NETWORK INTELLIGENCE / 2022—2026

One partner now carries the rebuild.

POET’s partner graph has widened, but its commercial proof has narrowed. Trace the companies behind the announcements—customers, investors, suppliers and competing routes to market.

Research snapshot: 16 Oct 2026 UTC · Focus window: Aug–Oct 2026

Lumilens is the fulcrum

A $50M initial PO and $500M+ five-year ceiling anchor the rebuild, but delivery remains conditional on development, qualification and manufacturing scale.

Adtran is the proof point

It is the only named partner with stated 2026 production orders. The rest remain at collaboration, sample or prototype stage.

The window is compressing

OCP 2026 showed 102.4T CPO/NPO systems converging quickly, while optical-component supply is reported sold out into early 2029.

RELATIONSHIP TOPOLOGY

Follow the chain outward

Select a node to inspect why it matters. Orange markers denote developments from the recent focus window; orange outlines denote a disputed or contested link.

18 connections shown

confirmed / primary-sourcedAug–Oct 2026 developmentdisputed or single-source claim

WHAT CHANGED

Signals from the outer rings

The strongest new evidence is not another POET announcement. It is capital, capacity and product movement one or two companies farther out.

06 AUG 2026 · VALIDATION

Lumilens exits stealth

>$700M @ $5.51B

Its Series C pushed total funding above $900M. It says products are already shipping under a multi-billion-dollar agreement with one of four hyperscalers—but does not identify the customer.

31 JUL 2026 · STRATEGIC OVERLAP

LITEON buys into DenseLight

$34.3M / ~21%

DenseLight makes InP lasers and was POET’s subsidiary from 2016 until its 2019 sale. The new stake strengthens LITEON’s own component path; it is historical adjacency, not a POET asset.

09 OCT 2026 · SUPPLY

InP capacity stays tight

sold out to early 2029

Lumentum’s CEO said optical parts were completely sold out, with inability to meet roughly 70% of some demand through 2027. This validates scarcity—but favors scaled incumbents.

SEP 2026 · CAPACITY

Sivers scales Glasgow

$30M / >100M lasers

The InP fab expansion targets more than 100M CW-DFB lasers annually and Q4 2027 operation. Sivers still disclosed no data-center purchase orders.

21 SEP 2026 · PRODUCT

Lessengers advances alone

3.2T glass NPO

Its ECOC platform puts DOW on glass with through-glass vias and claims up to 50% power savings. No customer beyond POET was identified.

FY26 · MANUFACTURING RISK

Globetronics stays fragile

RM56.4M loss

The Malaysian manufacturing partner reported RM80.4M revenue and RM56.4M loss after tax, amid weak loadings, impairments and governance overhang.

12–15 OCT 2026 · SYSTEM SHIFT

102.4T becomes concrete

CPO / NPO

Meta, Lightmatter with Flex, Celestica and Micas showed open-rack or 102.4T-class systems. The ecosystem’s migration is no longer theoretical; POET’s execution window centers on 2027.

11 SEP 2026 · POLICY

Expected ban did not land

FCC rule narrowed

The final rule did not directly restrict Innolight, Eoptolink or TFC, defusing the feared near-term China-transceiver ban and its automatic “China Plus One” upside.

PARTNER LEDGER

Promise versus evidence

Lumilens / Rain TreeEOI platform · 2026
$50M initial PO; $500M+ ceiling over five years; samples due late 2026.
CONDITIONAL
  • Lumilens says it is shipping to an unidentified one-of-four hyperscaler under a multi-billion-dollar agreement.
  • Its LumiCore platform includes Lumilens’ own electrical-optical interposer technology.
  • The Rain Tree acquisition timing and performance claims come only from Night Market Research, a disclosed short seller.
BULL READ

A $900M+-funded, hyperscaler-qualified partner validates wafer-level assembly and can finance the ramp.

BEAR READ

The order is qualification-dependent; own-interposer capability may make POET a bridge rather than the core.

AdtranMicroMux Quattro · 2025–26
The only named relationship with stated production orders for H2 2026 delivery.
PRODUCTION ORDER
  • POET engines are used in the 4×100G LR4 MicroMux Quattro, demonstrated at OFC 2025.
  • At its June 2026 AGM, POET said Adtran placed new production orders for second-half delivery.
  • Night Market Research disputes future scope, describing use as minimal.
BULL READ

Closest public evidence that a design converted into a product order.

BEAR READ

100G LR4 is telecom-oriented and does not itself prove traction in AI interconnects.

LITEONOptical modules · Mar 2026
Prototypes due late 2026; no disclosed purchase order; separate DenseLight investment.
CO-DEVELOPMENT
  • Q2 sales NT$52.7B, cloud business +70% YoY, net income NT$7.1B and EPS NT$3.14.
  • LITEON invested $34.3M for roughly 21% of DenseLight and a board seat.
  • No public Q1 earnings material was found in which president Anson Chiu discussed POET; an alleged denial appears only in the short report.
BULL READ

A scaled, profitable manufacturer investing heavily in AI infrastructure offers a credible route to volume.

BEAR READ

No dollar commitment, while DenseLight gives LITEON an increasingly independent component stack.

Lessengers1.6T 2×DR4 · Mar 2026
Q2 2026 sample target passed without public confirmation; 3.2T NPO shown at ECOC.
SAMPLE STAGE
  • Its 3.2T platform uses a glass packaging core, through-glass vias and DOW.
  • Lessengers claims up to 50% power savings and thermal stability.
  • No customer other than POET was publicly identified.
BULL READ

Independent NPO progress supports the underlying packaging direction.

BEAR READ

No confirmed 1.6T samples, no named customer, and the standalone platform may eventually bypass POET engines.

Sivers SemiconductorsELS modules · Sep 2025
$1.2B claimed pipeline and $30M fab build; no data-center purchase order disclosed.
PROTOTYPE / PIPELINE
  • Q2 2026 net sales SEK53.8M, down 12% YoY; adjusted EBITDA −SEK35.5M.
  • Parallel design-ins include GlobalFoundries SCALE, Jabil, O-Net/Enablence and SemiNex.
  • The Glasgow expansion targets more than 100M CW-DFB lasers per year, operational Q4 2027.
BULL READ

Capacity investment and multiple credible counterparties position its InP lasers into a constrained market.

BEAR READ

Shrinking revenue and unconverted pipeline mirror POET’s own commercialization gap; POET is one of several platforms.

Quantum Computing Inc.3.2T TFLN · Nov 2025
POET funds development; H2 2026 completion target has no public status update.
R&D
  • QCi’s October presentation showed $1.3B cash, zero debt and $42.5M backlog.
  • It acquired Luminar Semiconductor for $110M in February 2026.
  • The target is a 400G/lane TFLN modulator for 3.2Tbps optical engines.
BULL READ

A well-funded TFLN partner with acquired fab capacity can finance a leap to 400G/lane.

BEAR READ

POET is paying for development and there is no committed offtake or confirmed milestone delivery.

Foxconn / FIT800G / 1.6T · 2024
FIT is central to Nvidia rack optics; POET’s participation in that chain is unverified and disputed.
DISPUTED
  • Reports place FIT in Kyber connectors, Spectrum-6 SPX racks and Nvidia CPO cabinet supply.
  • A single Seeking Alpha bull article argues FIT’s ELSFP specs align with POET Starlight.
  • Night Market Research alleges a Foxconn executive said use of POET stopped two years earlier.
BULL READ

If Starlight qualifies into FIT ELSFP, it could reach Nvidia rack volume without a direct Nvidia contract.

BEAR READ

The commercial link itself may be stale; the Kyber path is an unverified inference, not a purchase order.

Luxshare Tech400G / 800G · 2023–24
Three announcements and a demo, but no reported revenue or 2026 progress update.
STALE / DISPUTED
  • Scope expanded in 2024 after “exceptional” 800G 2×FR4 test results.
  • Luxshare was to manufacture and sell modules globally.
  • NMR alleges two Luxshare engineers at OFC 2026 had never heard of POET.
BULL READ

Repeated technical engagement with a top-tier module maker would make qualification meaningful.

BEAR READ

Three years without reported revenue and no current update suggest a stalled program.

Mitsubishi ElectricEML engines · 2025–26
Teralight 1.6T uses four 2×200G EML chips instead of the standard eight.
DEMONSTRATION
  • Joint work targets 1.6T and 3.2T pluggable transceivers.
  • The four-laser architecture addresses cost in the most expensive transmitter component.
  • No purchase order or revenue has been disclosed.
BULL READ

A concrete bill-of-materials advantage paired with a major laser supplier can open roadmap discussions.

BEAR READ

Customer demos are not production, and a disclosed short report says nothing commercial exists.

GlobetronicsManufacturing · 2024–26
Claims mass production since Oct 2025; FY26 loss and governance issues weaken the foundation.
MANUFACTURING RISK
  • FY26 revenue RM80.4M; loss after tax RM56.4M; cash RM26.0M.
  • Its November 2025 AGM minutes said the silicon-photonics line achieved mass production in October.
  • Combined GMSB and SPX capacity is claimed above one million engines per year.
BULL READ

Malaysia creates a non-China manufacturing path and cited mass-production capability.

BEAR READ

Losses, impairments and governance overhang make it a fragile volume-ramp partner.

Sanan / SPXFormer JV · ended 2024
POET acquired full control; the former partner’s funding and exit remain a bear-case reference point.
PARTNER EXIT
  • POET acquired the remaining 24.8% of SPX on Dec 31, 2024.
  • Assembly continues in cleanrooms leased from SAIC.
  • NMR alleges Sanan contributed roughly $7M of a $25M commitment; POET filings recorded a $6.85M acquisition loss.
BULL READ

Full ownership gives POET control of China assembly while Malaysia supports diversification.

BEAR READ

A former capital partner’s withdrawal is a historic vote of no confidence.

Celestial AI / MarvellELS · 2022–26
Marvell canceled all orders in April 2026 after a disclosure/confidentiality breach.
ENDED
  • The relationship began with co-development and an advance purchase order in February 2022.
  • Marvell acquired Celestial AI for $3.25B in December 2025.
  • All POET orders were canceled April 23, 2026 and disclosed April 27.
LESSON

Technical selection can disappear before volume; disclosure discipline and direct customer confirmation matter.

STATUS

No confirmed shipments from the program are evidenced in the research record.

BEARINGS

The relationships that define the case

The recurring pattern is visible across years: launch, demo, sample target—then silence. Lumilens is large enough to break that pattern; Adtran is the one current instance of stated production orders.

FEB 2022

Celestial AI begins

Co-development plus an advance purchase order for Starlight ELS modules.

FEB 2023

Luxshare begins

400G/800G engines promised for 2H23 production; an 800G demo follows, then broader scope in August 2024.

2024–OCT 2025

Manufacturing becomes the story

POET signs Globetronics, takes full control of SPX, and Globetronics later says its silicon-photonics packaging reached mass production in October 2025.

SEP–NOV 2025

Sivers and QCi expand the roadmap

External laser sources for CPO and 400G/lane TFLN engines move the pitch beyond pluggable optics; both remain pre-production.

23 APR 2026

Marvell cancels every order

The flagship Celestial relationship ends; the cancellation is disclosed four days later.

14 MAY 2026

Lumilens becomes the rebuild

$50M initial PO, $500M+ potential ceiling, warrants and a late-2026 sample target—conditional on development and qualification.

26 JUN 2026

Adtran orders are stated

At the AGM, POET says Adtran placed production orders for delivery in H2 2026.

AUG–OCT 2026

Outer-ring evidence accelerates

Lumilens’ funding, LITEON’s DenseLight stake, Sivers’ fab build and the OCP 102.4T wave all strengthen the market thesis—without yet proving POET pull-through.

OPEN QUESTIONS

What the evidence does not settle

These are not missing footnotes. Each one is a gating diligence item for whether the partner graph converts into revenue.

01Who is Lumilens’ hyperscaler?
No source names the customer beyond one of Amazon, Alphabet, Microsoft or Meta.
02Did Lessengers ship the Q2 samples?
No public update confirms the 1.6T 2×DR4 milestone.
03Did QCi complete the H2 milestone?
No public status update was found for the 400G/lane TFLN modulator.
04Is FIT still a commercial relationship?
The Starlight-to-Kyber thesis is single-source and the partnership status is contested.
05Is GoldiLink the unnamed $5M customer?
This is a Night Market allegation only; no independent confirmation was found.
06What did LITEON’s president say?
No public Q1 material was found containing a POET comment; the alleged denial is unverified.
07When does “mass production” show in POET revenue?
Globetronics’ October 2025 claim has not been matched by disclosed POET product revenue.
08Does the $50M Lumilens PO survive qualification?
Fulfilment remains conditional on engineering, qualification and manufacturing scale.

SOURCE DESK

Primary reading

Company releases establish announced terms; independent reporting provides counterparty context. Night Market Research is a disclosed short seller and is labeled wherever used.