Today’s three AI news stories,
initially seem completely unrelated.
One involves:
AI chips
being sent to space.
Another involves:
ads
being placed inside ChatGPT.
The third is about:
raising a massive
$11.1 billion.
But in fact, all three address:
three critical questions as the AI industry moves into its next phase.
Where will computing power be located?
Where will revenue come from?
Where will the funding come from?
① Google to Send TPU to Space Next Week: Not Building a Space Data Center Yet, But Testing If AI Chips Can Survive Launch, Radiation, and Space Cooling
The first news
sounds like science fiction,
but the test starts next week.
On September 24, Google announced:
Project Suncatcher
will conduct its first-ever
in-orbit test.
Google plans to send its own
Tensor Processing Units
— TPU AI chips —
inside a prototype satellite,
launching it into
Low Earth Orbit.
The goal isn’t to immediately run a full Gemini Data Center in space.
The first step is to answer:
Can AI chips operate reliably in space?
The satellite will fly on SpaceX’s Transporter-18 rideshare mission
This mission is a collaboration between Google
and satellite company Planet.
The prototype satellite will be launched via
SpaceX’s Transporter-18
Rideshare Mission.
According to Reuters,
the launch is planned for
next week.
Once in orbit,
Google will test the TPU’s ability to withstand
the extreme vibrations and forces of launch,
space radiation,
extreme temperatures,
and the vacuum environment,
to determine if it can operate stably.
Why are AI companies suddenly looking to put data centers in space?
Because on Earth, AI is facing a
very physical challenge.
It’s not that
models aren’t smart enough,
but rather
electricity is becoming harder to access.
Large AI data centers
require
massive power,
land,
cooling,
power grids,
substations,
long construction times,
and government approvals.
The bigger the model,
the more severe these challenges become.
Google’s Project Suncatcher
poses an extreme question:
If AI needs so much energy, could we one day get solar power directly from space?
Google says low Earth orbit satellites receive nearly continuous sunlight
Google states that
satellites orbiting in low Earth orbit
can access
near-constant sunlight.
The company estimates the potential to generate up to
eight times more solar energy than on the ground.
The long-term vision is not just one satellite,
but many connected satellites carrying TPUs,
forming an
orbital compute cluster
working together to handle
large-scale AI workloads.
But we're still far from a “space AI data center”
This must be clear.
Next week isn’t
a soft launch of a Google space data center.
Nor is it
Gemini permanently running in orbit.
It’s simply
a first prototype test satellite.
Many problems must be solved first.
One of the most immediate challenges is
heat dissipation.
Space is cold—but chips don’t cool easily
This is counterintuitive.
We often think,
“Space is so cold; AI chips should cool easily up there.”
But space is a vacuum.
There’s no air.
Common data center cooling methods on Earth, such as
fans, airflow,
and air conditioning,
don’t work in a vacuum.
TPUs generate
a lot of heat in a confined space,
so Google must design
heat pipes,
and radiators
to dissipate heat via thermal radiation.
This orbital mission will also verify if these designs are effective.
Radiation is another concern
Servers on Earth are shielded by
the atmosphere,
the magnetic field,
and data center buildings.
Space provides no such protection.
High-energy particles may cause
bit flips,
errors,
and even hardware damage.
Google has already run radiation tests on the ground,
but admits
some questions can only be answered by going into space.
So the real takeaway isn’t just “Google is cool for sending AI to space”
but
that AI infrastructure
is being pushed into new, rarely considered territory.
In the past, AI companies competed on
GPU chips,
models,
and tokens.
Now competition expands to include
power grids,
nuclear power,
solar energy,
data center land,
cooling,
and even
orbit.
This shows AI’s bottleneck
is moving
from purely software
into the
physical world.
② ChatGPT Ads Arrive in Taiwan: Under 200 Days Since Launch, Annualized Revenue Hits $1 Billion
The second story
directly affects Taiwanese users.
On September 23, OpenAI announced:
ChatGPT Ads
are expanding to:
Taiwan,
Indonesia,
Malaysia,
the Philippines,
Singapore,
Thailand,
and Vietnam.
This adds
seven Asian markets
bringing ChatGPT Ads presence to
more than 60 countries.
Not all ChatGPT users see ads
OpenAI’s policy is:
ads only appear for users on
Free
and
Go
plans.
Users on
Plus,
Pro,
and Enterprise
plans stay
ad-free.
So
“ChatGPT Ads expand to Taiwan”
does not mean
all Taiwanese ChatGPT accounts will see ads.
Whether ads appear also depends on
the subscription plan
and relevance to the current conversation.
Ads won’t be mixed into ChatGPT’s answers
This is a key principle OpenAI stresses.
Ads are
clearly labeled,
separate from ChatGPT’s responses.
Advertisers cannot pay to
influence how ChatGPT answers.
OpenAI also states
user conversations are not
directly shared with advertisers,
and user data is not sold
to advertisers.
This boundary is crucial,
because ChatGPT Ads differ greatly from
Google Search Ads.
Search ads typically rely only on your keywords
For example, if you search
“coffee shops in Tainan,”
the engine knows you’re likely looking for
coffee.
But a ChatGPT conversation might include
“I’m looking for a coffee shop in Tainan for four people on Sunday afternoon, someone needs power outlets to work, quiet environment, budget 300 TWD per person.”
The intent behind this
is much richer than just a keyword.
So for advertisers,
conversation-based advertising
is valuable not because
ad slots are prettier,
but because users are
making decisions.
OpenAI views this as a new advertising channel
OpenAI says people use ChatGPT mostly when
planning trips,
choosing software,
renovating,
shopping,
learning new skills,
or job hunting.
In other words,
it’s not just random feed browsing.
Many conversations are
decision journeys.
Advertisers want to engage
when users are comparing options.
And the business is growing faster than expected
OpenAI reports that by the end of August,
less than
200 days
since its launch,
ChatGPT Ads reached
$1 billion in annualized revenue scale.
It already has
tens of thousands of advertisers.
This figure should not be taken to mean OpenAI collected $1 billion in full-year revenue yet.
Annualized revenue run rate means
the current revenue pace extrapolated to a full year.
Still,
reaching a $1 billion scale in under 200 days is a very rapid commercialization pace.
Taiwanese businesses can now start buying ads
OpenAI says advertisers
can access ChatGPT Ads through
OpenAI Ads Solutions,
agency partners,
technology partners,
and qualified companies via
Ads Manager self-service.
Agency partners include
dentsu,
Havas,
Omnicom,
Publicis,
and WPP,
large global ad groups.
This indicates ChatGPT Ads
is no longer a small beta experiment,
but is evolving into
a true global ads platform.
What Taiwanese small businesses should watch is not immediate ad buying, but measuring effectiveness
For example,
a small Taiwanese SaaS company
that previously ran
Google Search Ads
and Meta ads
now faces a new option:
ChatGPT Ads.
What really matters is comparing:
how much is spent per lead,
cost per genuine trial,
and final conversion cost.
Not simply
“ChatGPT is trending, so ROI must beat Google.”
Another challenge: can ads and answers trust remain separate long-term?
This might be
the toughest long-term question for ChatGPT Ads.
People accept
sponsored results in search engines.
But AI assistants
provide a more conversational relationship.
If users begin to wonder,
“Are you recommending this because it paid for ads?”
the whole product’s trust
will be affected.
So OpenAI repeatedly stresses:
Ads do not affect answers.
This is not marketing fluff,
but the core to whether ChatGPT’s ad model can sustain long-term.
③ SoftBank Issues $11.1 Billion High-Yield Bonds Mainly to Complete Its OpenAI Investment
The third news story
shows
another side of the AI boom’s tab.
On September 24, SoftBank completed
an issuance of approximately
$11.1 billion
in U.S. dollar and euro bonds.
According to Reuters,
This is
the largest
high-yield corporate bond sale ever recorded worldwide.
The US dollar bonds have coupon rates up to 9.75%
The dollar bonds were issued in three tranches:
$1 billion for 3.5 years,
with an 8.625% coupon,
$4.5 billion for 5.5 years,
with a 9.25% coupon,
and another $4.5 billion for 7.5 years,
with a 9.75% coupon.
In addition, two €500 million bonds
were issued,
with yields around 7.125% and 8%, respectively.
Clearly, this financing is
not cheap for SoftBank.
The largest use of funds is to pay OpenAI
Previously, SoftBank pledged to invest
$30 billion in OpenAI by 2026.
This new bond issuance will fund the last
$10 billion tranche
of that commitment.
Reuters notes that after this,
SoftBank’s total investment in OpenAI
will reach
$64.6 billion.
This is no ordinary tech investment,
but an extremely concentrated, large-scale capital allocation.
SoftBank also borrowed a large amount earlier this month
Reuters reports that earlier in September, SoftBank
issued
1 trillion yen
in corporate bonds to Japanese retail investors,
roughly $6.3 billion.
Now followed by $11.1 billion more,
this represents a massive influx of new debt funding
in a short period.
The reason is simple:
AI investment capital needs
have grown so large
that existing cash flow may not suffice.
This shift shows the AI boom is moving from equity markets into bond markets
In recent years,
AI investment discussions
focused on venture capital,
rising stock prices,
and corporate capital expenditures.
Now, another player emerges:
bond investors.
AI companies need cash for
chips,
land,
data centers,
power,
model training,
acquisitions,
and investments.
If internal cash
is insufficient,
they must
borrow.
Thus, the AI boom
is spreading from
the tech sector
into the
credit market.
The key difference with debt: interest must be paid
If stock prices fall,
companies don’t necessarily need to
pay cash immediately.
But bond coupons
must be paid on schedule,
and principal must be repaid.
For example, SoftBank’s
7.5-year US dollar bond
carries a high coupon of
9.75%.
Investing in OpenAI requires
not only the promise it will be valuable someday,
but also sufficient returns
to justify
the current financing cost.
So OpenAI faces not only “Who has the best model?”
but also
the
entire capital chain’s
return expectations.
SoftBank,
Microsoft,
Nvidia,
data center investors,
bond holders,
and other capital providers —
the bigger the investment,
the more they ask:
Where is the revenue?
This links back to
the second news story.
ChatGPT Ads
are important because,
increasing model capabilities is one thing,
turning a billion users into sustainable revenue is another.
These three news stories actually form a connected storyline
Google is tackling
where to place compute power,
even testing sending TPUs into space.
OpenAI is figuring out
how to monetize massive user volume.
Therefore, ChatGPT Ads
are officially expanding to Taiwan and more markets.
SoftBank is solving
where the money comes from to fund this AI expansion.
Hence the $11.1 billion bond issuance.
This is the real change happening in the AI industry
Previously,
looking at model leaderboards
gave a rough sense
of who was ahead.
That’s no longer sufficient.
A company might have
excellent models,
but can’t secure enough power.
Another might have
lots of users,
but insufficient revenue to sustain compute costs.
Yet another may have
grand AI ambitions,
but rising financing costs.
So the real future AI competition will occur across
models,
compute,
energy,
distribution,
advertising,
capital,
and even
space infrastructure.
AI is becoming less of just a software business
A typical app
is developed,
deployed to the cloud,
and charges monthly fees.
This is simple.
Cutting-edge AI is different.
It requires
advanced chips,
massive data centers,
power,
cooling,
networks,
satellite research,
massive financing,
and global commercialization.
So what we’re seeing
is not just a few software companies competing,
but
a whole new industrial infrastructure taking shape.
Today’s three news stories
make this very clear.
AI’s next phase
is no longer just
“How much smarter can it get?”
but
“With intelligence this massive, where does it run, how do you make money, and how much capital does it take to sustain?”
Today, we progress a little with AI.
Learn an AI skill daily.
Save a bit of time every day.
Improve your abilities step by step.
SasaDaily, growing with you.
