Sanmina builds other companies' electronics in its own factories
AI racks
whole rack-sized computers, from boards to cooling
Network gear
for telecom and cloud companies
Medical and defence
electronics in certified factories
Sanmina makes nothing under its own name. It buys the parts, assembles circuit boards, fits them into a rack with power and liquid cooling, tests it at full load and ships it.
The company · 2 · What exactly does it sell?
Networks, cloud and AI are two thirds of sales
Each $1 of sales, nine months to June 2026
64%36%
Networks, cloud and AIIndustry, medical, defence and cars
It is paid per product built. For AMD it will assemble the Helios rack, which holds 72 AI chips, in factories it bought from AMD.
The company · 3 · How does it make money?
It keeps about a dime of each sales dollar after building
10.7 centskept per $1 after parts and labour, a record
7.0% to 7.5%profit per dollar expected next year, before some costs
$541Mspare cash in the year, after factory spending
Most of each sales dollar goes on parts. Office costs barely rise with sales; materials and factory spending do. Much of its money sits in stock and unpaid customer bills.
Is it a good business? · 4 · What gives it an edge?
Moving a product to another builder costs time and approvals
Helios
trial runs done here would have to be repeated elsewhere mid-launch
AMD's choice
its preferred American partner for new AI racks
In medical, defence and aerospace a line is certified to each product, so moving it means new approvals. AMD bought ZT Systems, kept the designers and sold the factories to Sanmina.
Is it a good business? · 5 · What is the catch?
Bigger rivals, and AMD decides the volume
+81%Celestica's hardware sales growth in 2025
41%of Jabil's sales come from data centres
5 yearsthe AMD factory agreement, with no minimum volume
AMD kept the more valuable design work. It could hand Helios volume to a second builder, and the top ten customers already buy 63% of Sanmina's sales.
Is it a good business? · 6 · What does its record show?
Sales jumped after years of going sideways
Sales, $B
7.9
8.9
7.6
8.1
14.5
Oct 2022Sep 2023Sep 2024Sep 2025Oct 2026
Sales, fourth quarter+84%to $3.86B
Sales for the year+79%to $14.5B
Tax rate19.5% → 28.5%as production moved to America
Sales were stuck near $8B for years, then the ZT factories added AI work. Growth adds profit in step with sales, no faster.
Is it a good business? · 7 · How much room is there to grow?
Nobody counts the racks, but the spending is visible
$173.0BAmazon's equipment spending in a year, from $107.7B
$5.0BCelestica's hardware sales in 2025, a similar builder
Outside builders already earn billions from cloud spending. How many AI racks the market needs, nobody publishes.
Is it a good business? · 8 · Who is in charge?
The co-founder still runs it, paid on adjusted figures
Since 1980Jure Sola, co-founder and chief executive
About 3%owned by insiders, nearly all his
19 of 20open-market insider trades in a year were sales
Pay depends on profit figures that leave out share pay and the cost of the ZT deal. Nothing pays on the return the money earns.
The story right now · 9 · Why does AMD need Sanmina?
AI chips now ship as whole rack-sized computers
72 AI chipsin one AMD Helios rack
$17.5B to $19.5BSanmina's sales forecast for next year
AI chipmakers no longer sell single chips; they sell whole racks wired, cooled and tested. Someone has to build them by the thousand, and AMD picked Sanmina.
The story right now · 10 · Why is next year's forecast risky?
Next year's jump depends on racks not yet shipped
This quarterSales of $3.86B.
Next quarterForecast level, $3.8B to $4.1B.
Then the racksMust lift weekly sales 29%.
If they slipThe year falls short.
Management says rack sales are weighted to the second half of its year. That leaves time for delays. The price already allows for some shortfall, which is our cushion.
The story right now · 11 · What changed lately?
Sales rose 84%, and next year's forecast cleared our bar
+84%fourth-quarter sales
$17.5B to $19.5Bnext year's sales forecast
$3.8B to $4.1Bthe first quarter's forecast, level with the fourth
118,368shares the chief executive sold in May
Strong quarter
The forecast is the test we were waiting for, and it passed with no sign of a slip. The step up still comes late in the year.
The story right now · 12 · What is everyone arguing about?
Will the AMD racks ship on time?
Many investors think
A thin-profit assembler with one big customer and AI sales not yet shipped
We think
Right about the thin profit, but the price already pays for a shortfall
We are wrong if first-quarter sales miss $3.8B, or if the racks are pushed to the second half or later.
The price and our view · 13 · What is a share worth?
The price sits about two-fifths below our value
$255what we think a share is worth · the last close is of that
$255our value
Attractive below $204
Neutral
Unattractive above $306
$150gloomy: growth halves and profit per sale slips
$255our estimate: the middle of the forecast, then slowing
$416sunny: 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?
One programme carries it, and AMD holds the volume
Few customersBiggest
The top ten are 63% of sales.
AMD holds the volumeWatch
No minimum is disclosed for Helios.
A heavier tax billShowing now
Tax rose as production moved to America.
Fees that endUnknown
How much profit is one-off engineering fees.
The payments tied to the ZT deal are valued twice: AMD expects $348M, while Sanmina now carries $261M for the same agreement.
The price and our view · 15 · How will we know?
February shows whether the racks ship
First-quarter sales of $3.95BGood sign
Rack sales counted in the first quarterGood sign
AMD names a second builder for HeliosWe turn negative
FEB 20271
Results for its first fiscal quarter of 2027. 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?
A thin-profit assembler about to build AMD's AI racks
$255our value
Attractive below $204
Neutral
Unattractive above $306
AttractiveMedium conviction
Sanmina's forecast cleared our bar and the price already allows for a shortfall. Conviction is Medium because one programme carries the growth, AMD controls the volume and nothing has shipped yet.
↑ ↓ to move · C for contents
What it does
More about the company
Who it is
Sanmina builds electronics to other companies' designs in its own factories. For AI that means a whole rack-sized computer: it buys the parts, assembles the circuit boards, fits them into a rack with power and liquid cooling, runs it at full load and ships it to a data centre. AMD's Helios rack, which it is preparing for production, holds 72 AI chips.
It also makes circuit boards and other parts that go inside, which most assemblers buy in.
How it makes money
It earns a thin cut on everything it builds: even June's record was 10.7 cents of adjusted gross profit per sales dollar. Before volume starts, customers also pay engineering fees to prepare their products for production. Because Sanmina buys the parts, much of its money sits in stock and unpaid customer bills: most of what it paid AMD bought exactly that.
Growth and profit
What growth costs
Growing costs Sanmina little in overhead: office costs and share pay barely rise with sales. What does rise is materials and, since ZT Systems, spending on factories. So growth adds profit roughly in step with sales rather than faster, and the margin has not widened since the deal.
Its record
The numbers
Is it growing?Fast
Sales grew 79% in the year to Oct 2026, faster than the 7.4% in the year before.
Is it profitable?Thin
Operating margin is 4.7%, about where it was (5.1%) in the year to Sep 2023.
Does growth pay off?Flat
Running costs take 5¢ of each sales dollar, up from 3¢ in the year to Oct 2022.
Does it earn good returns?Decent
Each dollar put into the business earns 12% a year after tax, about where it was (13%) in the year to Oct 2022.
Is it making cash?Thin
It turned 3.7% of sales into free cash, $540M over the year.
Is the balance sheet safe?Light debt
It owes $412M more than the cash it holds, about 0.4 years of the cash its operations bring in.
Are shares being diluted?Shrinking
The share count fell 13% in 4.0 years, and pay in shares costs 0.7% of sales.
Why it wins
The evidence
Two kinds of lock-in, of different strength. In medical, defence and aerospace a factory line is certified to the product, and moving it means certifying again. In AI racks the cost is time: a customer that put a product through Sanmina's pre-production work would have to repeat it elsewhere mid-launch.
AMD's choice is the evidence. Having bought ZT Systems for $4.4B, it kept the designers, sold the factories to Sanmina and named Sanmina its preferred American partner for launching new AI racks.
Who could beat it
The rivals in full
Bigger rivals do the same work. Celestica's hardware business grew 81% to $5.0B in 2025, and Jabil's data-centre segment was 41% of its $29.8B of sales. And AMD kept ZT Systems' designers, so the higher-value half of the work stays with AMD, which decides where the volume goes.
Each rival
Celestica, through its Hardware Platform Solutions businessCLS
The rack and system-level build a hyperscaler would otherwise place with Sanmina. Celestica is engaged directly as an original design manufacturer, so it takes the design work as well as the assembly.
Jabil, through its Intelligent Infrastructure segmentJBL
The cloud and data-centre infrastructure assembly Sanmina is competing for, at roughly twice Sanmina's whole-company scale in that end market.
AMD's own retained ZT Systems design organisationAMD
The design-for-manufacture and new-product-introduction engineering that is the profitable part of what Sanmina does today. AMD kept those functions when it sold the factories, so the higher-value half of ZT Systems already sits with the customer.
How big it could get
The market and who has signed
The market
The market is rack-scale AI computers built by outside manufacturers, and nobody counts them. The spending behind them is visible: Amazon spent $173.0B on property and equipment in the year to June 2026, against $107.7B a year earlier. Celestica's $5.0B of hardware sales and Jabil's data-centre segment show what outside builders already earn from it.
Who has signed
AMD signed a five-year manufacturing agreement with the ZT business when Sanmina bought it, and management says it holds orders from large cloud companies and equipment makers for the next AI programme. No volume, minimum or customer name has been disclosed for any of it.
Who runs it
Leaders, pay and the cash
Who runs it
Jure Sola co-founded Sanmina in 1980 and has run it for most of the time since; Jonathan Faust has run the finances since 2023. Insiders own about 3%, nearly all of it Sola's. Their bonus rides on sales and an adjusted profit margin, their shares on adjusted earnings per share over three years. Nothing pays on the return the money earns, and the adjusted figures leave out the cost of share pay and of the ZT earn-out.
What they did with the money
It borrowed to pay for ZT, taking debt to $2.17B, though $1.84B of cash covers most of it. It bought back 1.59 million shares for $239.2M, more than the shares issued to AMD. Free cash flow is on course to be no higher than fiscal 2025's $478M, on nearly twice the sales.
The last report
The latest report
Sanmina's sales rose 84% and it guided next year to $17.5B to $19.5B
What it did to the thesis Strong quarter
It cleared the test we set: a forecast above the promised target and a margin well above the line. The call kept the picture of a late-year step and gave no sign of a slip. With the shares well below their September price, the view moves up to Attractive.
The report
The fourth quarter of fiscal 2026 brought in $3.86B, up 84%, about 12% a week above the company's own forecast once the extra week is removed. For the year, sales were $14.5B, up 79%, and free cash flow was $541M, above last year's $478M.
Next year's forecast is $17.5B to $19.5B at an adjusted margin of 7.0% to 7.5%. The first quarter's $3.8B to $4.1B is level with the fourth, so the step comes late in the year. On the call management said the AI sales are still weighted to the second half, with a second customer's platform in the forecast and a third outside it.
What's happening now
The background
The forecast Sanmina gave on 9 November is the test the thesis was waiting for. Getting to its middle needs weekly sales 29% above the fourth quarter's, from an AI programme not yet counted as revenue. On the call management said the AI rack sales are still weighted to the second half.
Meanwhile insiders have sold: the chief executive sold 118,368 shares in May, and 19 of 20 open-market insider trades in the past year were sales.
Where we disagree
Both sides in full
The market
An assembler deserves an assembler's price: thin margins, one large customer, and AI sales that have not yet shipped.
We believe
The market is right about the thin margin. At November's price it already pays for the forecast falling short, which the forecast itself does not.
The argument
The first quarter's forecast is level with the fourth quarter, so the whole step to $18.5B rests on the AI programme, which has shipped nothing.
We are wrong if first-quarter sales miss $3.8B, or if the racks are pushed to the second half or later.
What it's worth
How we value it
Valued on the guide's middle and 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 earn-out owed to AMD, carried at $261M in the annual report, is subtracted with the debt; 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
1.7%
Our estimate
27%
Measured at the close of
17 Nov 2026
The three cases
Bear case$150
Growth runs at half the base path, 14% and 10%, the margin settles a fifth lower at 4.8%, and new capital earns only the 8% rate: the downturn test.
Our value$255
Revenue grows 27% and 20% over the next two years: the middle of the $17.5B to $19.5B guide (C57) on fiscal 2026's $14.53B, which is where the analysts now sit, then their fiscal 2028 revenue (C58); it then fades toward 4%, closing 30% of the gap each year. The margin moves to 6.0%, the analysts' year-three margin after stock pay, below the 6.4% best it has filed.
Bull case$416
Growth runs a quarter above the base path, 34% and 25%, the margin reaches 6.6%, and new capital earns 11%.
The bet
Seen as a bet
The bet is on AMD's Helios racks alone: complete AI server racks Sanmina is preparing to build in America, which have not shipped and carry no disclosed volume.
Today's business, valued at an ordinary contract manufacturer's price, covers under half of today's price, so buyers already pay a lot for the racks. Even if they bring the step in sales management promises for next year, the whole company would be worth less than today: an expensive bet.
If the racks never ship in volume, shareholders keep today's business.
What could go wrong
Every risk
The risks
One customer carries it
The ten largest customers were 63% of the year's sales, against about half before, and one relationship carries most of the growth.
AMD holds the volume
AMD kept the design team and has disclosed no minimum, so it could hand Helios volume to a second manufacturer.
A heavier tax bill
The tax rate rose to 28.5% from 19.5% as production moved to the United States.
An earn-out valued twice
AMD expects to receive $348M from the earn-out; Sanmina carries $183M for the same agreement, capped at $450M.
What we still do not know
How much of the record margin is engineering fees that will not repeat once racks ship in volume. Management has not said.
How many Helios racks AMD will order, and when. No volume or minimum is disclosed.
What changes our view
Every route to a new view
Every route to a new view
→ Neutral
The share price rises to about $204, 80% of what we think the business is worth, so it is no longer cheap enough.
→ Unattractive
The share price rises to about $306, 20% above what we think the business is worth.
→ Unattractive
AMD names a second manufacturer for Helios volume, or Sanmina's share of the programme falls. The ten largest customers pass 75% of sales while total sales stop growing.
→ Neutral
First-quarter sales fall below $3.8B, or the full-year forecast is cut below $15B. Management says the AI rack volume ships in the second half or later.
The next test
It supports the case if
Revenue at or above $3.95B, adjusted operating margin at or above 7.0%, and the accelerated-compute programme recognised as contributing in the quarter.
It pressures the case if
Revenue below $3.8B, or adjusted operating margin below 7.0%, or the programme's first revenue pushed to the second half or later.
The price is about three-fifths of our value and level with the bear case.
BusinessNegative
A thin-margin assembly trade, now dominated by AI work, with its best margin leaning on fees paid before shipments.
GrowthPositive
The guide for next year sits above the promised target, and the price needs only a modest margin on it.
RisksNegative
Ten customers make up most of the sales, the chip maker keeps control of volume, tax is rising and the earn-out owed to it has grown.
RivalsMixed
Rival assemblers chase the same racks; holding the factories the chip maker sold gives it a head start.
The groupMixed
Server and network builders move together on AI spending news; the rack programme is its own fact.
Latest newsPositive
The latest report cleared our test with a forecast above target and a margin well above the line; the racks must still ship.
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.