Markets Confusing? Ask Edgen Search.
Instant answers, zero BS, and trading decisions your future self will thank you for.
If you opened the markets Monday wondering why POET Technologies stock crashed 47%, the answer is brutally simple: the company's own CFO talked his way out of the most important contract on its books. On April 23, 2026, Marvell Technology sent written notice cancelling every purchase order POET had inherited through Marvell's February acquisition of Celestial AI. The trigger was not a product failure or a yield miss — it was a Stocktwits interview in which CFO Thomas Mika named the customer, described the workload, and walked retail investors through what should have been confidential supply terms. By Monday's close, POET (NASDAQ: POET) printed a -47.35% session to $7.95, the largest single-day decline in its history.
This is a diagnostic on what happens when a pre-revenue silicon-photonics name with a 20-year unprofitability streak loses, in 30 seconds of unscripted speech, the only commercial validation it had.
What happened to POET on April 27, 2026: a 47% single-day crash explained
POET opened the week near $15 after rallying ~70% in the prior eight sessions on the Marvell-Celestial inheritance thesis. By 9:35 a.m. ET Monday, after the 6-K disclosed the cancellation, the bid disappeared. POET printed a 47.35% loss, closing at $7.95 on ~9× average volume.
Three things made the crash this violent:
- No prior warning. The cancellation was dated April 23; POET disclosed April 27. Retail rally buyers went from a 70% paper gain Friday to a double-digit loss Monday morning.
- No fundamental cushion. POET's 2025 Form 20-F reported just $1 million in total sales and a $62.96 million net loss. The Marvell-Celestial pipeline was, in management's words, the "potentially revolutionary cash stream." Strip it out and the equity reverts to pure optionality.
- A class-action lawsuit hit Monday afternoon. Securities firm Gibbs Mura began soliciting impacted investors the same day, citing disclosure-process failures.
The market is not pricing a contract dispute — it is pricing the loss of POET's only credible commercial reference for its Optical Interposer platform.
The Stocktwits interview: how POET CFO Thomas Mika triggered the NDA violation
The POET CFO Thomas Mika Stocktwits blunder is now a textbook case of how not to communicate a confidential semiconductor contract. On April 21, 2026, Mika sat for a high-profile Stocktwits video titled "Capacity. Capability. Credibility." The framing was investor-relations 101: validate POET's Optical Interposer to a retail audience that had been waiting years for a tangible customer name.
What Mika said, on camera:
"We're a supplier to Marvell now that they've acquired Celestial AI, who has been a customer of ours for a couple of years. And what we supply to Celestial AI are light sources — high-bandwidth, multi-frequency, high-power light sources that light up the photonic fabric that Celestial AI talks about as being the communication device between GPUs and one GPU and another GPU, a GPU and a memory device."
In one sentence, Mika disclosed the customer (Marvell, post-Celestial), the product (multi-frequency, high-power light sources), the application (GPU-to-GPU and GPU-to-memory photonic fabric), and the IP context (Celestial AI's publicly described core fabric). In separate comments he hinted the order could "exceed $5 million" — small in absolute terms but competitively sensitive shipment data.
Every one of those points is the kind of supply-chain detail standard semiconductor NDAs forbid. Marvell, like NVIDIA and Broadcom, treats photonic interconnect roadmaps as a strategic moat: naming the "light source" vendor in Celestial AI's fabric tells competitors which third parties to court, what bandwidth class is in play, and roughly when product is shipping. There is no plausible reading in which Mika's interview did not breach the NDA.
POET Marvell Celestial AI cancellation NDA: walking through the contractual logic
The POET Marvell Celestial AI cancellation NDA chain is mechanically clean once you lay out the timeline:
| Date | Event |
|---|---|
| Apr 2023 | POET begins shipping production-unit orders to Celestial AI. |
| Feb 2026 | Marvell closes Celestial AI acquisition; inherits POET orders. |
| Apr 21, 2026 | CFO Mika gives Stocktwits interview disclosing customer, product, order size. |
| Apr 23, 2026 | Marvell sends written cancellation notice citing confidentiality breach. |
| Apr 27, 2026 | POET files 6-K. Stock crashes 47.35% to $7.95. |
From Marvell's perspective, the cancellation logic is clean. The inherited Celestial AI purchase orders carried standard confidentiality terms that survive a change of control. Once POET's CFO disclosed customer identity, product specification, and shipment magnitude on a public investor platform, Marvell had two choices: accept the disclosure and lose leverage with every other photonics vendor, or invoke the confidentiality clause and exit. Marvell chose to exit, and almost certainly had the contractual standing to do so.
There's also a strategic angle. In its own AI custom-silicon disclosures (see our Marvell stock analysis), Marvell has flagged optical I/O as the bandwidth wall it intends to attack with in-house IP — and Celestial AI was the centerpiece of that strategy. A vendor that publicly outs Marvell's photonic-fabric supply chain is a vendor Marvell did not want in the next-gen platform anyway. Some observers ask whether the NDA breach was the reason or merely a reason — a clean legal exit from a relationship Marvell wanted to bring in-house. For investors, the distinction is academic: the orders are gone.
What POET loses: $5M+ in disclosed orders, but the bigger AI data-center pipeline is the real damage
The headline dollar loss is small. The cancelled order book exceeded $5 million against POET's $313 million in cash, cash equivalents and short-term investments at year-end 2025. POET financed itself aggressively through 2025 (total assets jumped to $328.6 million from $69.7 million) specifically to fund the AI photonics ramp. So solvency isn't the issue.
What's destroyed is harder to put on a balance sheet:
- The reference customer. POET's entire pitch since 2023 has been "we supply the AI data center photonic fabric." Celestial AI was the proof point. Without a named flagship, POET reverts to being a pre-revenue R&D platform with engineering samples.
- The 2027–2028 ramp curve. Consensus did not expect POET to reach profitability before 2028 with the Celestial-Marvell pipeline intact. Without it, break-even slips by at least a year — and only if POET signs a comparable customer in 2026, a meaningful ask after a public NDA breach.
- Counterparty trust. Every optical-I/O procurement organization — Broadcom, NVIDIA, AMD, the hyperscalers — just watched POET's CFO give away the most sensitive details of an existing supply relationship on a podcast. They remember.
- Cap-table fragility. POET has 5.79M options and 37.36M warrants outstanding — a 28.35% potential dilution overhang — plus a $135.6 million derivative warrant liability and $5.8 million of convertible debt. Refinancing those instruments against a damaged commercial story is meaningfully harder.
The right way to think about the $5M figure: it is not the loss. It is the floor on the loss. The ceiling is the entire AI-photonics narrative.
Spillover to silicon-photonics peers: who else fell, and why
Optical-comm peers — Lumentum (LITE), Applied Optoelectronics (AAOI), and Fabrinet (FN) — sold off intraday Monday but recovered most of the move by Tuesday's open. The market read this as POET-specific, not a sector revaluation: none has a Stocktwits-disclosure problem and all three have diversified customer bases. Lumentum posted 65.5% YoY revenue growth last quarter on a multi-year NVIDIA framework — different business: scale production, multi-customer, profitable. Marvell (MRVL) slipped because the cancellation costs it the integration option it paid for; our Marvell coverage remains Buy — the optical I/O strategy is intact, the execution just got messier.
Where the spillover matters is the AI-networking complex broadly. Investors who held POET as a low-priced call option on photonic fabric now have one fewer pure-play, and that capital rotates into proven names — see Credo (CRDO) for the electrical-interconnect counterpart, LRCX for upstream equipment, and Seagate (STX) for the storage side of the same AI-infra build-out — a reminder that execution discipline is the dividing line between a re-rating and a 47% gap-down.
The path forward: what POET needs to do before 2028 to survive
POET is not bankrupt. With $313 million in cash and ~$60 million annual burn, runway is ~5 years absent dilution. The question is relevance, not solvency.
Three things need to happen by mid-2027 for POET to reclaim the photonics narrative:
- Take Mika off public investor comms. A CFO transition or an announced "no media without legal review" policy is step one — without it, no procurement team signs an NDA with POET in good faith.
- Sign a new named customer with pre-cleared disclosure language. A second optical-I/O design win, structured so management can talk about it without breaching, is the only path back. There are 4+ hyperscaler-and-ASIC programs in the market — POET has to win one after the headline.
- Tighten the cap-table overhang. The 28.35% dilution potential and $135.6 million warrant liability are the second-order kill switch. A forced raise before a replacement customer prints would price punishingly. Opportunistic warrant buybacks or convertible restructuring buy time.
If those land, POET still has a shot at the 2028–2029 photonic-fabric design wins. If they don't, it becomes a takeout candidate at a fraction of pre-crash valuation — possibly to Marvell itself.
Our view: at $7.95, POET trades near cash-per-share, but the operating story has been gutted. We initiate Sell, 12-month PT $5 — loss of the only commercial reference, elevated forced-raise dilution risk, and a governance discount until disclosure controls are visibly fixed. We revisit constructively only on a named replacement customer with a contract structure that prevents a repeat. Track the live consensus and our updated price target on the POET forecast page.
Frequently asked questions
1. Why did POET Technologies stock crash 47% on April 27, 2026? POET filed a 6-K disclosing that Marvell Technology had cancelled all purchase orders POET inherited through Marvell's February 2026 acquisition of Celestial AI. The cancellation notice was dated April 23, citing breach of confidentiality obligations. The trigger was an April 21 Stocktwits interview by CFO Thomas Mika that publicly disclosed Marvell as customer, the product supplied, and the photonic-fabric application — exactly the supply-chain detail the underlying NDA prohibited.
2. What did POET CFO Thomas Mika actually say? Mika said: "We're a supplier to Marvell now that they've acquired Celestial AI… what we supply to Celestial AI are light sources — high-bandwidth, multi-frequency, high-power light sources that light up the photonic fabric." He also hinted the order size could exceed $5 million. Naming the customer, product, and application in one breath was the breach. Marvell's cancellation notice landed within 48 hours.
3. Was the NDA breach real, or was it a pretext for Marvell to exit? Both. The contractual breach is genuine — every standard semiconductor NDA prohibits the disclosures Mika made. But Marvell had been signaling for months it wanted optical I/O in-house, and the Celestial AI acquisition was about owning that stack. The interview gave Marvell a clean, low-litigation exit. For investors, the distinction doesn't matter: the orders are gone.
4. How big was the dollar loss? The cancelled order book was Mika's own "potentially exceeding $5 million" — small against POET's $313 million cash. The real loss is the only named flagship customer for the Optical Interposer platform, which materially weakens POET's position in any 2027–2028 hyperscaler design-win cycle and tightens the dilution overhang.
5. Is POET a buy at $7.95? Not in our view. The stock trades near per-share cash, but the operating story has been gutted: 20 consecutive unprofitable years, $1 million in 2025 revenue, a damaged reference customer, and a 28.35% potential dilution overhang. We rate POET Sell, 12-month PT $5. The catalyst to revisit is a named replacement customer with a properly scoped disclosure framework — not a bounce off oversold technicals.
Disclosure: For informational purposes only; not investment advice. The author and Edgen hold no positions in POET, MRVL, LITE, AAOI, or FN at publication.
Your money person, finally.
Try Ed free. No credit card. No commitment.









