Summary

$900M market cap on $41K trailing revenue. The bull case — POET's Optical Interposer becomes the standard platform for AI datacenter optical interconnects — is real technology chasing a real TAM. But every testable claim in the thesis fails on primary sources. Eight named customers with stated production timelines have delivered zero material revenue eighteen months later. The marquee relationship (Marvell/Celestial AI) has never been confirmed by the counterparty in any filing or transcript. Insiders are selling every share they can exercise. Retail is loaded with 270K OTM LEAPS calls. No forced buyer exists on the upside; PIPE holders at $7.25 create mechanical selling pressure on the downside.

The manufacturing transition from China to Malaysia (completed Q2 2025) explains the timeline slippage without killing the pipeline. That's the one finding that keeps this from being an outright short. But "delayed, not dead" is not an investable thesis when the key variable — customer pipeline conversion timing — is unknowable from public information.

Wait for March 31 earnings. That single data point resolves more than any amount of additional research.

What the Price Requires

Strip the $300M+ cash position and the market values the operating business at ≈$600M. To justify that:

Current reality: $41K TTM revenue (declined 91% YoY from $466K). $5.6M in firm production orders. The market implies 50-67% probability of successful commercialization at scale within 2-4 years.

Base rate for pre-revenue semiconductor companies at $500M+ market cap achieving $50M+ revenue within 2 years: approximately 15%. The market is pricing 3-4x the base rate. That's either conviction or delusion. I can't tell which from the available evidence.

Factor Decomposition

iev regress POET (250 trading days)

SPY          β = -0.67     (-5.0% of variance)
MTUM         β = +1.62     (16.0% of variance)
XLK          β = +0.60     (6.8% of variance)
Idiosyncratic              (82.2% of variance)

α = 39.7%   σ_idio = 88.6%   R² = 17.8%

82% idio passes the 75% target. Returns are stock-specific, not factor-driven. But the 39.7% reported alpha is above the no-arbitrage orthogonal bound (realistic: 5-15%). Model is incomplete — missing small-cap, AI-thematic, and speculative-growth factors. Peers (LITE 61% idio, COHR 45%, GFS 57%) all have much higher factor loadings, suggesting POET's idio purity is partly an artifact of pre-revenue illiquidity rather than genuine stock-specific edge.

Momentum loading at +1.62 beta (16% of variance) is the crash risk factor. This accumulated passively from the stock's run from $3 to $9. If momentum reverses sector-wide, POET takes outsized damage.

Edge audit by factor:

α calculation:

Bull target: $8.00 (analyst high)
Current: $5.69
Horizon: 18 months
r_f: 5%

Raw return = ($8.00/$5.69)^(1/1.5) - 1 - 0.05 = 21.0%
Edge% = 82.2% (idio only, no factor edge)
Confidence = 35% (pipeline unverified, counterparty silent, insiders selling)

α = 21.0% × 0.35 × 0.822 = 6.0%

Bear case: $3.50 target (pipeline fails, PIPE overhang liquidates)
Raw return = ($3.50/$5.69)^(1/1.5) - 1 - 0.05 = -30.6%
α_bear = -30.6% × 0.35 × 0.822 = -8.8%

Probability-weighted (bull 20%, base 30%, bear 50%):
EV(α) = 0.20 × 6.0% + 0.30 × 0% + 0.50 × (-8.8%) = -3.2%

Note: the bear α of -8.8% assumes the same 35% confidence and 82% edge%. If you grant higher confidence in the bear case (the evidence is stronger — insiders selling, zero revenue, counterparty silence), bear α deepens. At 50% confidence: α_bear = -12.6%, EV(α) = -5.1%. The asymmetry favors the downside.

Expected alpha is negative under all reasonable confidence assumptions. Not enough to short (89% vol will kill you), not enough to buy.

Primary Source Findings

The Customer Pipeline (From SEC Filings and POET Blog, Jun 2024)

Eight named customers with stated production timelines. Results after 18 months:

One unnamed "leading systems integrator" placed a $5M PO in October 2025 (6-K filed Oct 22). A 9th customer, Lessengers (South Korea), was added in Q2 2025. Neither is named as a Tier 1.

Root Cause: Manufacturing Transition, Not Product Rejection

The timeline misses are explained by a concurrent manufacturing relocation:

• Late 2024: POET acquired remaining equity in SPX (China JV with Sanan), initiated wind-down
• Dec 2024: Signed master agreement with Globetronics (GMSB) in Malaysia
• Q2 2025: "All equipment now installed and operational" at GMSB
• Jun 2025: Shipped backlog of sample engines to existing customers
• Jun 2025: Signed NationGate Solutions (second Malaysian manufacturer) for "one key customer"
• Q2 2025: "Customers actively qualifying both manufacturing facilities"
• Oct 2025: First production PO ($5M) received

This is the most important finding. It shifts the interpretation from "customers rejected POET's products" to "the factory wasn't ready." The customer relationships may be intact, just delayed 6-12 months by the China-to-Malaysia transition. Qualification of the new facilities (6-12 months industry standard) puts the earliest volume production at H1 2026.

However: POET's own 20-F acknowledges a previous "fab-light strategy which was not successful" (p.458). They are attempting the same model again with a different partner. Historical base rate for this management team on outsourced manufacturing: 0 for 1.

The Marvell/Celestial AI Gap

This is the binding constraint on the entire bull thesis.

Celestial AI's Photonic Fabric is the industry's most anticipated CPO platform. Marvell acquired it for $3.25B+ (completed Feb 2, 2026). Marvell expects $500M ARR by Q4 FY2028 and $1B by Q4 FY2029. If POET supplies the external light sources for this platform, POET's revenue trajectory transforms.

Every claim that POET is "designed into" Celestial's platform traces to POET's own 2022-2023 press releases. Primary source chain:

1. Feb 2022: POET announces supply agreement with Celestial AI (POET press release)
2. Apr 2023: POET announces "advanced purchase order" from Celestial AI (POET press release)
3. Dec 2025: Marvell announces Celestial AI acquisition (Marvell 8-K) — does not mention POET
4. Dec 2025: Marvell Q3 FY2026 transcript — does not mention POET
5. Feb 2026: Marvell completes acquisition (Marvell press release) — does not mention POET
6. All POET 6-Ks post-acquisition — do not mention Marvell

No counterparty has ever confirmed this relationship in a binding disclosure. The "advanced purchase order" (Apr 2023) may represent pre-production samples worth thousands, not volume production worth millions. POET's FY2024 revenue of $41K is consistent with negligible or zero Celestial shipments.

Worse: Marvell CEO Matt Murphy described Celestial's architecture as enabling "photonic connection directly into the XPU rather than the edge of the die" in "sharp contrast to many other CPO implementations where the photonics engine is adjacent to the XPU." POET's optical interposer IS an adjacent engine. If Marvell views POET-style approaches as architecturally inferior to what Celestial is building internally, the relationship isn't delayed — it's dead.

Meanwhile, zero peer companies mention POET in any earnings transcript. In an industry where everyone name-drops partners (GFS cites "nearly double silicon photonics revenue," COHR touts "massive CPO purchase order," AXTI reports record $60M+ InP wafer backlog), POET is invisible. The TAM validation from peers actually makes POET's failure to convert more damning, not less — the market is moving, just without them.

This gap is unfillable from public information. Probability that POET Starlight is production-locked into Marvell's platform: 30%. Probability Marvell internalizes, requalifies with another supplier, or delays the light source decision: 70%.

Positioning and Forced Actors

Insider Activity

All transactions are exercise-and-sell (M/S codes). Zero open market purchases in visible history.

Total insider selling in October 2025: ≈473K shares, $3.6M. In the same month the company raised $225M from outside investors.

Capital Raises

$405M raised in 8 months across four transactions:

Shares outstanding: ≈153M common. Fully diluted (warrants + options): ≈194M.

MMCAP International (Cayman-based, 9.99% holder) received 13.6M warrants at C$9.78 (≈$6.80 USD). This is typical PIPE deal structure with embedded upside, not conviction buying.

Options Positioning

Total options OI: ≈518K contracts (51.8M notional shares = 34% of outstanding). Jan 2027 LEAPS at 41:1 call-to-put ratio. This is retail levered speculation concentrated at $10-12 strikes that need a double.

Short interest: 7.3% of float, 1.1 days to cover. Not a squeeze setup.

Forced Actor Map

If right (stock to $8-10):

• Short covering: Weak. 7.3%, exits in one session.
• MM delta hedging: Moderate above $7. Accelerates if 55K $7-strike calls move ITM.
• Index inclusion: Unlikely (Canadian issuer, Russell US domicile rules).
• New institutional flow: Slow. Two small-shop analysts, no institutional mandate triggers.

If wrong (stock to $3.50-4.50):

• PIPE holders: Strong. $300M bought at $7.25, currently -21%. One liquidation = multi-day selling event.
• MMCAP position: Moderate. $75M at $5.50, near breakeven. Warrants at C$9.78 deep OTM.
• Retail call decay: Steady pressure as Apr/Jul/Jan OTM calls expire worthless.
• MM hedging unwind: As calls decay toward zero, delta hedges unwind = mechanical selling.

Asymmetry is wrong-way. Downside has concentrated forced sellers. Upside has no mechanical buyer. The $7.25 offering price creates a supply wall — 41.4M shares (27% of outstanding) with break-even incentive to exit at that level.

Catalyst Calendar

No catalyst with high conviction within 6 months. Both near-term high-impact events (earnings, Marvell transcript) are binary and unpredictable.

What Would Change This Assessment

Observable triggers, not narrative. Each is verifiable in SEC filings or earnings transcripts:

1. Q4 product revenue >$3M (March 31 earnings) — proves pipeline conversion. Multiple customers shipping, not just one PO. Shifts probability distribution materially.

2. Marvell names POET as Celestial AI supplier (any transcript or filing) — validates the single largest component of POET's implied value. Institutional re-rating follows.

3. Named Tier 1 customer PO >$10M (any 6-K) — "unnamed systems integrator" is unverifiable. Named = checkable. Foxconn or Luxshare disclosure would attract analyst coverage.

4. Insider open market purchase >$500K (any date) — first P-code transaction in visible history would be a conviction signal from people with the most information.

If none of these occur by Q2 2026, the thesis is dead for the foreseeable future. Four years of "revenue coming soon" without delivery is a pattern, not bad luck.

Kill Test Results

Recommendation

No position.

The technology may be real. The TAM is real. The customers are real companies. But $900M market cap on $41K revenue, with an unfillable gap on the marquee relationship and wrong-way positioning asymmetry, is not an allocation I can make.

March 31 earnings resolves more uncertainty than any amount of additional research. One number — Q4 product revenue — tells us whether 18 months of "pipeline development" produced anything beyond press releases and industry awards.

If the answer is yes, re-engage immediately. The factor profile (82% idio, low beta, low coverage) is ideal for stock-specific edge. The problem isn't the vehicle. It's the absence of verifiable evidence that the engine starts.