Tower Semiconductor makes other companies' analog chips
Analog chips
for radio, power and sensing, made to customers' designs
Silicon photonics
chips that carry data as light between AI servers
Six fabs
in Israel, the US and Japan
Many chip companies design chips but own no factory. Tower is one of the factories they hire, specialising in chips that handle signals, power and light rather than computing.
The company · 2 · What exactly does it sell?
Light-carrying chips are now nearly half of sales
Each $1 of sales, September 2026 quarter
54%46%
Other analog chipsLight-carrying chips
The light-carrying chips go into the plug-in links between AI servers and network switches. The rest are radio, power and camera chips for phones, cars and factories.
The company · 3 · How does it make money?
In 2025 it kept 12 cents of each sales dollar
Of every $1 of sales in 2025
23.2 centsleft after the cost of making the wafers
12.4 centsleft after research, offices and sales staff too
$1.57Btotal sales in the year
Tower is paid per finished wafer. A fab's costs are mostly fixed, so with fabs partly empty in 2025 little was left over. New tools are the big cost: 28% of sales.
Is it a good business? · 4 · What gives it an edge?
Customers paid in advance to keep its factories
$341M
paid ahead by customers to reserve capacity
Through 2028
how long the filed reservations run
First
to make these chips at high yields on both wafer sizes, it says
A customer designs its light chip on Tower's process and spends months proving it in Tower's fabs, so moving means starting again. Prepaying for years of capacity shows they plan to stay.
Is it a good business? · 5 · What is the catch?
A bigger rival with US aid, and a weak record
$955Mof US chip aid GlobalFoundries is moving into the same chips
3 suitspatent cases GlobalFoundries filed in March 2026
Tower's own photonics sales already run at about half of Yole's 2030 figure, so that count misses what Tower sells. No outside measure fits, so the size of its market is not established.
Is it a good business? · 8 · Who is in charge?
The same leaders for two decades, spending on factories
Since 2005Russell Ellwanger, chief executive
0.50%of the company owned by insiders
$920Mof new tools to expand photonics
Tower pays no dividend and buys back no shares; the cash goes into tools. The chief executive sold shares many times from July to September 2026, none under a pre-set plan.
The story right now · 9 · Why do AI servers need Tower's chips?
AI servers talk over light, and light needs these chips
$240Mphotonics sales in the September 2026 quarter
Over $1.1Byearly pace Tower targets by December
5xits 2025 photonics sales
AI training links thousands of chips that must swap data fast. Over the distances inside a data centre that data travels as light, and each link needs chips that turn electrical signals into light and back.
The story right now · 10 · Why can profit rise faster than sales?
Each extra wafer in a full fab is mostly profit
Fab partly empty12 cents per $1 in 2025.
Costs stay fixedTools and staff cost the same.
Orders fill it85% to 90% used.
Profit jumps24 cents per $1 by September 2026.
So we value Tower on about 30 cents of each sales dollar as operating profit, what fuller fabs should earn, below the 38 cents management targets.
The story right now · 11 · What changed lately?
Sales up 33% in a year, and the shares halved
$526Msales in the September 2026 quarter
24 centsoperating profit per $1, up from 13
$600Mexpected for the December quarter
−58%share fall from the June high
Strong quarter
Twice since July the shares fell more than a fifth in a week, the second time after this record quarter, on fears that a funded rival and an AI spending pause end the boom.
The story right now · 12 · What is everyone arguing about?
A boom about to fade, or two years already paid for?
Many investors think
An AI optics winner whose boom may peak soon, with a funded rival arriving
We think
Capacity is prepaid through 2028 and profit is rising as the fabs fill
We would be wrong if operating profit fell below 20 cents of each sales dollar in a quarter.
The price and our view · 13 · What is a share worth?
The price sits about a quarter below our value
$173what we think a share is worth · the last close is of that
$173our value
Attractive below $139
Neutral
Unattractive above $208
$63.12gloomy: half the expected growth, less kept per $1
$173our estimate: experts' path, about 30 cents kept per $1
$283sunny: growth beats the forecast by a quarter
Value per share in three futures. None is a prediction of the share price.
We estimate what a share is worth from the cash the business should earn over many years. Our rule: buy only when the price is at least a fifth below our estimate, and Tower is just inside it.
The price and our view · 14 · How could we be wrong?
A funded rival is the biggest risk
GlobalFoundriesBiggest
US aid, same chips.
Patent suitsShowing now
Could bar its wafers.
An AI pauseWatch
Fabs empty fast.
Who the buyers areUnknown
Tower names none.
Our value also needs profit per dollar above any full year Tower has filed, and management is the main witness that it will get there.
The price and our view · 15 · How will we know?
February tests the December quarter
Sales of $570M or moreGood sign
26 cents or more of profit per $1Good sign
A March quarter forecast below DecemberBad sign
FEB 202710
Fourth-quarter 2026 results, around 10 February. Companies report every three months; each report is a test of the story.
Our view also changes on price alone: a fifth below our value we stay positive, nearer to it we turn neutral, a fifth above it negative.
The price and our view · 16 · So, what do we think?
A prepaid AI optics ramp, priced just cheap enough
$173our value
Attractive below $139
Neutral
Unattractive above $208
AttractiveMedium conviction
Customers have paid ahead for Tower's factories and profit is rising as they fill. The price sits about a quarter below our value of about $173 a share, just inside our safety rule, so we rate it Attractive.
↑ ↓ to move · C for contents
What it does
More about the company
Who it is
Many chip companies design chips but own no factory. Tower is one of the factories they hire. It specialises in analog chips, which handle radio signals, power and light rather than computing, and most of all now silicon photonics for data-centre links.
It runs fabs in Israel, in Newport Beach and San Antonio in the US, and in Japan. Silicon photonics puts the optical parts of a data link onto one silicon chip, made with ordinary chipmaking tools.
How it makes money
Tower is paid per finished wafer, and more for harder processes. In 2025 sales were $1,566M and it kept about 12 cents of each sales dollar as operating profit, with its fabs partly empty.
Most of a fab's costs are fixed, so each extra wafer adds far more profit than the average one, and the margin rises fast as photonics fills the fabs. New tools are the big cost: 28% of sales in 2025.
Growth and profit
What growth costs
Growth has made Tower more profitable: research and overheads grew more slowly than sales, so operating profit rose from 10 cents per sales dollar in the first half of 2025 to 24 cents in the third quarter of 2026. What grows with sales is factories: plant spending rose from 21% of sales in 2021 to 28% in 2025.
Its record
The numbers
Is it growing?Steady
Sales grew 9.1% in the year to Dec 2025, faster than the 0.9% in the year before.
Is it profitable?Healthy
Operating margin is 12%, down from 19% in the year to Dec 2022.
Does growth pay off?Pays off
Running costs take 11¢ of each sales dollar, about where it was (11¢) in the year to Dec 2021.
Does it earn good returns?Low
Each dollar put into the business earns 6.0% a year after tax, about where it was (9.2%) in the year to Dec 2021.
How long can it keep going?Long runway
It used $48.9M of cash in the latest year, so the $1.15B it holds lasts about 23.5 years at that pace.
Is the balance sheet safe?Net cash
It holds $1.15B of cash against $162M of debt, $990M more cash than it owes.
Are shares being diluted?Stable
The share count rose 3.5% in 4.0 years, and pay in shares costs 2.4% of sales.
Why it wins
The evidence
A customer designs its optical chip on Tower's process and then spends months proving it works in Tower's fabs, so moving it to another factory means redesigning and re-proving it. Customers have gone further and prepaid to reserve years of Tower's capacity, money they get back only as later purchases.
Tower says it was first to make these chips at high yields on both wafer sizes. The research firm Yole expects the market for silicon photonic chips to grow about 31% a year to 2030.
Who could beat it
The rivals in full
GlobalFoundries, a larger contract chipmaker, is moving $955M of its US government chip aid into silicon photonics, and calls optical networking a growth driver. It also filed patent suits against Tower in March 2026. TSMC, the largest chipmaker, offers photonics too. Specialty rivals such as X-Fab, Vanguard and Hua Hong compete for Tower's older lines.
Each rival
GlobalFoundries silicon photonics and SiGeGFS
a transceiver maker designs its next photonics chip on GlobalFoundries' process instead of Tower's
TSMC's photonics and packaging for co-packaged opticsTSM
chip and switch makers put optics inside TSMC's own packages, bypassing the pluggable transceiver
GlobalFoundries RF, plus Vanguard, DB HiTek, X-Fab and Hua HongGFS5347.TWO000990.KSXFAB.PA1347.HK
a chip designer moves an RF, power or sensor product to another specialty foundry
How big it could get
The market and who has signed
The market
Yole puts the silicon photonic chip market at $404M in 2024, rising to more than $2.1B in 2030.
That count is too small to size Tower's market: Tower's own photonics sales are already running at about half of Yole's 2030 figure. Nobody publishes a measure that fits, so the size of the prize is not established.
Who runs it
Leaders, pay and the cash
Who runs it
Russell Ellwanger has been chief executive since 2005, after years at Applied Materials, and Oren Shirazi finance chief since 2004. Directors and managers own 0.5% of the shares. Most of the chief executive's share award pays on profit. Between July and September 2026 he sold shares many times at $215 to $250, none under a pre-set plan, and no insider bought.
What they did with the money
Tower puts its cash into new tools rather than paying it out: no dividend and no buybacks. In the nine months to September 2026 it spent $534M on plant, mostly for photonics, while operations brought in $872M, helped by customers' prepayments. It is spending about $920M to roughly triple photonics capacity by late 2027.
The last report
The latest report
Tower's quarterly sales rose 33% to $526M as profit margin nearly doubled
What it did to the thesis Strong quarter
It showed the photonics ramp real and the margin rising as the fabs fill, the thing our value most needs. It did not show the margin reaching about 30 cents per sales dollar, which only the next quarters can.
The report
In the three months to September 2026, sales rose 33% on a year earlier to $526.2M and 14% on the quarter before. Operating profit reached 24.2% of sales, against 12.8% a year earlier, as fuller fabs spread their fixed costs. Photonics brought in about $240M.
Management guided about $600M for the December quarter, which would make 2026 sales about $2.0B, and said it expects no dip in early 2027. The shares fell over the following days anyway.
What's happening now
The background
The next test is the December quarter's results, due around 10 February 2027.
Where we disagree
Both sides in full
The market
An AI optics winner whose boom may peak soon, with a better-funded rival arriving, so worth selling on any wobble.
We believe
The capacity is prepaid through 2028 and the margin is rising as the fabs fill; at the price, the next two years are underpriced.
The argument
Analysts are nearly all positive, yet the shares have more than halved: the market is pricing how long the AI optics boom lasts and whether GlobalFoundries takes share. We think the prepaid capacity makes the next two years more certain than the price assumes.
Against us: our value needs a margin above any year Tower has filed, and management's own target is the main witness for it.
What it's worth
How we value it
Valued on the analysts' revenue path for three years, each later year closing 30% of the gap to 4%, on GAAP operating profit with stock pay a cost, the margin rising to the analysts' 2028 figure on the filed basis, and tax at the 15% minimum that now applies in Israel.
What the price assumes
Sales growth in the first year, then slowing as ours does, at the same profit per sale.
The price assumed
37.7%
Our estimate
28%
Measured at the close of
16 Nov 2026
The three cases
Bear case$63.12
Growth runs at half the analysts' path, 13.9%, 19.9% and 16.6%, as the AI optics boom pauses or a rival takes share, the margin settles a fifth lower at 23.7%, and new capital earns only the 8% rate: the downturn test.
Our value$173
Revenue follows the analysts' path (C29), 27.8%, 39.8% and 33.2%, matching the $2.0B guided for 2026 and management's capacity plan, then fades toward 4%, closing 30% of the gap each year; the margin rises from 12.4% to 29.6%, the analysts' 2028 figure less 2.4% stock pay and a 2.7-point basis gap (C35); tax is 15% under Pillar Two (C24); new capital earns 12%.
Bull case$283
Growth runs a quarter above the analysts' path, 34.8%, 49.8% and 41.5%, as near-package optics and Track-1 add to the pluggable ramp, the margin reaches 32.6%, still below management's 38% model, and new capital earns 15%.
What could go wrong
Every risk
The risks
A funded rival
GlobalFoundries is putting $955M of US aid into the same chips and could win Tower's customers' next designs.
Patent suits
GlobalFoundries is suing at the US trade commission, which can bar imports of wafers it finds infringing.
The cycle
Sales fell 15% in 2023 and were flat until 2025; an AI spending pause would empty fabs fast.
Israel
Its main fab and head office are in Israel, exposed to conflict there.
What we still do not know
Who the photonics customers are: Tower names none, so how concentrated that line is cannot be checked.
How much of the contracts into 2029 is binding; the filed part runs through 2028.
How big Tower's market is: no outside count fits what it sells.
What changes our view
Every route to a new view
Every route to a new view
→ Neutral
The share price rises to about 80% of what we think the business is worth.
→ Unattractive
The share price rises to about 120% of what we think the business is worth.
→ Neutral
Operating profit falls below 20 cents of each sales dollar in a quarter.
The next test
It supports the case if
Revenue of $570M or more, the bottom of the guide, an operating margin of 26% or more, and a first-quarter 2027 guide above $600M.
It pressures the case if
Revenue below $570M, an operating margin below 26%, or a first-quarter guide below the fourth quarter.
We would also be wrong if
Quarterly photonics sales fall below the third quarter of 2026's level during 2027.
A top photonics customer moves a high-volume part to GlobalFoundries or TSMC.
The trade commission or a court bars Tower's wafers in the patent cases.
The verdict
Scores and how we weighed it
The scores
Growth
3 of 4
Business quality
3 of 4
Financial position
4 of 4
Execution record
3 of 4
Per share capture
3 of 4
Potential
3 of 4
Risk profile
2 of 4
Valuation
3 of 4
Evidence quality
3 of 4
How we weighed it
ValuePositive
The price sits just under the line where the margin of safety starts, a thin cushion with the bear case far below.
BusinessPositive
Customers prepaid to reserve its photonics capacity for years, and it was first to make these chips at volume.
GrowthPositive
Sales grow fast on capacity already bought and contracted, and the price asks for less growth than analysts expect.
RisksNegative
Its main fab is in Israel, the cycle has turned before, a few unnamed customers carry the photonics line, and the boss sold shares.
RivalsMixed
GlobalFoundries is pouring government money into the same chips and is suing over patents.
The groupMixed
AI optics names fell together on fears the spending boom is peaking; the fall was the rating, not the sales.
Latest newsPositive
The latest report was a record with a higher guide, and the shares fell anyway.
In plain words
Our estimate of value
We forecast how much cash the business can earn over the next ten years and beyond, then ask what that stream is worth today to an investor who wants an 8% yearly return. We do it three times: a gloomy future, an expected one and a sunny one. We call a share Attractive only when its price is at least 20% below the expected estimate, and Unattractive when it is 20% above.