](http://www.qsfp-dd.com/wp-content/uploads/2019/07/QSFP-DD-Hardware-rev5p0.pdf)
|
|
- [LPO MSA](https://www.lpo-msa.org/home.html)
|