POET offers a cheaper way to build optical engines
Optical engines
turn data into light for AI links
Built by machine
on a wafer, not by hand
81 patents
49 on its method
Optical engines send data as light between AI computers. POET places lasers, light detectors and electronics on a wafer by machine, without the slow hand alignment most makers use.
The company · 2 · What exactly does it sell?
All of today's revenue is fees and samples
Each $1 of sales, latest year
100%engineering fees and samples
About $1.07Mrevenue in 2025
POET has no product sales yet; customers pay it to develop and test engines.
The company · 3 · How does it make money?
It sells almost nothing and lives on its cash
$796.3Mcash and deposits
$11.3Mloss last quarter
$43.2Mrunning costs in 2025, from $20.9M
POET plans to sell engines to transceiver makers by the thousand. Until then it spends on staff and equipment from the $400M it raised in May.
Is it a good business? · 4 · What gives it an edge?
A cheaper method, not yet proven by buyers
About 60%
possible cost cut from machine assembly
81
patents
Celestial AI committed to production, then Marvell bought it and cancelled every order.
Is it a good business? · 5 · What is the catch?
Giant transceiver makers grow with their own designs
InnoLightthe largest transceiver maker
Lumentum, Coherentlasers and their own light chips
A newcomer may arrive just as the market slows: it has swung between booms and flat years.
Is it a good business? · 6 · What does its record show?
Spending doubled while revenue stayed near $1M
Costs$43.2Mrunning costs, 2025, from $20.9M in 2023
RevenueAbout $1.07Mrevenue, 2025, from $0.47M
BuildingAbout $50Mequipment planned this half
Staff rose past 115 with about 50 more planned, so costs run far ahead of sales.
Is it a good business? · 7 · How much room is there to grow?
A $26B market where it has almost nothing
$26BAI optical links market, 2026
$50MLumilens first order, conditional
$500Mover five years, POET says
None of these orders is binding beyond firm purchase orders.
Is it a good business? · 8 · Who is in charge?
A long-serving chief; insiders own almost nothing
Since 2016Suresh Venkatesan in charge
Under 0.1%owned by directors and officers
$7.3Mpaid to key managers in 2025
A new finance chief from Silicon Labs started on 2 November. No set rule fixes bonuses.
The story right now · 9 · Why do AI data centres need optical engines?
AI computers talk through light, by the million
$26Bspent on AI light links this year
1 millionengines a month POET plans by end-2027
AI data centres join thousands of chips with light in glass fibre, and each link needs an optical engine. Making them by hand is slow; a cheaper method could win big orders.
The story right now · 10 · What is POET worth before it sells?
Today the shares are worth about its cash
Cash in the bank$796.3M.
Costs burn it$11.3M a quarter, rising.
A customer must qualifyLumilens samples due.
Then repeat ordersNone have stayed so far.
The cash puts a floor under the value for some years; orders that last would add to it.
The story right now · 11 · What changed lately?
Revenue passed $1M; the engine ramp slipped
$1.03Mquarterly revenue, up 244%
$7.6Mquarterly loss
$774.3Mcash and investments
Mid-202730,000 engines meant for 2026
Mixed quarter
Shortages of wafers and parts and a yield ramp held output back. Lumilens's first products are on schedule for checks this year.
The story right now · 12 · What is everyone arguing about?
Will a customer qualify and keep buying?
Many investors think
The cash will leak away before the engines sell
We think
The cash covers the downside for years
Short sellers hold about a seventh of the shares. The open question is whether Lumilens qualifies and keeps buying.
The price and our view · 13 · What is a share worth?
No single value here: it is priced as a bet
2.5×if it works
74%kept if it fails
Multiples of the company's value when the bet was last struck, 16 Nov 2026.
We put no single value on a share here, because the business does not yet earn a lasting profit. Instead we ask what the shares pay if the plan works, and what is left if it fails.
The price and our view · 14 · How could we be wrong?
Vanishing orders, a tax trap and more shares
Orders can vanishBiggest
As the April cancellation showed.
A US tax trapWatch
It may be taxed as a passive investor.
More sharesShowing now
The count more than doubled in 18 months.
The $30M loanUnknown
Who borrowed it, and why.
A class action alleges misstatements about the tax risk and customer relationships.
The price and our view · 15 · How will we know?
Product sales and Lumilens samples decide it
Fourth-quarter shipments beat the first nine months'Good sign
Product sales of five million dollars a quarterWe turn positive
Cash below $500M, revenue still under $2MWe turn negative
MAR 20271
Fourth-quarter results and first earnings call Companies report every three months; each report is a test of the story.
Our view moves on these business events, not on a price line, since we put no single value on a share here.
The price and our view · 16 · So, what do we think?
A cheaper optical engine, still without a lasting customer
Speculationthe kind of bet this is
NeutralLow conviction
POET's way of building optical engines could be cheaper, but it sells almost nothing, its one production customer walked away in April, and its next big order waits on samples. The shares are worth about its cash. Neutral.
↑ ↓ to move · C for contents
What it does
More about the company
Who it is
POET designs optical engines and light sources. Its trick is the Optical Interposer: a wafer on which lasers, light detectors and electronics are placed and wired by machine, without the slow hand alignment most makers use.
Engineers working on an industry standard estimated this kind of passive assembly could cut a transceiver's cost by about 60%. POET holds 81 patents, 49 of them on the interposer, and says its 800G receive engine has been designed into several customers' modules.
How it makes money
Today it barely does. Revenue was about $1.07M in 2025 and in the first half of 2026, all engineering fees and sample sales. The plan is to sell engines to module makers by purchase order, and those orders can be cut or delayed without much penalty.
Growth and profit
What growth costs
So far growth costs far more than it brings in. Operating expenses doubled to $43.2M over three years while revenue stayed near $1M, staff rose past 115 with about 50 more planned, and management plans about $50M of equipment this half.
Why it wins
The evidence
If it works, the edge is cost and power: an engine assembled by machine on a wafer should be cheaper to build at volume than one aligned by hand, and the patents cover the method.
That edge is not yet proven by repeat buyers. The customer that had committed to production, Celestial AI, was bought by Marvell, which cancelled every order in April citing a breach of confidentiality. Lumilens is now the case's main test.
Who could beat it
The rivals in full
The biggest transceiver makers, InnoLight above all, are growing fast with their own designs, and Lumentum and Coherent sell lasers, packaging and silicon photonics into the same links. The market also runs in booms followed by flat years, so a newcomer may arrive at a slowdown.
Each rival
InnoLight and Eoptolink, the largest transceiver makers300308.SZ300502.SZ
Building their own engines with aligned assembly, which already ships in volume and is scaling fast.
Coherent and Broadcom silicon photonics and co-packaged opticsCOHRAVGO
The engine itself, made on silicon by suppliers with fabs, customers and balance sheets far larger than POET's.
Lumentum and Coherent high-power laser sourcesLITECOHR
POET's external light sources for co-packaged optics.
How big it could get
The market and who has signed
The market
The market is big. LightCounting puts sales of AI cluster transceivers and co-packaged optics at $16.5B in 2025 and $26B in 2026, with up to six transceivers per GPU.
POET's revenue was about 0.007% of that. How much it could win depends on qualifications it has not passed yet, so nobody can size its share honestly today.
Who runs it
Leaders, pay and the cash
Who runs it
Suresh Venkatesan has run POET since 2016 and chairs the board. Sandeep Kumar became operating chief in May, and Giovanni Pacelli, from Silicon Labs, took over as finance chief on 2 November.
Directors and officers together own under 0.1% of the company. Key managers were paid $7.3M in 2025, about half in shares, and the annual report says no set rule fixes bonuses.
What they did with the money
Most of the money raised is still in the bank. Beyond that, POET paid a financial adviser $9.5M in the first half, lent $30M to an unnamed company and bought a small stake in a module partner. None of these has been explained in detail.
The last report
The latest report
Revenue passed $1M for the first time; the engine ramp slipped to mid-2027
What it did to the thesis Mixed quarter
It cleared the bars we set, but moved nothing that changes the view: sales are still a fifth of the level that would, Lumilens has not qualified, and the volume ramp slipped half a year.
The report
In the third quarter of 2026 POET booked $1.03M of revenue, up 244% on a year earlier and its first quarter above $1M. It lost $7.6M, helped by a $6.4M paper gain on warrants, and used $12.7M of cash in operations.
Cash and investments were $774.3M. The 30,000 engines it had aimed to ship in 2026 now ship by mid-2027, blamed on wafer and part shortages and yields, while a new production line worth over $45M is built. Lumilens's first products are on schedule.
What's happening now
The background
The next tests are close: Lumilens samples by the end of 2026, fourth-quarter shipments promised to beat the first nine months', and the first earnings call with the year's results on 1 March 2027.
Where we disagree
Both sides in full
The market
The cash will leak away on advisers, loans and spending before the engines ever sell in volume.
We believe
The cash covers the downside for years; the open question is only whether Lumilens qualifies and keeps buying.
The argument
Short sellers hold about a seventh of the shares, and the price says the market expects much of the cash to be spent without a business to show for it. The two analysts covering it rate it a buy.
We sit between them. The cash puts a floor under the value for some years, but the engines have not yet won a production customer that stayed.
What it's worth
How we value it
No fair value: the analysts' forecasts show no lasting operating profit after stock pay, so a value built on earnings would rest on a guess; the bet prices it instead.
The bet
This is a bet on POET becoming a real supplier. If Lumilens qualifies the engines and its first order ships, sales of a modest size would value the business at more than twice today's price.
If the engines never sell, holders keep most of their money, because the cash left after a few years of spending is still most of today's market value.
The bet
Seen as a bet
This is a bet on POET becoming a real supplier. If Lumilens qualifies the engines and its first order ships, sales of a modest size would value the business at more than twice today's price.
If the engines never sell, holders keep most of their money, because the cash left after a few years of spending is still most of today's market value.
What could go wrong
Every risk
The risks
Orders can vanish
Customers buy by purchase order and can cancel or delay without much penalty, as the April cancellation showed.
A US tax trap
Interest income is most of its gross income, so POET may be a passive foreign investment company, which taxes US holders harshly.
A lawsuit
A class action alleges misstatements about that tax risk and its customer relationships in April 2026.
More shares
The share count more than doubled in eighteen months, and 78.7M warrants could add more.
Unexplained spending
Large adviser fees and a $30M loan to an unnamed company draw on cash that belongs to holders.
What we still do not know
Lumilens publishes no accounts, so whether it can pay for $500M of engines, and who its customers are, cannot be checked.
Who borrowed the $30M, and why, has not been disclosed.
Whether the plan for up to 1 million units a month by end-2027 has buyers behind it.
What changes our view
Every route to a new view
Every route to a new view
→ Attractive
Quarterly revenue reaches $5M from product sales, not engineering fees. Lumilens's engines pass qualification and production orders follow.
→ Unattractive
Cash and deposits fall below $500M with quarterly revenue still under $2M.
The next test
It supports the case if
Q4 revenue above $2M, Lumilens verification completed in 2026, and cash and investments above $700M.
It pressures the case if
Q4 revenue below $1.2M, Lumilens verification slipped into 2027, or cash below $680M or spent on loans or acquisitions.
The verdict
Scores and how we weighed it
The scores
Growthnot scored
Business quality
2 of 4
Financial position
4 of 4
Execution record
2 of 4
Per share capture
2 of 4
Potential
2 of 4
Risk profile
2 of 4
Valuation
3 of 4
Evidence quality
2 of 4
How we weighed it
ValueMixed
No fair value is drawn: the forecasts show no lasting profit after stock pay, so the bet prices it.
BusinessNegative
A mixed business near the weak end: almost no sales and its production customer left.
GrowthMixed
Growth cannot be measured until a customer qualifies the samples.
RisksNegative
Orders can be cancelled, a US tax trap may apply, a class action is open and more shares may come.
RivalsMixed
No filed rival compares directly, so the price is judged on its own.
The groupMixed
Optical names move together on AI spending news.
Latest newsMixed
The latest report cleared revenue and cash and kept Lumilens on schedule, but the engine ramp slipped half a year on shortages and yields.