Inside POET's Optical Interposer: A PIC Springboard for Tier 2 Optical Module Makers — Can It Last Until CPO Volume?

2 days ago

8 min read

The most dramatic single company in optical communications over the past two months is a small Canadian PIC designer — POET Technologies (Nasdaq: POET).

The storyline is packed: in late April, Marvell (which had just acquired Celestial AI) canceled all of its orders, citing the CFO's "breach of confidentiality obligations on Stocktwits"; the stock fell 46% in a single day and a securities class action followed. In mid-May, POET signed a $50M PO plus a five-year, $500M collaboration framework with Lumilens, a startup founded in 2024, and the stock rebounded 43%. Revenue disclosed on the Q1 2026 earnings call was $503,389 — not a typo: half a million, not fifty million.

With numbers this small and swings this large, is it a price bubble, or is there a real technical position that can carry it some distance? This piece breaks down three things: where POET's Optical Interposer really sits on the 2026 optical map, what problem it solves that others can't, and what its biggest ceiling and risks are. Conclusion first: the technology direction is right, but commercial scale lags far behind the narrative, and its real ceiling isn't the Marvell cancellation — it's that the 3D photonic interposer route will press directly down on it after 2028.

1. Why POET Is Only Getting Attention Now

POET isn't a new company; its technical papers go back at least to an Optical Interposer presentation at ECOC 2023. It has suddenly drawn attention in 2026 because the optical industry is stuck in an awkward middle ground.

Looking up, CPO (co-packaged optics) is not yet truly in volume production. Lightmatter's Passage M1000, revealed at HOT CHIPS 2025, is about "the next physical frontier," not near-term shipments; Marvell's acquisition of Celestial AI is also a post-2028 story. Looking down, the EML shortage is already the dominant theme of 2026 — Mitsubishi's annual capacity is only about 50 million units, Coherent and Momentum together can't fill the whole gap, a 1.6T optical module needs 8 EMLs, and the whole industry is queuing for every unit it can get.

That leaves a gap no one else has filled — Tier 2 optical module makers (Foxconn FIT, Luxshare, LITEON) want to catch up with tier-1 players like Innolight and Eoptolink, but they don't do pure SiPh themselves and can't wait for CPO. What they need is something "usable right now, able to bypass upstream bottlenecks, and requiring no in-house investment in a PIC design team."

That is exactly where POET stands. It reframes the problem from "I need to integrate EML, SiPh and PD onto one chip" to "I'll place these materials onto one platform for you and ship them in batches for your modules" — essentially shifting the integration burden from the module maker to the PIC designer.

2. What the Optical Interposer Is: A Hybrid Layer "Placed" onto a CMOS Platform

To understand POET's position, first look at what pure SiPh players can't do.

The advantage of pure SiPh (silicon photonics) is that it builds waveguides, modulators and PDs with standard semiconductor processes, leveraging the entire silicon wafer ecosystem. But silicon doesn't emit light — the laser source for a SiPh module must be attached externally (an external DFB or EML block), and that attachment step requires active alignment and wire bonding, making back-end packaging costly and yields unstable. Tower Semiconductor's own Q1 2026 earnings call named the modulator route battle, laser integration approaches and CPO PIC design as the main arenas for the next three years.

POET doesn't take the pure SiPh route. Its Optical Interposer is essentially a "multi-layer, CMOS-fabricated electro-optical hybrid interposer":

Bottom layer: high-speed metal traces, just like a standard electronic interposer
Top layer: waveguides + passive optical components + a precision-aligned placement area for active devices
Active devices: InP lasers, SiPh modulators and PDs, "placed" on via hybrid integration, rather than grown monolithically on silicon

The key word is "placed" — instead of competing on pure SiPh integration density, POET uses each material for its strengths: InP emits light (unbeatable performance), SiPh modulates (lowest cost) and the PD receives. The interposer then connects their optical paths and routes the signals out.

POET isn't selling "the most advanced PIC" — it's selling "the PIC with the fewest assembly steps."

The numbers in its technical documents are bold: eliminating wire bonds, eliminating active alignment, enabling wafer-level mass production. If it truly delivers, what it means for Tier 2 module makers is: no need to maintain a SiPh design team or build back-end packaging lines — just buy POET's PICs and assemble 800G/1.6T transceivers.

3. Mitsubishi 2×200G EML: Cutting 1.6T from 8 EMLs to 4

The one card POET can really play in 2026 — and that the industry recognizes — is its partnership with Mitsubishi Electric.

In the BOM of a standard 1.6T optical module, the EML is the most expensive single component — one 1.6T transceiver needs 8 EMLs, which alone eat up 15-20% of the BOM. In the 2026 EML shortage, those 8 EMLs aren't just a cost issue; they're a question of whether you can get supply at all.

The solution POET developed with Mitsubishi is called Teralight — using Mitsubishi's exclusive 2×200G EML (one EML outputs two 200G channels) to cut the EMLs a 1.6T transceiver needs from 8 to 4. Cut in half.

The brilliance of this solution isn't a technical breakthrough — it's the timing. For the same 1.6T and the same reach spec, the POET + Mitsubishi combination tells module makers: "You don't have to queue for 8 EMLs, just 4; your per-module cost comes down, and the margin goes to you." For Tier 2 module makers short on EMLs yet eager to win 1.6T orders, that's a direct incentive.

The question is whether Mitsubishi's own 2×200G EML production schedule can keep up. That's the biggest uncertainty in this storyline — POET can push its PICs, but the light-emitting component isn't in its hands.

4. Competitive Positioning: Three Comparisons to Put POET in the Right Place

Many in optical communications put POET on the same map as Lightmatter and Celestial AI. That's wrong. The three routes tell three different stories:

A. POET vs. the 3D photonic interposer route (Lightmatter Passage, Celestial AI, Ayar Labs)

Lightmatter is 3D — the compute die is bonded face-down onto the photonic layer, optical TX/RX is the core of the package structure, and the target customers are ASIC/GPU makers. POET is 2.5D — the electro-optical hybrid interposer is still planar, active devices are placed via hybrid integration, and the target customers are optical module makers. One is chasing the package-layer position next to the ASIC; the other is chasing the PIC position inside the optical module — different races.

This means POET's ceiling is capped. When CPO truly enters volume production (2028 at the earliest), POET's 2.5D hybrid will be a mid-tier solution, not the top tier.

B. POET vs. pure SiPh players (Marvell SiPh, Cisco SiPh, Rockley)

POET's integration is looser than pure SiPh — active devices are "placed" rather than monolithic — but its material choices are more flexible. It claims to eliminate wire bonds and active alignment and to cut costs with wafer-level packaging, but there is currently no independent third-party data to back this up; it is purely the company's own claim. This is the gap between technical papers and production BOMs; outsiders need to see actual yield numbers from module makers running POET PICs to verify it.

C. POET vs. tier-1 module makers' in-house PICs (Innolight, Eoptolink, Coherent, Lumentum)

Tier-1 module makers do their own PICs plus module packaging; POET "sells PICs to module makers." For a tier-1 player like Innolight with 23% market share, POET is entirely unnecessary — their own PICs run at far higher volume than POET's, with a deeper learning curve. POET's real market is Tier 2 makers such as Foxconn FIT, Luxshare and LITEON that want to catch up with tier 1 but lack PIC design capability.

Simply put: POET isn't on the front line of optical communications' must-watch list — it's a "technology springboard for Tier 2 module makers catching up with Tier 1".

5. Reality vs. Narrative: -46% Then +43% in Two Months

Next comes corporate governance. Two events at POET in April–May 2026 fully exposed the gap between the company's narrative scale and its commercial scale.

Event 1: Marvell cancels orders (2026/04/23). After acquiring Celestial AI, Marvell notified POET in writing that it was canceling all POs because "POET breached confidentiality obligations regarding purchase orders and shipment information" — more specifically, this involved POET's CFO publicly responding on Stocktwits to comments related to order details. The stock fell 46% in a single day, and a securities class action followed.

That's a negative reference for every future customer. What customers in optical communications — especially hyperscalers, ASIC makers and tier-1 module makers — are most sensitive to in a supplier isn't technology or price, it's "will you use my order details for marketing." The signal the Marvell cancellation sent to the whole industry: sign orders with POET and be prepared for leaks.

Event 2: Lumilens signs a $50M PO + five-year $500M framework (2026/05/14). Lumilens is a San Francisco startup founded only in 2024, backed by Mayfield and Spark Capital, building scale-up/scale-out optical interconnect systems. POET simultaneously granted Lumilens warrants for 22,921,408 shares (exercise price $8.25, vesting in tranches based on cumulative purchase amounts).

Do the math — at the current share price, the warrants POET granted are worth roughly the same order of magnitude as the $50M PO. This looks more like a "mutual capital endorsement + startup locking in a supplier" arrangement than a pure commercial design win. Discounting the Lumilens order, POET's truly arm's-length disclosed production orders are still around the $5M level, leaving a big gap to fill before reaching the "30,000 optical engines shipped annually" narrative.

Back to the financials for its real position: Q1 2026 revenue of $503K, net loss of $12.3M, cash of $429M. The cash position can fund 3-5 years of burn — POET has no immediate survival issue. But "able to survive" and "able to ramp" are two different things.

6. Industry View: Two Signals Will Decide POET's Next 18 Months

Placed on the 2026-2030 optical communications timeline, POET sits between two routes, ZR and Coherent Lite — the former is happening now, the latter is a story that won't take off until 2028. We broke down this timeline in full in ZR is the meat, Coherent Lite is the bone: the next decade's battle in depth for the optical transceiver market. What POET wants is the 800G/1.6T pluggable window of 2026-2028 — a window that genuinely exists, catalyzed by the EML shortage, with real demand from Tier 2 module makers.

The question is whether it can win enough design wins before the window closes to lift revenue from the current "millions of dollars" level to "tens of millions." Two signals decide everything:

The first signal: any of Foxconn FIT, Luxshare or LITEON publicly disclosing which optical module model uses a POET PIC. This matters more than any PO amount disclosure — a module maker listing the PIC source on a model's spec sheet is ironclad proof of a design win. That information hasn't appeared yet.

The second signal: Mitsubishi delivering on its 2×200G EML production schedule. POET's 1.6T story is built on that Mitsubishi EML; if it doesn't reach volume production on time, POET's Teralight is just a demo.

If neither signal materializes within 18 months, POET will face the squeeze of true CPO volume production — once 3D photonic interposers arrive, module makers' demand for 2.5D hybrid PICs will be diluted quickly. POET's window will close faster than expected.

If both signals materialize, POET can last until 2028-2029 as the standard PIC supplier to Tier 2 module makers before CPO, earning money from this transition period. But even that isn't a "next optical giant" story — it's a "window-period supplier" story.

POET's technology direction is right — hybrid integration genuinely has a place in 2026. But the company's biggest problem right now isn't technology, it's scale. Annualized revenue of $2M supporting a $500M framework narrative leaves a gap that must be filled by real optical module shipments, not by PR and capital-markets maneuvers. The Marvell cancellation has already reminded every future customer of that.

POET should be classified as a "PIC enabler for Tier 2 optical module makers — on watch". Understanding POET means understanding how Foxconn FIT, Luxshare and LITEON are trying to catch Innolight — POET itself isn't a front-line must-watch, but what it represents, "the window for hybrid integration before CPO volume production," is a real piece of territory in the 2026-2028 optical industry.

As for whether its narrative scale matches its commercial scale — that's for POET to prove over the next 18 months. STT will watch from the sidelines, log the tracking signals, and issue no investment rating.

This article is for technology and industry trend analysis only and does not constitute investment advice.