Celestica builds the hardware inside the cloud giants' data centres
Network switches
boxes that pass data between computers, built to each customer's design
AI racks
servers and storage for cloud companies, its fastest-growing line
Other electronics
for aerospace, defence and industry, an older business
A cloud company that wants a switch built to its own design cannot buy one off the shelf, and does not want a factory. Celestica buys the parts, then assembles, tests and ships the finished system.
The company · 2 · What exactly does it sell?
Four fifths of sales go to cloud and computer companies
Each $1 of sales, June quarter
56%25%19%
Switches and network gearServers, AI racks and storageAerospace, defence and industry
It is paid for each system it ships. More and more it helps write the design too: that work was 41% of sales in 2025, up from 21% in 2023.
The company · 3 · How does it make money?
It keeps under nine cents of each sales dollar
8.7 centskept per $1 of cloud sales, its highest ever
41%of sales include its own design work, up from 21%
$879Mof customer bills sold early in a half-year, flattering cash
Plain assembly earns a thin fee for the work. A design Celestica helped write earns more, because the price also carries its own design content. Either way most of each sales dollar goes on parts.
Is it a good business? · 4 · What gives it an edge?
Moving a programme elsewhere restarts two clocks
10
next-generation switch programmes it is already approved on
Over 52 weeks
the wait for parts Celestica has already ordered
Because Celestica helps design the switch and orders parts more than a year ahead, a customer that moves restarts approval and loses time. Customers ask it to retake work. But each new programme is bid for again.
Is it a good business? · 5 · What is the catch?
Three customers are most of sales, and rivals share the work
63%of sales come from three customers
Hon Hai, QuantaTaiwanese builders often approved on the same programmes
A buyer of NVIDIA's systems needs no separate switch. Thailand also made 59% of 2025 sales, up from 46% in 2023, so one country carries most of the production.
Is it a good business? · 6 · What does its record show?
Sales have more than doubled since 2022
Sales, $B
7.2
8.0
9.6
12.4
18.2
Dec 2022Dec 2023Dec 2024Dec 2025last 12 mo
Sales, latest quarter+83%to $5.83B
Spare cash this year$600M → $460Mthe forecast was cut
New shares9.7%added to the share count by a $3.45B sale
Sales and profit per sale both rose. But new factories now take the cash, and in the summer Celestica sold new shares nine days after saying operations would pay for them.
Is it a good business? · 7 · How much room is there to grow?
It takes about 5% of what four customers spend
$407.8Bspent on buildings and equipment by four cloud customers in a year
$20.5Bthe sales Celestica has forecast
$40.0Bits first sales forecast for 2027, up 90%
Much of that spending is land, buildings, power and chips, which Celestica does not sell. The market it actually bids for is unmeasured.
Is it a good business? · 8 · Who is in charge?
A chief who turned an assembler into a designer
Since 2015Robert Mionis as chief executive
About 1%of the company owned by insiders
1 Octobera new finance chief, as plant spending rises
The bonus pays on sales, profit and cash, and most share awards on profit per share. Nothing pays on the return earned from the money spent on new plants.
The story right now · 9 · Why is demand growing so fast?
Cloud giants are filling new data centres with AI hardware
$407.8Ba year of spending on buildings and kit by its four largest buyers
Up 90%Celestica's first sales forecast for 2027
OpenAI and AMDnewer programmes, with no signed volume yet
Companies building AI need rooms full of computers and the switches that connect them. They design the hardware themselves and hire builders like Celestica to make it. Orders grow for as long as that spending does.
The story right now · 10 · Why does fast growth need money?
Each big order needs a factory before it brings cash
Bigger orders arriveThe 2027 sales forecast is up 90%.
Factories come first$1.5B to $2.0B of plant spending in 2027.
New shares are sold$3.45B raised, 9.7% more shares.
Cash comes laterManagement expects spare cash from 2028.
Profit is what the accounts say was earned. Free cash is what is left after paying for factories. Celestica's profit is rising while its spare cash shrinks, and that gap is the thing to watch.
The story right now · 11 · What changed lately?
Sales beat forecasts, but the cash forecast was cut
+83%September-quarter sales, to $5.83B
$40.0Bfirst sales forecast for 2027
$460Mspare cash expected this year, cut from $600M
-28.8%the shares since our review began
Mixed quarter
The doubt about growth eased. The doubt about cash did not: management called 2027 a year of investment and gave no cash figure for it.
The story right now · 12 · What is everyone arguing about?
Will all this growth ever turn into spare cash?
Many investors think
An AI winner worth far more than an ordinary contract manufacturer
We think
The growth is real, the profit is flattered, and growth now needs far more money
Part of the reported profit is a financial contract tied to Celestica's own share price, booked as lower overhead. We are wrong if spare cash turns positive on 2027 sales without heavier plant spending.
The price and our view · 13 · What is a share worth?
The price is about half our value
$527what we think a share is worth · the last close is of that
$527our value
Attractive below $421
Neutral
Unattractive above $632
$248gloomy: growth halves and profit per sale slips
$527our estimate: experts' forecasts, capped at 60% growth
$930sunny: growth runs a quarter faster
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, as a cushion for being wrong.
The price and our view · 14 · How could we be wrong?
Three customers, nothing signed, and growth that costs cash
Three customersBiggest
They make up 63% of sales.
Nothing is signedWatch
OpenAI and AMD programmes have no signed volume.
Growth eats cashShowing now
Plant spending was paid for with new shares.
Supplier billsUnknown
Whether doubled bills owed must be paid down.
Thailand made 59% of 2025 sales, so trouble in one country would reach most of the business.
The price and our view · 15 · How will we know?
January's cash figure for 2027 decides it
Spare cash of $460M or more this yearGood sign
A 2027 cash figure near $0.9B, and no new sharesGood sign
No 2027 cash figure, and more new shares soldBad sign
JAN 202727
Fourth-quarter results and the 2027 targets. Companies report every three months; each report is a test of the story.
Our view also changes on price alone: if the shares rise to within a fifth of our estimate, the cushion is gone and we move to Neutral.
The price and our view · 16 · So, what do we think?
The cloud giants' builder: fast growth that swallows cash
$527our value
Attractive below $421
Neutral
Unattractive above $632
AttractiveMedium conviction
Celestica's sales are soaring and the price sits under even our gloomy estimate. That cushion is needed. Conviction is Medium because three customers carry the business and spare cash is not expected until 2028.
↑ ↓ to move · C for contents
What it does
More about the company
Who it is
A cloud company that wants a switch built to its own design cannot buy one off the shelf, and does not want a factory. Celestica buys the chips, boards and power supplies, then assembles, tests and ships the finished system.
More and more it helps write the design too. That design-led work was 41% of sales in 2025, up from 21% in 2023. An older business, 18.9% of sales, builds electronics for aerospace, defence and industry.
How it makes money
Celestica is paid for each system it builds. On a plain assembly job it earns a fee for the work; on a design it helped write, the price also carries its own design content. Margins are thin either way: the cloud business kept 8.7 cents of each sales dollar as profit in the June quarter, the highest in its history.
Growth and profit
What growth costs
Growth has made Celestica more profitable: overhead and share pay barely rose while sales soared, so margins widened. Part of that comes from the share-price bet booked as lower overhead. The real cost of growth is now new factories, which take cash rather than margin, and that bill is rising faster than sales.
Its record
The numbers
Is it growing?Fast
Sales grew 62% over the twelve months to Sep 2026, faster than the 28% in the year to Dec 2025.
Is it profitable?Thin
Operating margin is 8.3%, up from 4.2% in the year to Dec 2023.
Does growth pay off?Pays off
Running costs take 3¢ of each sales dollar, down from 5¢ in the year to Dec 2022.
Does it earn good returns?Strong
Each dollar put into the business earns 26% a year after tax, up from 13% in the year to Dec 2022.
Is it making cash?Thin
It turned 3.0% of sales into free cash, $545M over the last twelve months.
Is the balance sheet safe?Net cash
It holds $3.96B of cash against $819M of debt, $3.14B more cash than it owes.
Are shares being diluted?Stable
The share count fell 1.5% in 3.8 years, and pay in shares costs 0.5% of sales.
Why it wins
The evidence
Because Celestica co-designs the switch and has already ordered the parts, often more than a year ahead, a customer that moves a programme restarts two clocks: approving a new supplier's design, and rejoining a year-long queue for components. Management says customers now ask it to take back work they had given to second suppliers.
The limit is that this protects a programme already running, not the next one, which is bid again.
Who could beat it
The rivals in full
The rivals it loses work to are Taiwanese builders such as Hon Hai and Quanta, often already approved as second supplier on the same programme. Jabil has built a whole division around the same buyers. And NVIDIA sells its own network inside the rack, so a buyer of NVIDIA systems needs no separate switch.
Each rival
Nvidia's NVLink and InfiniBand rack-scale interconnectsNVDA
The buyer stops purchasing a separately designed Ethernet scale-up switch for the inside of the rack, because the interconnect arrives with the accelerators as one vertical system.
Hon Hai, Quanta, Wiwynn and Accton2317.TW2382.TW6669.TW2345.TW
The buyer stops giving Celestica the build of the rack or the switch and gives it to a Taiwanese original-design manufacturer already qualified on the same generation - the second source Celestica's own chief executive says these programmes routinely carry.
The buyer stops using Celestica for the design-plus-build engagement and uses a rival contract manufacturer that has reorganised an entire reportable segment around the same artificial-intelligence infrastructure, cloud and networking demand.
The operator's own hardware teamin house
The buyer stops outsourcing the design and keeps it, using Celestica or anyone else only for assembly - the case Arista's annual report names when it lists customers who develop switches for internal use among its competitors.
How big it could get
The market and who has signed
The market
Four of its customers, Amazon, Microsoft, Alphabet and Meta, spent $407.8B on capital in their last filed years, and Celestica's guided $20.5B is about 5% of that. But much of that spending is land, buildings, power and chips, which Celestica does not sell, and no buyer publishes the split. The market it actually bids for is unmeasured.
Who runs it
Leaders, pay and the cash
Who runs it
Robert Mionis has run Celestica since 2015 and turned a low-margin assembler into a designer of cloud hardware. The finance chief of nine years moved to a sales role on 1 October; Todd Ankenmann, a finance executive there since 2018, replaced him. Insiders own about 1% of the company. The bonus pays on sales, profit margin and cash; most shares pay on earnings per share over three years, none on return on the money spent on new plants.
What they did with the money
On 27 July its filing said operations would pay for this year's plant spending. Nine days later it sold $3.45B of new shares, 16.5% below the market price, adding 9.7% to the share count. The earnings forecast of $11.30 a share was never restated for them.
The last report
The latest report
Celestica beat its forecast, guided 2027 sales to $40B, and cut its cash forecast
What it did to the thesis Mixed quarter
The growth doubt eased: sales and margin beat and the 2027 plan is far larger. The cash doubt did not: the call gave no 2027 cash figure and called 2027 a year of investment, so conviction fell to low.
The report
The cash side moved the other way: free cash for 2026 was cut from $600M to $460M and 2027 plant spending set at $1.5B to $2.0B. On the call, managers called 2027 a year of investment and expected free cash to turn positive in 2028.
What's happening now
The background
The shares have fallen 28.8% since our review began, and the 27 October call and investor day have now come. Management gave no 2027 free cash figure and called 2027 a year of investment, with positive free cash expected in 2028.
The next test is the fourth-quarter report in late January, which should bring the 2027 guidance and any free cash figure for next year.
Where we disagree
Both sides in full
The market
An AI infrastructure winner whose faster growth and rising margins in 2027 justify paying far more than for a contract manufacturer.
We believe
The growth is real. The profit it is priced on is flattered by a share-price swap, and the growth now costs far more capital.
The argument
Most analysts rate the shares a buy, and their 2027 sales figure sits 75% above this year's forecast. What troubles us sits where headlines do not reach: the swap in a note to the accounts, and the cash the growth swallows.
We would be wrong if free cash turns positive on 2027 sales with plant spending at or below $2.0B, the opposite of what management said on the 27 October call.
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 official operating profit with stock pay a cost and tax paid in full; the thesis's own evidence is the check on that path.
What the price assumes
Sales growth in the first year, then slowing as ours does, at the same profit per sale.
The price assumed
6.6%
Our estimate
60%
Measured at the close of
6 Nov 2026
The three cases
Bear case$248
Growth runs at half the analysts' path, 30%, 26% and 11%, the margin settles a fifth lower at 6.9%, and new capital earns only the 8% rate: the downturn test.
Our value$527
Revenue grows 60%, 52% and 22% over three years, the analysts' path (C56), held to at most 60% a year, then fades toward 4%, closing 30% of the gap each year; the margin moves to 8.7%, the analysts' year-three figure after stock pay.
Bull case$930
Growth runs a quarter above the analysts' path, 75%, 66% and 28%, the margin reaches 9.5%, and new capital earns 11%.
What could go wrong
Every risk
The risks
Three customers, 63% of sales
The top ten went from 64% of sales to 79% in two years, and the two new programmes deepen that rather than spread it.
Nothing is contracted
The OpenAI and AMD programmes are described as multibillion-dollar 2027 opportunities, with no volume, term or signed agreement behind them.
Growth that eats capital
Plant spending is about 6% of sales this year, against 1.5% to 2% historically, and was paid for with new shares.
One country makes most of it
Thailand made 59% of 2025 sales, up from 46% in 2023.
What we still do not know
How much of the half-year's cash came from selling unpaid customer bills early: $879M went that way, against about $10M a year earlier.
Whether the doubled bills owed to suppliers, which paid for this year's inventory build, stay or have to be paid down.
What 2027 plant spending will be. About $1.5B is indicated, and the finance chief who would set it is leaving.
What changes our view
Every route to a new view
Every route to a new view
→ Neutral
The share price rises to about $421, 80% of what we think the business is worth, so it is no longer cheap enough.
→ Unattractive
The share price rises to about $632, 20% above what we think the business is worth.
Research reopens
Once 2027 sales have landed, free cash reaches at least a quarter of adjusted earnings, with no more shares sold.
The next test
It supports the case if
2026 free cash flow at or above the $460M outlook, a 2027 free cash flow figure at or above $0.9B or a fourth-quarter adjusted operating margin above 8.4%, and no new equity.
It pressures the case if
2026 free cash flow below $460M, or no 2027 free cash flow figure, or new equity issued, or a fourth-quarter margin at or below 8.4%.
We would also be wrong if
Plant spending settles above 6% of sales with no matching rise in operating margin.
Any of the three biggest customers falls below 10% of sales for two quarters running.
Celestica sells shares again before 2027 sales reach the guided level.
The verdict
Scores and how we weighed it
The scores
Growth
3 of 4
Business quality
2 of 4
Financial position
3 of 4
Execution record
3 of 4
Per share capture
2 of 4
Potential
2 of 4
Risk profile
2 of 4
Valuation
4 of 4
Evidence quality
3 of 4
How we weighed it
ValuePositive
The price sits far under our value and below the bear case, which halves the growth and trims the margin.
BusinessNegative
A mixed business near the weak end: thin margins, growth that eats capital, and a profit line that includes a gain tied to its own share price.
GrowthMixed
The price asks for far less growth than the analysts expect, but that path counts next year's large programmes as done.
RisksNegative
Three customers are most of sales, the new programmes are not contracted, most output comes from one country, and new shares were sold to fund plant.
RivalsMixed
Its own filed rivals cost less for each dollar of profit, though they have fallen since; the gap now says little.
The groupMixed
Server and network builders move together on AI spending news; its fall with the group is the group's fact.
Latest newsMixed
The last report beat on sales and margin and the next year's plan grew, but no cash figure came with it.
In plain words
Selling new shares
A company can raise money by creating new shares and selling them. The company gets cash, but the same business is now divided into more slices, so each existing share owns a little less of it. This is called dilution.
In plain words
Free cash flow
The cash a company actually has left in a year after paying its bills and its investments. Profit can be shaped by accounting rules about when a cost is counted; cash cannot. When profit rises but free cash does not, investors ask why.
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.