Engineering Tool Landscape
v1 · silicon photonics, pluggable transceivers, cpo · 39 tools · 13 personasWho uses what, all day. Hover tool chips for why-it’s-standard / why-it’s-rising. Clicks open vendor site.
Power flow & interconnectionInside the data centerData-center thermal + coolingOptics (inside DC)DCI + fiber networksEDA + advanced packagingCivil + structural (hyperscale DC)Natural gas (E&P → LNG)
Color:● Incumbent● Disruptor● Emerging● Open sourceGlyph:↑↑ Accelerating↑ Growing→ Stable↓ Losing
The physical layer connecting GPUs is in a once-in-15-years architectural shift. Pluggable optics are at peak revenue (20M+ units of 400G/800G in 2024, +60% in 2025), but power economics break at 1.6T and are catastrophic at 3.2T. Co-packaged optics (Broadcom Bailly shipping; Nvidia Quantum-X announced 2H 2025) promise 70% power savings but require a packaging supply chain (TSMC CoWoS, GF Fotonix, InP laser chips from Lumentum) that doesn’t yet exist at required volume. The core tension: pluggable incumbents (Coherent, Innolight, Lumentum, Marvell DSP) defend a $16B+ market; CPO disruptors (Ayar Labs, Lightmatter, Celestial, Ranovus, Nubis) plus vertically-integrated giants (Broadcom, Nvidia) race to own the 2027–30 transition.
Where the battles are
Every lane of work: who owns it, who's challenging, and why.
CONTESTEDCPO vs LPO vs pluggable (three-way race)· Nvidia-era scale-up architecture
Owns:Rising:
Pluggables win 2024–26 on ease-of-use; LPO gets 2025–27 niche for shortest reach; CPO takes over at 3.2T+ from ~2027 when power budgets break.
CONTESTED800G pluggable optics (DR8/FR4/SR8)· Volume market for Nvidia Hopper/Blackwell scale-out
Owns:Rising:
800G shipped ~20M+ units in 2024. Chinese vendors (Innolight + Eoptolink) captured ~60% of Nvidia’s volume. Western vendors hold high-margin variants.
CONTESTED1.6T pluggable ramp· 2026 volume crossover
"800GbE optics shipments to grow 60% in 2025" while 1.6T pilot begins on 200G/lane EMLs that only Lumentum + a few others yield well.
CONTESTEDAI-cluster optical fabric (Nvidia vs Broadcom)· NVLink + InfiniBand vs Ethernet
Owns:Rising:
Nvidia’s CPO-based Quantum-X + Spectrum-X (Mar 2025) is a direct attack on Broadcom’s Ethernet + Bailly model. Meta + Oracle adopt Broadcom; Nvidia owns InfiniBand.
CONTESTEDSilicon photonics foundry· Merchant SiPh process leadership
GF owns the merchant SiPh foundry slot today; TSMC’s COUPE is the Nvidia/AMD preferred path for 2026–27 because it piggybacks on CoWoS capacity.
SETTLEDOptical circuit switching· All-optical vs packet-switched reconfiguration
"Optical Circuit Switching Market to Exceed $2.5B in 2029." Google deployed thousands of OCS in production; HUBER+SUHNER Polatis + Calient in merchant.
CONTESTEDHollow-core fiber· Air-core NANF vs silica
Owns:
— no clear incumbent
Rising:— no serious challenger
Microsoft Lumenisity acquisition (Dec 2022) pushed HCF into production ramp 2025. No intra-DC HCF spec yet at OCP — whitespace.
FALLINGOptical EDA consolidation (Synopsys + Ansys)· Photonic-EDA near-monopoly post-merger
Owns:Rising:
Synopsys-Ansys deal (closed Jul 2025) creates near-monopoly on commercial optical EDA. Keysight got OSG divestiture. Open-source Python stacks (GDSFactory, Luceda) are the real alternative for startups.
CONTESTEDPython-native PIC layout· Open-source parametric layout vs proprietary GUI
Owns:Rising:
GDSFactory has 2M+ downloads; Luceda IPKISS offers Python + GUI. Python-first matches how photonic engineers write physics — eroding proprietary EDA seats at startups + academia.
CONTESTEDSilicon photonics foundry· Merchant SiPh process leadership
Owns:Rising:
↑↑TSMC COUPE (TSMC)
↑Tower Semiconductor PH18DA (Intel (ex-acquisition target))IMEC iSiPP (UMC licensed 2025)
GF Fotonix owns merchant SiPh slot today (Gen 2 at 200G/λ). TSMC COUPE = Nvidia/AMD preferred for 2026–27 (piggybacks on CoWoS). UMC licensed IMEC iSiPP300 Dec 2025 for 2026–27 risk production.
CONTESTEDCo-packaged vs linear pluggable vs pluggable (3-way race)· Nvidia-era scale-up architecture
Owns:
↑Coherent Corp (Coherent)
↑Lumentum (Lumentum)
↑↑Innolight + Eoptolink (combined)
↑↑Marvell Nova 2 (1.6T DSP) (Marvell)Rising:
Pluggables win 2024–26 on ease-of-use; LPO (Macom + Semtech) gets 2025–27 niche for shortest reach; CPO takes over at 3.2T+ from ~2027 when power budgets break.
CONTESTEDAI-cluster optical fabric· Nvidia NVLink+InfiniBand vs Broadcom Ethernet
Owns:Rising:
Nvidia’s CPO-based Quantum-X + Spectrum-X (Mar 2025) is a direct attack on Broadcom’s Ethernet + Bailly. Meta + Oracle adopt Broadcom; Nvidia owns InfiniBand.
Who uses what, all day
13 personas across 5 org types. Hover any tool chip for WHY.
~17,100 engineers across tracked personas. Utilities dominate raw count; hyperscalers are tiny but growing fastest.
Developers
10 roles~14,050 (82%)
Utilities
1 role~2,000 (12%)
Hyperscalers
2 roles~1,050 (6%)
Silicon Photonics Design Engineer
Ayar / Lightmatter / Celestial / foundry
15–60 at startups; 50–200 at foundries
~2,500 US
Does: FDTD sweeps (ring modulator Q/FSR); grating coupler layout; PDK DRC against Fotonix/COUPE rules; Python parametric sweeps for MPW shuttle tape-outs.
Why: Commercial FDTD (Lumerical) + IPKISS Python layouts; 3-6 month foundry shuttle cycles.
Pressure: 200G/lane transition forcing every modulator + driver redesign. Failed MPW = 1-year delay.
Transceiver Module Engineer
Innolight / Coherent / Source Photonics
200–800 at Coherent/Innolight
~3,000 US
Does: EML bin by wavelength/power; TIA/driver validation on reference PCB; 85°C reliability soak; debug PAM4 eye closure.
Why: Keysight FlexOTO for statistical-process-control on optical yield.
Pressure: 1.6T transition: EML at 200G/lane requires new device structures; yield well below 400G/lane.
DSP Engineer
Marvell / Broadcom / Inphi legacy
100–250 per firm
~1,500 US
Does: Design equalizer taps for 224G PAM4 linear Rx; characterize EML + driver non-linearity; tune FEC margin.
Why: MATLAB + Virtuoso + VPIphotonics channel modeling.
Pressure: LPO threatens DSP jobs at 224G/lane; hybrid LPO+DSP architectures emerging.
Optical Networking Engineer (hyperscaler)
Meta / Google / MSFT / AWS
50–200 per hyperscaler
~800 US
Does: Allocate wavelengths on OCS-based TPU pod; trade DAC vs AEC vs optics per link; negotiate 1.6T qualification with Coherent/Innolight.
Why: Internal control planes; standard MSA transceivers plug in underneath.
Pressure: GPU ship dates set deadlines. If Blackwell ships and optics aren’t qualified, the cluster doesn’t light.
Optical Packaging Engineer
Coherent / Lumentum / Ranovus
30–150 per firm
~2,000 US
Does: Fiber-array alignment to edge coupler (sub-μm); measure coupling loss vs temperature; develop epoxy-free laser attach.
Why: Ansys mechanical + thermal + Lumerical multiphysics (now Synopsys); active-alignment stations (Ficontec, Physik Instrumente).
Pressure: CPO yield must go from 60–70% (2024) to >95% for economic shipment — entirely a packaging problem.
Optical Test Engineer
Module vendors + test houses
30–100 per manufacturer
~1,500 US
Does: 4-corner compliance test on each 1.6T module; MPO connector inspection; CMIS-compliant test-report generation.
Why: Keysight + VIAVI + EXFO; OIF CMIS + IEEE 802.3dj compliance = vendor-locked test equipment.
Pressure: 1.6T test time is >2× 800G; throughput is a bottleneck for 20M+ modules/year.
AlsoPython / LabVIEW automation
Standards Body Participant (OIF / IEEE / OCP)
Hyperscalers + module vendors (chief architects)
2–5 delegates per co
~500 US
Does: Interops at OFC/ECOC; draft CEI-224G-VSR/LR clauses; chair IEEE 802.3dj subgroups.
Why: Paper + running-code + interop demos matter most. OIF + IEEE + OCP Optics are the three standards bodies.
Pressure: Race to freeze 1.6T / 224G specs before Blackwell-Ultra + Rubin ships (2025–27).
PrimaryOIF contribution templatesMATLAB + channel sim
AlsoInterop demos (OFC/ECOC)
DSP Engineer (coherent/linear)
Marvell / Broadcom / Cisco Acacia / Inphi legacy
100–250 per firm
~1,500 US
Does: Equalizer tap design for 224G PAM4 Rx; EML + driver non-linearity characterization; FEC margin tuning.
Why: Channel-impaired PAM4 signal recovery = mixed-signal problem requiring SPICE + system-level co-sim.
Pressure: LPO threatens DSP jobs: if linear pluggable wins, DSP moves from module to switch ASIC. 1.6T stresses linear — hybrid LPO+DSP likely.
AlsoVPIphotonics (channel modeling)
Quantum-Dot / III-V Laser Chip Engineer
Lumentum / Coherent / Quintessent
30–80 per firm
~500 US
Does: MOCVD reactor capacity + crystalline growth; facet coating yield; laser modulation bandwidth characterization.
Why: The physics bottleneck of 800G+ is EML / VCSEL chip yield; MOCVD + facet coating is the hardest step.
Pressure: Lumentum + Coherent EML shortages in 2024 trace directly to MOCVD reactor capacity. Hardest piece of 800G to scale.
AlsoVPIphotonics for laser modulation bandwidth
Optical Systems Architect (Hyperscaler)
Meta / Google / AWS / Microsoft / Oracle
15–40 per hyperscaler
~250 US
Does: Defines bandwidth-per-rack roadmap, pluggable-vs-CPO mix, fiber-plant SLA. Morning NetQ telemetry review; vendor syncs with Broadcom/Marvell; writes 3.2T pluggable RFQs.
Why: Owns internal link-budget + intent-based orchestration policy; PathWave is the sign-off reference.
Pressure: CPO vs LPO vs linear-drive still unsettled. Power-per-bit target below 5 pJ/bit. Every Meta-Lumen $6B fiber deal locks physical plant for a decade.
Optical Network Automation Engineer
Tier-1 carriers + hyperscaler net-automation
20–100 per carrier
~2,000 US
Does: Writes YANG/OpenConfig intent models + Python automation that provision optical paths, run pre-FEC BER sweeps, manage ROADM wavelength allocation across multi-vendor gear.
Why: ODTN/OpenROADM disaggregated optical networks; someone has to write the multi-vendor glue.
Pressure: Rollbacks expensive because optical path changes ripple service-impacting events; every change must pass Blue Planet digital-twin validation.
Optical Bring-Up Engineer (1.6T Compliance)
Innolight / Eoptolink / Coherent / Lumentum
5–20 per module OEM
~650 US
Does: Runs overnight TDECQ sweeps on 8x200G lanes; opens escalation when lane 5 shows 1e-3 pre-FEC BER; runs CMIS firmware handshake vs hyperscaler switch NOS.
Why: The gatekeeper. Hyperscaler won’t accept modules until they pass ONE-1600ER interop against their specific switch silicon.
Pressure: 802.3dj final standard expected 2026; every module shipped before then is “hoping” it will be compliant.
Laser / III-V Reliability Engineer
Coherent / Lumentum / Sivers / Macom / Ayar / IQE
5–15 per III-V vendor
~400 US
Does: Pulls 2000-hr HTOL data on DFB batches; runs MTTF regression; partners with foundry on ESD escapes; reviews RIN on CPO ELS lasers.
Why: CPO shifts reliability from replaceable pluggable to soldered package — a laser failure = RMA the entire switch.
Pressure: Remote laser modules (Broadcom Bailly, NVIDIA Quantum-X) demand >100,000 FIT per laser; this data gates CPO qualification.
Positioning: footprint × momentum
39 tools grouped by market share vs growth rate. Hover chips for WHY.
↑ HIGH momentum, NARROW
Challengers to watch · 14
↑ HIGH momentum, BROAD
Incumbents with momentum · 15
→ LOW momentum, BROAD
Sleeping giants · 8
← narrow · broad →top = rising · bottom = stable/declining
Disruptor timeline, 2023 → 2026
20 events with cited quotes.
Jul '22
M&AII-VI + Coherent merger closes ($6.56B)
II-VI acquires Coherent Inc; renames to Coherent Corp Sep 8, 2022. — Coherent
Dec '22
M&AMicrosoft acquires Lumenisity (hollow-core fiber)
"Hollowcore fiber cables can reduce latency by 47%." Strategic bet on low-latency optics. — Microsoft
Apr '23
SHIPBroadcom launches Jericho3-AI
Up to 32,000 GPUs at 800 Gb/s Ethernet each; the merchant-silicon Ethernet AI fabric. — Broadcom
Oct '23
M&AIntel divests Silicon Photonics to Jabil
"Jabil took over the manufacture and sale of Intel’s Silicon Photonics-based pluggable transceiver product lines." — Optics.org
Oct '23
M&AIntel divests Silicon Photonics to Jabil
"Jabil took over manufacture + sale of Intel’s Silicon Photonics pluggable transceiver lines." — Optics.org
Jan '24
M&ASynopsys announces $35B ANSYS acquisition
Includes Lumerical photonic EDA. — Optica OPN
Mar '24
SHIPBroadcom ships Bailly 51.2T CPO switch
"Industry’s first 51.2 Tbps co-packaged optics Ethernet switch." — Broadcom
Mar '24
SHIPNvidia GB200 NVL72 + Marvell Nova 2 (1.6T DSP)
5,184 passive-copper DAC cables; drives 1.6T optical roadmap. — Nvidia + Marvell
Mar '24
SHIPRanovus 6.4T CPO with integrated QD laser (OFC 2024)
"Industry’s first 6.4 Tbps co-packaged optics with integrated laser for AI/ML." OFC 2024. — Ranovus
Mar '24
$Astera Labs IPO ($700M+ raised)
Mar 20, 2024 NASDAQ IPO at $36/share. 86% retimer share + PCIe Gen6 / CXL platform. — SEC
Mar '24
SHIPMarvell Nova 2 1.6T optical DSP
"Industry’s first 1.6T optical DSP featuring 200G per lane electrical + optical I/O." — Marvell
Oct '24
$Lightmatter $400M Series D at $4.4B
"Lightmatter raises $400M Series D; quadruples valuation to $4.4B." — Lightmatter
Oct '24
$Xscape Photonics Series A $44M (Nvidia strategic)
"$44M Series A … with investment from Cisco Investments, Nvidia." Multi-λ laser source for CPO scale. — Xscape
Nov '24
SHIPTower Semiconductor 300mm silicon photonics
PH18DA at 300mm with heterogeneous InP integration. Announced Nov 26, 2024. — Tower Semi
Dec '24
$Ayar Labs $155M Series D (AMD + Intel + Nvidia)
Investors include AMD Ventures, Intel Capital, Nvidia. — Ayar Labs
Mar '25
SHIPNvidia unveils Quantum-X + Spectrum-X Photonics
"4× fewer lasers, 3.5× more power efficiency, 63× greater signal integrity." — Nvidia GTC 2025
Mar '25
$Celestial AI $250M Series C1 at $2.5B
New investors: BlackRock, Maverick Silicon, Tiger Global, Lip-Bu Tan. — Celestial AI
Jul '25
M&ASynopsys closes ANSYS acquisition
Lumerical + other optical EDA now inside Synopsys (with Keysight divestiture). — Synopsys
Dec '25
M&AUMC licenses IMEC iSiPP300 for volume production
Dec 2025. UMC targets 800G + 1.6T pluggable risk production 2026–27. — IMEC
Where the time actually goes
Estimated planning-engineer hours per stage (LBNL, MISO, NERC).
Silicon photonic design
28%Commercial FDTD + Python layouts; MPW shuttle cycles 3–6 months
Owns:↑↑Marvell Nova 2 (1.6T DSP) (Marvell)
↑Ansys Lumerical FDTD (Synopsys)
↑GlobalFoundries Fotonix (GlobalFoundries)↑Tower Semiconductor PH18DA (Intel (ex-acquisition target))
↑Synopsys OptoCompiler (Synopsys)
↑Keysight PathWave ADS 2026 (Keysight)→Ansys Lumerical INTERCONNECT (Synopsys)→Synopsys OptoDesigner + OptSim (Synopsys)
→AIM Photonics (DoD + state consortium)→Synopsys OptSim + OptSim Circuit (Synopsys)
Rising:
↑↑Broadcom Bailly (CPO) (Broadcom)
↑↑Nvidia Quantum-X Photonics (Nvidia)
↑↑Ayar Labs TeraPHY
↑↑Lightmatter Passage
↑↑Celestial AI Photonic Fabric↑↑TSMC COUPE (TSMC)
↑↑GDSFactory (open-source)
↑↑Nubis Communications XT1600
↑↑Xscape Photonics
↑Luceda IPKISS
↑Ranovus Odin
↑VPIphotonics Design Suite
→Nazca Design (open-source)
→KLayout + SiEPIC-Tools
Optical packaging
22%Alignment is the #1 CPO cost driver; mechanical + optical co-sim
Module test + compliance
20%Keysight + VIAVI + EXFO compliance; 4-corner test per module
System integration
15%Hyperscaler fabric integration; OCS + Ethernet spine planning
Fabric deployment
8%MPO patch-panel + cross-connect install
Monitoring + support
7%Live fabric health + OTDR + RMA
Workflow map
Tools owning each interconnection-study stage.
Software stack, by category
Every software category a utility / developer / hyperscaler runs. Hardware + physical-layer vendors at the bottom.
Optical Chip / IP
10 tools
Whitespace
Gaps incumbents handle badly and SaaS hasn't closed.
CPO field-serviceability: Soldered CPO is a warranty nightmare; field-replaceable CPO sockets are unbuilt.
Multi-wavelength laser scalability: Only Xscape announced a productionable 8-λ source; need 32-λ roadmap for 3.2T+.
On-chip laser (heterogeneous InP on SOI): Tower PH18DA + imec iSiPP300 in pilot; no 300mm high-volume yield yet. Biggest reliability risk in CPO.
Optical BMC/telemetry stack: Optical links need out-of-band health monitoring at switch scale — hyperscalers build this in-house; no merchant product.
Open photonic EDA backbone: Post Synopsys-Ansys-Keysight, gap for Python-native open flow (GDSFactory, Luceda) serving hundreds of CPO startups.
Intra-DC hollow-core spec: Microsoft proved value metro/campus; no OCP HCF spec for intra-DC form factors.
Why now
Forces moving the market 2024–26.
Per-lane rate cliff: 224G PAM4 electrical in copper can’t go beyond ~1.5m without retimers — optical is physically required at GPU scale-out distances.
Power envelope at rack: NVL72 uses 120+ kW/rack; pluggable optics at 30W per 1.6T port = 50%+ of switch power. 70% CPO savings returns 15% of rack power to GPUs.
IEEE 802.3df + OIF CEI-224G stable: Specs frozen, so high-volume tooling (Keysight M8050A, VIAVI ONE-1600) shipping in volume.
GPU shipment deadlines: Blackwell-Ultra (2025) + Rubin (2026) force networking to match.
Foundry capacity unlocks: TSMC COUPE qualifies 2025; UMC licenses iSiPP300 for 2026–27; GF Fotonix Gen 2 at 200G/λ.
Geopolitical hedging: AWS-AOI $4B deal; MSFT-AOI supply agreement — hyperscalers pay premium for non-Chinese supply.
Hardware + physical-layer vendors
The physical vendors that sit under the software stack \u2014 cold plates, conductors, sensors, transceivers, cables, foundries, EPCs.
Optical Switch
3 tools
Silicon Photonics Foundry
12 tools
April 2026 snapshot. Headcounts are mid-point estimates. Data in
src/lib/data/research-tools.ts.