Semiconductors
PrivateA real photonics IP asset wrapped in a promotional, serially-diluting shell — $430M cash and a genuine 800G/light-source ramp on one side, a lost marquee customer (Marvell/Celestial), a live securities-fraud class action, a Wolfpack short and 92% share-count growth in five months on the other. Option-value long only; the equity is a narrative-and-dilution machine, not yet a business. WATCHING, not owning, until FY26 revenue actually prints against the "≥$40M" guide.
Research
The verdict
A real photonics IP asset wrapped in a promotional, serially-diluting shell — $430M cash and a genuine 800G/light-source ramp on one side, a lost marquee customer (Marvell/Celestial), a live securities-fraud class action, a Wolfpack short and 92% share-count growth in five months on the other. Option-value long only; the equity is a narrative-and-dilution machine, not yet a business. WATCHING, not owning, until FY26 revenue actually prints against the "≥$40M" guide.
Primary sources
SEC filings
Source documents — open to read in full
POET Technologies is a pre-scale-revenue photonics design house built around one asset: the POET Optical Interposer™, a wafer-level platform that integrates electronic and photonic devices on a single chip using semiconductor-style packaging (the industry calls the general approach "wafer-level chip-scale packaging") ``. The pitch: eliminate the "costly components and labor-intensive assembly, alignment, and testing methods employed in conventional photonics," and thereby make optical engines cheaper and more scalable for AI data-centre interconnect.
What it actually sells today. Almost nothing, in revenue terms. FY2025 total revenue was $1,074,865 — against a $62.96M net loss . Revenue is recognised as (a) **non-recurring engineering (NRE)** milestone payments — sample products, design reports, test reports — and (b) early sample sales of optical engines . This is not a product business yet; it is a design-and-qualification business being funded by the equity market until (management hopes) production revenue arrives in 2026–27.
Product line ``: 100G LR4, 200G FR4, 400G/800G FR4 (Tx with integrated driver, Rx with integrated TIA), 1.6T 4×FR4 Rx, 200G/lane Tx & Rx for 1.6T and 3.2T, plus LightBar / Blazar™ C-Band and O-Band external light sources (the laser-source business management is prioritising for 2026).
Customers. The 20-F names no customer and books 100% of FY2025 revenue through the Asia (Singapore) entity . The *named* commercial relationships are all post-year-end : a $5M 800G production order (Oct 2025, Globe Newswire); a now-cancelled Marvell/Celestial AI relationship (Apr 2026); a Lumilens $50M initial PO / $500M-framework (May 2026); a "Lumilens/Lumilens-adjacent" pipeline management describes as "more than 10 active customer engagements expected to exceed $100M in future annual revenue" ``. Note that several previously-touted partners (Foxconn, LITEON, Celestial/Marvell) are no longer active — see Lens 13.
Suppliers / manufacturing. Fab-light. POET divested its own fab (DenseLight, 2019) and now assembles/tests through Globetronics Manufacturing (GMSB) and NationGate Solution, both in Penang, Malaysia . It relocated out of China (dissolving the Super Photonics Xiamen JV with Sanan IC) explicitly to "mitigate the geopolitical risk in China." Some wafer/component suppliers are **sole-source** with no long-term contracts .
Bottom line for Lens 1: the moat claim is a manufacturing/integration-cost story on a large TAM; the P&L is a pre-commercial R&D burn. The gap between the two is the entire investment question.
Map, upstream → POET → end demand, with the named links unless tagged:
. POET's stated strategy is to sell *engines/light sources to module makers*, and deliberately **not** compete with them on finished modules ("sell our optical modules, once developed, into niche rather than mainstream applications") .Chokepoints: (1) sole-source wafer/laser suppliers with no contracts; (2) single Penang manufacturing cluster; (3) customer qualification — POET cannot swap suppliers/contract-manufacturers quickly because customers must re-qualify the line ``. This lens is not generic — the named links are real — but the whole chain is pre-volume, so "chokepoint" today means "single points of failure on a line that isn't yet running at scale."
The claimed moat is process/IP: 79 issued patents + 34 applications (45 issued patents directly on the Optical Interposer), covering device structures, the underlying interposer technology, applications, and fabrication processes ``. Management frames the moat as "the full semiconductorization of the photonics industry" — wafer-level integration that competitors using discrete assembly cannot match on cost/scale.
How durable is it, honestly?
Bargaining power: weak on both sides today. Suppliers are sole-source (they hold power); customers are large module makers and hyperscalers who can dual-source or vertically integrate (they hold power). POET's only lever is being early and cheap on wafer-level integration — a technology bet, not a structural moat. Verdict: the moat is a hypothesis (process cost + design-in stickiness), not yet a demonstrated one.
POET does not report product segments (it is effectively one pre-revenue segment). The only disaggregation in the 20-F is geography of revenue and geography of assets ``:
| Revenue by region (USD) | FY2025 | FY2024 | FY2023 |
|---|---|---|---|
| Europe & Asia | $505,361 | $41,427 | $191,225 |
| North & South America | $569,504 | $0 | $274,552 |
| Total | $1,074,865 | $41,427 | $465,777 |
By the operating-segment note, all $1.07M of revenue is booked in the Asia (Singapore) entity; the US and Canada entities booked $0 revenue and carry the cost base (SM&A $25.1M split US $8.0M / Canada $11.8M / Asia $5.2M; R&D $18.1M, of which Asia $15.5M) . **$312.8M of the $328.6M total-asset book value sits in the Canada entity** — i.e. it is the cash/short-term-investment pile, not operating assets .
Trend: revenue is directionally up (2024 was a near-zero trough because of the SPX/China wind-down), but the absolute numbers are noise. The only segment story that matters is the one that doesn't exist yet: 800G engines and light sources scaling in Asia in H2 2026. Q1 2026 revenue $503,389 (vs $166,760 PY) `` is the first data point on that curve.
FY2025 (audited, IFRS, the anchor) ``:
Post-year-end (the numbers the market actually trades on) ``:
Read: the FY2025 P&L tells you nothing except "burning ~$30M/yr of cash, no revenue." The story is entirely forward: does the "≥$40M" 2026 guide print? Q1's $503K is <2% of $40M, so essentially the entire year is H2-loaded and unproven. The balance sheet is the good news — this is a funded pre-revenue company, not a distressed one.
No transcripts on the shelf (transcripts/ empty; foreign filer, calls are less standardised). Sentiment reconstructed from 6-K press releases and the AGM recap ``:
The recurring pattern in the language is the tell: large framework numbers ($500M, $100M, "1M units/month") attached to early-stage POs, announced into a retail audience, with a cadence of new "landmark partnerships" replacing lapsed ones. Management says "on track" a lot; it does not yet say "shipped at scale." Bears (Lens 13) read this as a promotion; bulls read it as a genuine pre-ramp company that over-shared once and got punished. The honest sentiment read: high promotional intensity, low delivered proof, one serious self-inflicted governance wound.
POET is not valuable on any earnings/sales multiple — FY2025 revenue is $1.07M and FY2026 guidance is "≥$40M," so P/E and even EV/Sales are meaningless (EV/Sales on the guide would be ~35x on a number that is 98% un-earned). It trades on option value — a ~$1.5B equity claim on a photonics possibility, with ~$430M of that backed by cash. The right frame is to place it against the optics peers that do have revenue, to show what "success" would have to look like:
| Company | Ticker | Mkt cap | Revenue (ttm) | EV/Sales | P/E | 2026 YTD | Note |
|---|---|---|---|---|---|---|---|
| POET Technologies | POET | ~$1.54B `` | $1.07M FY25 / ≥$40M FY26E guide | n/a — not meaningful (pre-revenue) | n/a (loss-making) | ~flat / whipsaw `` | ~$430M cash ; 172.6M shares |
| Applied Optoelectronics | AAOI | $9.71B `` | $507M `` | ~18.8x `` | n/a | +441% `` | $1.1B FY26 rev guide; 800G volume shipping |
| Credo Technology | CRDO | n/a (mcap) | $1.34B `` | ~31–42x `` | ~profitable ($472M NI) `` | n/a | AI interconnect (AECs), profitable |
| Lumentum | LITE | n/a (mcap) | ~$665.5M last qtr `` | n/a | 159x `` | +166% `` | OCS backlog; laser supplier |
| Coherent | COHR | n/a (mcap) | ~$1.806B last qtr `` | n/a | n/a | +97% `` | $2B NVIDIA investment; laser/materials |
52-week range $3.87 – $20.81, now ~$8.93 — a >5x peak-to-trough. What actually moves POET:
Pattern: POET is a pure narrative/retail stock — it reacts violently to (a) new "landmark" partnership PRs and (b) short/litigation/dilution shocks, and barely at all to fundamentals (there are none yet). ~51.8% of shares are held by US holders of record, 66,512 US record holders `` — a heavily retail register. This is a name where positioning and narrative, not numbers, set the price. That cuts both ways: enormous upside on a genuine ramp, brutal drawdowns on any crack in the story.
— the 20-F officer table already flags "President (Resigned)". Losing your commercial/optics GM at the exact moment you're trying to prove commercialisation is a negative.Track record: 11 years, ~$297M accumulated deficit, and revenue still at $1M. That is either "patient deep-tech that is finally at the inflection" or "a company that has never commercialised." The bull needs the former; the history supports the skeptic.
Skin in the game — the red flag. After 11 years as CEO, Venkatesan directly owns just 7,500 common shares; all directors and officers combined own 128,715 shares (<1% of class) . His economic exposure is almost entirely **options** (strikes largely $1.27–$1.79, deep in-the-money) — i.e. upside-only, no downside alignment. Meanwhile 2025 CEO comp was **~$6.94M** (mostly a $6.0M equity award) on a company with $1M revenue . The one 5%+ holder is a hedge fund, MM Asset Management / MMCAP (1.53M shares + 12.96M warrants struck $4.26–$8.39) `` — a trading position, not a strategic anchor. No strategic/hyperscaler on the register.
Capital allocation: the defining competency here is raising money, not deploying it into returns — >$293M raised in 2025, ~$525M+ more across late-2025/2026 , funding an R&D burn. Paying **$8.2M/yr in "finance advisory fees"** to a firm that helps them raise capital is itself a tell about where management energy goes. Archetype: founder-technologist CEO + promotional capital-markets machine. The technology leadership is credible; the capital-markets/communications posture is the liability.
Acting as a forensic analyst on the FY2025 IFRS statements ``:
. But that language is a **historical-losses artifact**; as of the current balance sheet POET holds **~$430M cash** and management states cash is "sufficient to fund operating and investing activities beyond one year," with ~$313M+ "remaining to be spent" ``. Do not read the going-concern note as imminent insolvency — the real risk is dilution and commercialisation, not bankruptcy. At ~$31M/yr ops burn (call it $60–80M/yr as capex ramps), $430M is 5–10+ years of runway — an unusually well-funded pre-revenue name.Regulatory findings (required sub-section) ``:
):** a **securities-fraud class action** was filed after the April 2026 events (class period **April 1–27, 2026**; lead-plaintiff deadline **June 29, 2026**), alleging (a) POET **concealed likely PFIC status** (adverse US-shareholder tax consequences) and (b) the confidentiality breach that led to Marvell cancelling all orders . Wolfpack Research and Night Market Research have published short theses ``. This is not SEC enforcement, but it is a live, material legal + reputational overhang.No forecast.ts forecast logged (watchlist rule; and I would not commit a base EPS on a company whose revenue is 98% un-earned). EPS is the wrong metric — POET will be loss-making for years regardless. The scoreable variable is FY2026 revenue vs. the "≥$40M" guide and the cash-runway-to-scale question.
Framing the three paths on revenue, not EPS ``, all inputs labelled:
. Net loss widens toward ~$60–80M as manufacturing bring-up costs hit ``. Still burns cash; still ~$400M+ liquidity. Stock stays a narrative name.. This is the "becomes a real company" path — if 800G engines ship at volume and light sources ramp, POET re-rates toward the optics peer group (AAOI at ~19x EV/Sales on $507M — implying a much larger cap if revenue is real and durable). This is the entire long thesis.. Stock re-rates to cash value (~$430M / 172.6M ≈ ~$2.50/share of net cash, before further dilution) `` — i.e. >70% downside from ~$8.93 to hard book value. The floor is cash, and cash is being diluted.The asymmetry is genuinely two-sided and wide: net cash is ~$2.50/share, the stock is ~$8.93, and the bull case is a multi-bagger. You are paying $6.40/share ($1.1B) for the option on commercialisation. Whether that option is cheap depends entirely on your probability that the "≥$40M → $100M+" ramp is real and not the eighth pivot.
Bull case. POET is early on the single biggest hardware bottleneck in AI — moving from copper to optical interconnect — with a differentiated wafer-level integration platform, a real IP estate (79 patents), fully-qualified 800G engines already designed into customer modules, a Malaysia line rated at 1M engines/year, and — critically — ~$430M of cash to fund the ramp without a financing cliff . The Lumilens $50M PO / $500M framework and "10+ engagements > $100M" give a credible path to the "≥$40M FY2026 → $100M+ FY2027" story. If even *half* of that converts to shipped, durable revenue, POET re-rates toward peers trading at 19–42x sales — a multi-bagger from a ~$1.5B cap. The secular tailwind (AI-cluster optics market cited at ~$26B in 2026, +60% YoY ) is as strong as any in tech.
Bear case (2–3 permanent-impairment risks).
Pre-mortem (it's Jan 2028, thesis broke): FY2026 revenue came in ~$12M (guide missed on qualification slippage), Lumilens' PO never converted to volume (framework was non-binding), a second short report + the class action settled badly, POET raised again at a lower price to fund the burn, and the stock sits near net cash ~$2. The tell was always there: framework numbers, not shipped numbers.
Are multiples too high? There are no multiples — it's an option. The option premium (~$1.1B over cash) is rich for a pre-revenue name with this litigation/dilution profile, but not absurd given the TAM if you assign even ~20–30% odds to the bull.
Contrarian view (what the market is refusing to see): cuts both ways. Bulls refuse to see that this is a promotional, serially-diluting, litigation-scarred shell with a CFO who cost them their best customer. Bears refuse to see that it is genuinely well-funded ($430M is real, runway is long) and that the technology and the design-ins are real, not fabricated — a rare pre-revenue name that will not be forced to sell the story at a fire-sale price. The truth is it's both a real asset and a bad stock, and which one wins is decided by the FY2026 revenue print.
You are dismantling the bull case. The short case here is unusually well-developed because two funds have already written it ``:
A fortress-balance-sheet specialty-analog foundry that has been repriced overnight from a $45 cyclical into a $250 AI-silicon-photonics growth story — the operational inflection is real (Q1'26 op-profit +96%, $1.3B of 2027 SiPho contracts, $290M prepaid), but at ~66x forward EPS the stock already discounts most of the 2028 model, and a fresh GlobalFoundries patent-injunction suit at the ITC is an asymmetric, under-priced tail risk. Long the business, cautious on the multiple.
A real GaN/InnoSwitch franchise on a depressed-earnings base that the market has already re-rated +116% YTD onto a 2027-2028 AI-datacenter dream — at ~9x EV/sales and ~54x forward P/E with insiders selling and zero datacenter revenue today, the AI optionality is fully priced; WATCHING for the cyclical-recovery print, not the narrative.