Skip to content
primarysourced Photonics sector POET
POET
~9 min read · 2,091 words ·updated 2026-04-29 · confidence 35%

Confidence legend: ✓ verified-primary · ◐ partial / aggregator · ⚠ inferred / estimate.

This page lays out the structural ecosystem around POET Technologies: who buys POET’s Optical Engines, who buys their products, what alternative integration platforms exist, and which forces are reshaping the silicon-photonics merchant-supply-chain. POET sits two layers down from the demand signal — the hyperscaler buys a transceiver from a module maker, the module maker buys an optical engine from POET — so understanding both layers (transceiver makers’ competitive structure AND hyperscaler buying behavior) is essential to forecasting POET revenue.

1. Where POET sits in the supply chain

The pluggable-optical-transceiver supply chain has five layers, each governed by different competitive dynamics:

LayerPlayer examplesWhat they sell
End-customerAWS, Microsoft Azure, Google Cloud, Meta, OracleDatacenter compute services (purchase intent for optical interconnects is derived demand)
Network OEMCisco, Arista, Juniper, NVIDIA Networking (Mellanox), HPESwitch / router platforms — cage placements consume pluggable optics
Transceiver / module makerInnolight (Suzhou), Eoptolink (Chengdu), Coherent (US), Lumentum (US), Accelink (Wuhan), Hisense Broadband (Qingdao), HG Genuine, AOI (US/Taiwan)Pluggable transceivers (QSFP-DD800, OSFP-1.6T, etc.) — sell to network OEMs and direct to hyperscalers
Component / engine supplierPOET (Optical Interposer), Marvell (DSP + Inphi optics), Macom, Coherent Discrete (formerly II-VI), Broadcom (Sipho), Nokia / OpenLight, Intel Silicon Photonics (now spun-out)Optical engines, DSPs, modulators, drivers, TIAs
Wafer / III-V foundryTSMC, GlobalFoundries (silicon photonics), Tower Semiconductor, IMEC, AIM Photonics, Sanan IC (compound semis), Almae Technologies (III-V)Foundry services

POET’s place in this stack is at the component / engine supplier layer. Its primary customers are the transceiver / module makers. Its competitors are other integration platforms — Marvell+Inphi photonics, Intel/now-Silicon-Light spinout, Nokia/OpenLight, Broadcom Sipho-based modules — plus the in-house captive integration that some module makers do themselves (Coherent / Lumentum integrate III-V devices using their own back-end-of-line packaging).

The “two layers down” insight: POET’s revenue does not move 1:1 with hyperscaler capex; it moves 1:1 with transceiver-maker order flow, which itself moves 1:1 with hyperscaler capex minus qualification-cycle delays. A typical optical-engine design-in cycle is 12–24 months from sample shipment to volume revenue. ⚠ industry consensus / operating-cycle-norm

2. Hyperscaler vertical-integration pressure (the dominant strategic threat)

The single most-important structural variable for POET (and every merchant transceiver and optical-component supplier) over 2026–2030 is whether hyperscalers vertically integrate the optical engine layer.

Disclosed hyperscaler-side optical efforts (as of 2026-04-29)

  • NVIDIA Photonics initiative — At GTC 2024, NVIDIA disclosed silicon-photonics co-packaged-optics (CPO) prototypes. Spectrum-X family CPO switch was announced 2025; full ramp expected 2027–2028. ◐ NVIDIA Spectrum-X CPO (NVIDIA IR landing)
  • Amazon AWS Trainium — December 2024 Marvell five-year contract for Trainium custom silicon; AWS internal photonics roadmap not publicly detailed but inference is that AWS will use external pluggable optics through 2026, with CPO consideration starting 2028. ⚠ inferred
  • Google “Apollo” optical interconnect — Google has long deployed proprietary Apollo optical-circuit-switch technology in production, but this is for TOR-to-spine fabric, not pluggable optics; does not directly displace POET-class engines. ◐
  • Meta MTIA + photonic accelerator coupling — Per April 2026 Broadcom announcement, Meta is partnering with Broadcom on MTIA v3 plus a photonic interconnect path; details pending. ◐ Broadcom Press Release April 14 2026

Read-through to POET

The hyperscaler-vertical-integration risk operates over a 3–5 year horizon, not 12 months. Two tactical mitigants protect POET on that horizon:

  1. The merchant 1.6T pluggable cycle (2026–2028) is already past the design-in window — hyperscaler 1.6T deployments through 2028 will be served by merchant pluggables. Captive CPO does not start materially displacing them until 2028+.
  2. CPO is technically demanding — yields, thermal, and serviceability problems on first-generation CPO modules are well-documented. Even hyperscaler-led CPO programs are likely to coexist with merchant pluggables for redundancy / flex-capacity reasons through at least 2030. ⚠ industry consensus

POET’s competitive frame is thus: win the 1.6T merchant pluggable design-in cycle from 2026–2028 (volume revenue 2027–2029); the longer-term CPO transition is a 2028+ problem.

3. Transceiver-maker customer landscape

POET’s direct customers are the optical-transceiver / module makers. The competitive structure of THIS layer is what matters for POET’s pricing power, channel diversification, and revenue concentration.

The China-domiciled module-maker bloc

  • Innolight Technology Suzhou — Largest single optical-transceiver maker globally by volume (~25–30% global pluggable share by units in 2024 ⚠). Major customer for hyperscaler-grade 800G/1.6T modules. Privately held; financial detail limited.
  • Eoptolink Technology (Chengdu) — Second-largest by volume (~15–20% share ⚠). Public on Shanghai Stock Exchange; launched Gen2 1.6T OSFP family at OFC 2025. ✓ Eoptolink press release at OFC 2025
  • Hisense Broadband (Qingdao) — Subsidiary of the Hisense home-appliance giant; entered transceiver market through 2014 acquisition of Multiplex (a US-based transceiver maker). Maintains design centers in the US and China. Relevant POET interest because the Multiplex acquisition target was a former employer of POET’s Dan Meerovich. ✓ POET 20-F filed 2026-03-31, Item 6.A bio of Meerovich
  • Accelink Technologies (Wuhan) — China State-backed; large in carrier-class optics, smaller in datacom relative to Innolight/Eoptolink. ⚠
  • HG Genuine Optics (Wuhan / Shenzhen) — Smaller Chinese pluggable maker focused on lower-speed datacom; less direct POET overlap.

The Western module-maker bloc

  • Coherent Corp. (formerly II-VI Incorporated; merger closed 2022) — Public, NYSE: COHR. Vertically integrated — owns its own modulator / laser / TIA process flows. Coherent is both a customer and a competitor to POET-class platforms; competes by integrating III-V devices using internal back-end-of-line, plus by buying merchant DSPs (Marvell, Broadcom).
  • Lumentum Holdings — Public, NASDAQ: LITE. Owns CMS (Communications and Sensing) optical-component platforms; primarily a discrete-optics + III-V supplier; exposure to pluggable transceivers smaller than Coherent’s. Vertically integrated similar to Coherent.
  • AOI (Applied Optoelectronics, Inc.) — Public, NASDAQ: AAOI. US-listed but with significant Taiwan-based manufacturing. Smaller than Innolight / Eoptolink in absolute volume; focused on US-billed transceivers.
  • Cisco Optics (post-Acacia) — Cisco acquired Acacia Communications March 2021 ($4.5B). Acacia became Cisco’s coherent and PAM4 internal-engine source. Cisco-Acacia is primarily a captive consumer of its own optical engines, not a merchant POET customer. ◐ Cisco-Acacia acquisition press release

Implication for POET

POET’s revenue concentration risk runs through Innolight + Eoptolink + Hisense Broadband. These three customers likely account for the majority of POET’s design-in pipeline (POET does not disclose customer concentration in its 20-F as a percentage of revenue, because revenue is still nominal). The bear case worry is that:

  • (a) BIS export-control rules tighten on China-domiciled module makers (see regulatory landscape), forcing them to source from non-US optical-engine suppliers, OR
  • (b) Hyperscalers shift module-procurement away from China-domiciled vendors toward Coherent / Lumentum / AOI as a geopolitical hedge.

Either path compresses POET’s serviceable customer base. The primary mitigant: POET is building Penang-based PTM (April 2025) precisely to enable manufacturing-of-record outside China. ⚠ inferred mitigation pathway

4. Foundry-allocation politics

Silicon photonics requires either:

  • Standard CMOS-compatible foundries with photonic process modules — TSMC, GlobalFoundries 90nm SiPh, Tower Semiconductor SiPh — for the silicon waveguide layer.
  • Compound-semiconductor (III-V) foundries for the active-device layer — Sanan IC (China), Almae Technologies (France), AWSC (Taiwan), Smart Photonics (Netherlands), POET via SPX (Xiamen) — for InP / GaAs lasers and modulators.

POET’s Optical Interposer specifically:

  • Uses a passive silicon waveguide substrate that the 20-F describes as fabricated using “advanced wafer-level semiconductor manufacturing techniques” (the 20-F does not disclose the specific foundry; ⚠ multiple unnamed foundry relationships).
  • Mounts III-V active devices via flip-chip bonding; the III-V devices were historically supplied by DenseLight (POET’s former subsidiary, divested 2019) and now by external partners + the SPX subsidiary (post-consolidation 2025-12-31).

The foundry-capacity risk

  • TSMC SiPh capacity is the gating bottleneck for the broader silicon-photonics industry; allocation decisions favor large customers (Nvidia, Marvell, Broadcom) over smaller ones. POET, as a development-stage micro-cap, has limited TSMC negotiating leverage. ⚠ industry consensus
  • GlobalFoundries 90nm SiPh has been positioned as the merchant-friendly alternative; GlobalFoundries’ silicon photonics platform page ◐ describes the offering. POET has not publicly disclosed which foundries it uses for the interposer layer.
  • AIM Photonics (US) — Albany NY-based research-and-prototyping foundry funded by US DOD; relevant for early-stage prototyping but not volume production.

The Sanan dependency

Pre-2024, POET’s III-V active devices were partially sourced through SPX (a JV with Sanan IC). The 2024–2025 unwind (see jv history) gives POET full ownership of SPX, which in turn means POET’s III-V supply runs through a wholly-owned PRC subsidiary fed by Sanan IC merchant compound-semi services. The Penang subsidiary (PTM) provides one path to diversify away from Sanan dependency over time. ⚠

5. Merchant-vs-captive integration platform competition

The fundamental question for any merchant silicon-photonics platform is: why does a transceiver maker buy from POET vs. (a) build internally or (b) buy from a competing merchant platform?

POET’s value proposition (per 20-F Item 4.B)

The 20-F frames the Optical Interposer thesis explicitly: “POET’s Optical Interposer eliminates costly components and labor-intensive assembly, alignment, and testing methods employed in conventional photonics. We believe the cost-efficient integration scheme and scalability of the POET Optical Interposer brings value to devices or systems that integrate electronics and photonics, including high-growth areas of communications and computing.” ✓ Form 20-F filed 2026-03-31, Item 4.B

The economic argument: traditional optical-transceiver assembly is labor-intensive (active alignment of discrete optical components requires precise sub-micron alignment, typically done with active feedback alignment systems that scale poorly). Wafer-level chip-scale packaging (WLCSP, the term the semiconductor industry uses for the same concept) reduces unit assembly cost and improves yield by moving alignment to the wafer-fab front-end-of-line.

Competing merchant platforms

  • Marvell + Inphi optics — Inphi acquired by Marvell April 2021. Marvell sells optical DSPs (Spica / Ara family) plus integrates III-V devices via Inphi-origin processes. Vertically broader than POET (DSP + optics) but with a captive demand channel via Marvell-supplied DSPs.
  • Nokia / OpenLight — Nokia spun out OpenLight (joint venture with Synopsys) in 2022. OpenLight licenses heterogeneous integration on InP-on-Si platform; competitive with POET’s interposer at the platform level. ◐
  • Intel Silicon Photonics → Jabil — Intel divested its silicon-photonics business to Jabil in 2024; the spun-out entity continues to supply optical components.
  • Broadcom Sipho — Broadcom’s silicon-photonics platform powers Tomahawk-Ultra family CPO references; primarily used internally for switch ASIC + CPO integration, not sold merchant.

Relative positioning

POET differentiates on:

  1. Merchant-only — no competing module-business unit (Coherent/Lumentum compete with their own customers).
  2. Wafer-level integration approach — passive Si waveguide + flip-chip III-V (vs. monolithic InP-on-Si or full-CMOS SiPh).
  3. Multiple data-rate platform — 100G / 400G / 800G / 1.6T scaling claimed via the same interposer architecture.

POET is disadvantaged on:

  1. Scale — POET’s R&D budget (~$18M in 2025) is <1% of Marvell’s optical R&D.
  2. DSP integration — POET does not sell its own DSP; module makers must source DSPs from Marvell, Broadcom, or MaxLinear and integrate.
  3. Customer references — POET has not publicly disclosed volume design wins at named tier-1 customers; competitive merchant platforms have multi-quarter design-win lists.

6. Reading the silicon-photonics ecosystem signals

For an LWLG-style analyst tracking POET, the highest-value signals to monitor are:

  • Innolight / Eoptolink quarterly call commentary on optical-engine sourcing diversity — any mention of POET as a qualified supplier shifts probability mass toward 2026–2027 design-in monetization. ◐
  • Coherent / Lumentum quarterly calls on internal vs. external optical-engine balance — if either signals more captive integration, that compresses POET’s serviceable customer base.
  • NVIDIA / Broadcom CPO ramp commentary — accelerating CPO timelines compress POET’s merchant-pluggable monetization window.
  • BIS Federal Register publications on advanced-DSP / advanced-modulator export controls (see regulatory landscape).
  • POET 6-K announcements of customer engagements, foundry partnerships, or volume-shipment milestones.

The single most-important leading indicator: the first credible disclosure of POET volume revenue from a named tier-1 customer. POET’s revenue line in fiscal 2025 was nominal; the inflection comes when a 1.6T module program begins shipping with POET inside, which would first appear in a 6-K rather than the 20-F.

Cross-section pointers

  • tam sam — Volume / dollar sizing for the markets named here.
  • regulatory landscape — China BIS / CFIUS / Investment Canada Act exposure for the China-customer dependency.
  • jv history — SPX / Sanan supply-chain history.
  • overview — Customer and partner profiles for the named module makers.
  • overview — Optical Interposer technical detail; pairs with the merchant-vs-captive section above.
  • bear case — Hyperscaler-vertical-integration risk gets developed further in the bear case.