Three weeks ago,
SasaDaily covered an important milestone.
Foxconn’s Q2 revenue:
Cloud networking products including AI servers
accounted for the first time:
over half of total revenue.
At that time, the real takeaway wasn’t:
how much profit Foxconn made that quarter.
It was this:
a large manufacturer long associated with:
iPhones,
consumer electronics,
and contract manufacturing,
is seeing its revenue structure
rewritten by
AI servers.
Today,
that story has advanced another step.
Not through new quarterly earnings,
but via:
monthly revenue.
Foxconn’s August Revenue Hits NT$921.8 Billion
According to Reuters on September 5,
Foxconn’s August revenue reached:
NT$921.8 billion.
A year-over-year increase of:
51.98%.
This is not only a significant jump,
but also:
Foxconn’s highest August revenue ever.
Plus, it marks the second consecutive month
with monthly revenue surpassing:
NT$900 billion.
July Was Actually the First Month Over NT$900 Billion
In July this year,
Foxconn’s monthly revenue was about:
NT$946.5 billion.
With a year-over-year rise of around:
54.2%.
That set a new July record
and was the first time monthly revenue exceeded:
NT$900 billion.
In August,
though the figure dipped slightly compared to July,
it remained:
above NT$900 billion.
This is more meaningful than just:
a single month spike.
Because what the AI industry really needs to prove is never:
a one-time big order,
but
whether high-volume shipments can be sustained.
Foxconn Says Q3 May Outperform Market Expectations
In a statement, Foxconn said,
in Q3 as:
AI demand continues to grow,
and ICT products enter the seasonal peak in the second half of the year,
operational momentum is expected to gradually increase.
Particularly notable is the phrase:
the company currently sees:
better visibility for Q3 than a month ago,
and expects:
overall performance to beat market forecasts.
This is different from simply saying:
"Q3 will grow."
It means that from the company’s current view of:
orders,
demand,
and shipment plans,
the outlook is clearer than it was a month ago.
But Foxconn Did Not Provide a New Detailed Forecast
Don’t overinterpret this.
Foxconn did not announce:
how much revenue or profit Q3 will produce,
or how many AI servers will be shipped.
The company does not normally provide precise quarterly revenue guidance.
So what we can confirm today is:
the company is more optimistic about Q3 operational visibility,
but we cannot infer:
"Q3 will definitely hit new highs,"
or
"AI server revenue will definitely double again,"
as there’s no official data supporting those claims yet.
Why Is Today’s Number Worth Revisiting Foxconn?
Because the article on August 12
answered this question:
Has AI already changed Foxconn’s revenue structure?
At that time, the answer was becoming very clear.
Foxconn’s Q2 revenue:
NT$2.53 trillion,
up 41% year-over-year,
with net profit of:
NT$60 billion,
up 35% year-over-year.
More importantly,
cloud networking products made up:
51%
of revenue—
surpassing consumer electronics for the first time.
This shows that AI servers are not just:
a "new growth product,"
but are reshaping:
what drives the company’s revenue.
Today’s New Question: Can This Change Be Sustained?
If only Q2 was strong,
it could be due to:
concentrated one-off shipments,
customer pull-ins,
or product cycles.
But now,
July surpassed NT$900 billion,
August again sustained over NT$900 billion,
and the company says Q3 visibility has improved.
This shifts the story from:
structural change
to:
sustained scaling.
The AI Boom Must Ultimately Result in “Real Products Being Made”
Every day we see:
OpenAI,
Anthropic,
Google,
Meta,
new models,
new agents,
new products,
but behind these services are:
GPUs,
server racks,
networking,
power,
cooling,
storage,
and data centers—
real physical equipment.
When a model company says:
"I want more computing power,"
this doesn't just add cloud capacity.
It translates into:
chip orders,
PCB assembly,
power modules,
liquid cooling equipment,
high-speed networking,
server assembly,
rack integration,
testing,
shipping,
and data center installation.
This is why,
as the AI boom expands,
Taiwan’s manufacturing sector will directly feel the impact.
Foxconn’s Role Is Deep in This Supply Chain, Critical But Far Downstream
Nvidia designs GPUs.
Other companies supply memory,
network chips,
CPUs,
but all the chips and components
must be assembled into:
working systems in data centers.
Not single GPUs,
but complete servers,
full racks,
power, cooling,
networking,
and extensive integration.
This is the role of a major electronics manufacturer.
As AI servers grow more complex,
the importance of integration capabilities also rises.
Today’s AI Servers Are Far More Than Traditional Servers With a Few GPUs
High-end AI racks must handle:
very high power consumption,
many interconnected GPUs,
high-speed networks,
liquid cooling,
power distribution,
firmware,
system testing,
and full rack integration.
If any part of:
cooling, power, cables,
or networking
fails,
the expensive GPUs won’t function properly.
So the real bottleneck in AI infrastructure
is often not:
"Do we have a GPU?"
but
can the entire system be mass-produced stably.
This Explains Why Foxconn’s AI Growth Is More Than Just Adding a New Product Line
If AI servers were just another new product,
it might mean:
"We used to make phones, now we make servers too."
But the scale we see now is affecting:
revenue share,
capital expenditures,
global factory allocation,
capacity,
the supply chain,
and even
how the company defines itself.
At its August earnings briefing, Foxconn said,
the demand for AI production capacity in 2027
remains very strong.
The company is also preparing increased capital expenditures
in US states like:
Texas,
Wisconsin,
Ohio,
and California.
This means AI for Foxconn
is not just about a few server orders this year,
but about
the next round of factory and capacity planning.
Why Can AI Servers Boost Revenue So Fast?
Partly because unit prices are high.
A smartphone is a single consumer electronic product,
but a fully loaded high-end AI server rack
can include:
huge numbers of the latest GPUs,
networking equipment,
cooling systems,
and power infrastructure.
The system value is entirely different.
Each large AI data center deployment
generates very large revenue at every stage of the supply chain.
This explains why
even though there are only a few large data center customers,
they bring orders of extraordinary scale.
But “Revenue Growth” and “Profit Margin Growth” Are Different Things
This is critical when examining manufacturing figures.
Revenue reveals shipment scale,
while
profit depends on:
product mix,
costs,
yields,
customer negotiation,
exchange rates,
investment,
and factory utilization.
So seeing Foxconn’s monthly revenue of
NT$921.8 billion
doesn’t directly mean the company earned
50% more profit.
The 51.98% referenced by Reuters today
is
revenue growth,
not profit growth.
The real Q3 profit figures
will only be known
after official quarterly earnings are released.
Rapid AI Server Growth Also Brings New Costs
Such as:
liquid cooling production lines,
new factory equipment,
testing,
logistics,
localized manufacturing in the US,
talent,
power supply,
supply chain backup.
The company must first invest
capital expenditures
to turn market demand into
actual shipments.
So the “AI boom” for manufacturers is not just:
collecting revenue,
it also means:
having the capacity-building ability.
What to Watch Next Is Not Just Whether Monthly Revenue Breaks Records Again
NT$921.8 billion is huge.
But the real question is:
how long can the high demand for AI servers last?
If demand remains strong in 2027,
the AI infrastructure cycle
won’t just be a one-year spike,
but possibly a multi-year
hardware upgrade cycle,
like we saw with:
PCs,
smartphones,
and cloud computing.
AI might now be forming:
a new
infrastructure cycle.
Second: Watch Whether Customers Start Designing More ASICs Themselves
AI computing power isn’t only about:
Nvidia GPUs.
Major tech firms like:
Google,
Amazon,
Microsoft,
and Meta
are developing their own custom chips—
ASICs.
This isn’t necessarily bad for Foxconn,
because whether it’s:
GPU servers
or
ASIC servers,
they still need:
manufacturing,
integration,
rack assembly,
cooling,
and data center deployment.
Foxconn itself has said it’s
gradually raising market share in GPU and ASIC
categories.
So the key question is:
whether the total AI hardware market continues to expand,
not just one chip vendor.
Third: Whether AI Data Centers Will Face Power Constraints
Just recently, SasaDaily reported on how Texas
has started inspecting
data center grid connections,
water resources,
and local costs.
This reminds us that even if servers can be built,
data centers can’t necessarily be expanded without limits.
Ultimately, AI infrastructure will be limited by:
chips,
memory,
packaging,
server capacity,
power grids,
land,
cooling,
and capital.
If AI data center construction slows,
the supply chain will be impacted as well.
Fourth: Global Politics
Foxconn reminded today that we still need to watch:
the volatile global political and economic situation.
This is not just boilerplate.
AI servers involve:
US export controls,
the Chinese market,
chip sourcing,
supply chain localization,
tariffs,
US manufacturing,
and geopolitical risks.
A single AI rack’s components
may come from multiple countries,
and be delivered to data centers
in different regions.
As the AI server business grows,
geopolitical risks will rise accordingly.
For Taiwan, This Is a More Tangible Impact Than News About New AI Models
Most people don’t hold:
OpenAI stock,
Anthropic shares,
or own US data centers.
But Taiwan has a complete
semiconductor,
PCB,
server,
power,
cooling,
networking,
ODM,
manufacturing,
and supply chain ecosystem.
Every major AI compute order
ultimately translates into
products actually delivered
by Taiwan’s supply chain.
This is why AI is moving
from pure tech headlines
directly into
exports,
GDP,
corporate profits,
jobs,
and investments.
But This Doesn’t Mean Taiwan Can Rely Forever on “Making AI Servers for Others”
This raises a longer-term question.
The AI infrastructure boom benefits Taiwan greatly,
but the truly high-value portions
include:
models,
software,
cloud,
data,
applications,
and AI services.
If Taiwan only benefits from hardware demand spikes,
but doesn’t develop
its own AI applications and services capabilities,
its long-term value will remain limited.
So Foxconn’s record revenue is good news,
but it also reminds us:
Taiwan currently captures most AI gains through hardware and manufacturing.
How Should the Average Person Understand Today’s News?
No need to decide
whether to buy the stock first.
No need to memorize
the NT$921.8 billion figure.
Focus on understanding this:
If AI were only
lots of people chatting on ChatGPT,
a large manufacturer wouldn’t
have two consecutive months of
monthly revenue above NT$900 billion.
Today’s number really means:
AI has shifted into
large volumes of physical equipment,
major corporate spending,
massive data center construction,
and large supply chain orders.
In other words,
the AI boom is moving from screens to tangible products factories must build, ship, and customers must pay for.
This Is a Key Marker When Judging AI Bubbles
Markets can still:
overinvest,
inflate valuations,
and build redundancies.
Overcapacity may also occur in the future.
But we can no longer view the whole AI boom as
only stock prices and hype.
Because now we see:
real revenue,
real equipment,
real capacity,
and real shipments.
The more precise question is not:
"Is all AI just a bubble?"
but
"Can today’s huge physical investments be supported by future AI usage and revenue?"
This will be the real test ahead.
Three Weeks Ago, We Saw “AI Changing Foxconn’s Structure”
Q2:
Cloud networking products made up
51% of revenue,
becoming the
largest product category.
Today we see:
July:
monthly revenue topped NT$900 billion for the first time,
August:
again exceeded NT$900 billion,
year-over-year growth of:
51.98%,
and in Q3:
the company reports better visibility,
expecting to beat market estimates.
So tonight’s update is not just:
"AI servers are popular,"
but rather,
the structural shift is turning into sustained, large-scale shipments for two consecutive months.
Three Future Milestones to Watch
First:
September revenue,
to see if the high level holds.
Second:
Q3 earnings,
to see if high revenue translates into
solid profits.
Third:
2027 AI capacity demand,
to verify whether the massive orders now
turn into actual shipments.
Answers to these
will be clearer indicators than any AI benchmark
of how real this AI infrastructure boom is.
Tonight’s Most Important Takeaway
If the AI boom only stayed at the
model launch level,
it wouldn’t be an industrial revolution.
But when a major Taiwanese manufacturer
posts two consecutive months of over NT$900 billion in revenue,
and clearly attributes growth momentum to:
AI demand,
it means the story isn’t just:
"Everyone is optimistic on AI,"
but rather:
AI is turning into large volumes of real physical products factories must build and customers need to pay for.
The next questions are no longer whether AI has demand,
but:
how long can this demand last, and can the huge computing power built today produce enough value in the future.
Today, progress a little with AI.
Learn one AI skill daily.
Save a little time daily.
Improve a bit every day.
SasaDaily, growing together with you.
