After an AI model opens its weights,
shouldn't the norm be:
everyone downloads it,
deploys it themselves,
and then the model company earns through APIs or enterprise services?
Moonshot AI is now telling the market:
things might not be that simple.
Reuters on September 3 cited three insiders reporting that
the Beijing AI startup:
Moonshot AI
has secretly filed for a Hong Kong IPO.
One source says the company hopes to raise approximately
$3 billion.
But what’s truly worth noting tonight is not just:
“Another Chinese AI company preparing to go public.”
Rather, it is that
Kimi K3 has brought the next challenge of open-weight AI models to the forefront: free access to models does not mean running them is free.
Clarifying: Moonshot Has Not Officially Announced Its IPO
This news comes exclusively from Reuters.
Reuters cites three insiders familiar with the plans stating
Moonshot has secretly submitted its Hong Kong IPO application.
The company has not responded to Reuters’ requests for comment at this time.
So the most accurate statement now is:
“Reportedly submitted secretly.”
It cannot be stated as:
Moonshot officially announced its Hong Kong listing today.
The listing still requires
regulatory approval,
market conditions,
and subsequent procedures.
The fundraising amount may also change.
Who Is Moonshot?
If you have been following Chinese AI recently,
you have probably seen it many times already.
Moonshot AI was founded in
2023.
Its founder, Yang Zhilin, is an AI researcher who previously conducted PhD research at Carnegie Mellon University.
The company’s most well-known product is
Kimi.
And what truly brought Moonshot into the international AI competition spotlight this year is
Kimi K3.
Kimi K3 Took Moonshot to the Next Scale in July
Moonshot states that
Kimi K3 houses
2.8 trillion parameters.
The company claims it is the world’s largest
open-weight model.
Open-weight means
model weights are accessible,
developers can, under licensing terms,
download,
deploy,
modify,
and create their own services.
This differs from models only callable through closed APIs.
However, after Kimi K3 launched,
Moonshot quickly faced a reality:
demand exceeded capacity.
SasaDaily Reported in July: Kimi Was So Popular New Subscriptions Were Temporarily Paused
When Kimi K3 was released,
demand rapidly approached the company’s computing limits.
Moonshot temporarily paused accepting some new consumer subscriptions to prioritize existing users.
This already reveals that
AI model competition is not
“Model done means it’s over.”
You also need
GPUs,
data centers,
cloud infrastructure,
network,
inference capacity,
and
funding.
The more popular the model,
the bigger these challenges become.
Today’s $3 Billion IPO Raise Shouldn’t Be Seen Just as “An AI Company Wants More Money”
Moonshot’s real challenge might be that
turning Kimi K3 into a global commercial service requires massive capital.
Reuters reports Kimi K3’s demand is already putting pressure on Moonshot’s own computing resources.
The company has two paths:
First,
build more computing capacity themselves.
Second,
host the model on the world’s largest cloud platforms.
Moonshot appears to be pursuing
both simultaneously.
The More Crucial Direction: Moonshot Is Negotiating with Microsoft, Amazon, and Google
This is not the first time we hear this.
Reuters reported on August 26 that
Moonshot is
talking with
Microsoft,
Amazon,
and Google
to establish revenue-sharing agreements for Kimi K3.
The goal is to make Kimi K3 available as a service through
Azure,
AWS,
and Google Cloud.
Negotiations are still in early stages,
with no guarantee of signing.
But if successful, Reuters notes
it could be the first major revenue-sharing collaboration between a Chinese AI company and leading US cloud providers.
Why Is This More Important Than Just “Kimi on Three Major Clouds”?
Because Moonshot reportedly doesn’t just want to receive
fixed licensing fees,
but
revenue shares.
Insiders told Reuters that Moonshot aims for a share as high as
30%
of the service revenue generated by Kimi K3.
This means enterprises using Kimi K3 through
Azure,
AWS,
or Google Cloud
would pay Moonshot a portion of actual income.
This concept reshapes the commercial model for open-weight AI in ways people hadn’t expected.
Downloading the Model Doesn’t Mean Enterprises Will Run It Themselves
This is the most important insight to understand here.
In theory,
Kimi K3 is an open-weight model.
Enterprises can
download it,
deploy it,
and manage it themselves.
But it has
2.8 trillion parameters.
Models at this scale require enormous
GPU capacity,
memory,
network bandwidth,
inference infrastructure,
and maintenance capabilities
to run properly.
Therefore, Reuters quotes analysts saying that the number of clients who will actually host such large-scale models internally
may be limited.
This Leads to an Interesting Business Model
The model itself is
open,
but
the infrastructure to run it remains very costly.
So enterprises may ultimately choose
Azure,
AWS,
or Google Cloud
instead of buying massive GPUs themselves,
handling the maintenance,
and managing global capacity.
Here,
the cloud becomes the critical distribution gateway.
Model companies can say
“Our models are downloadable,”
but if a large number of enterprises use them via cloud,
they want to share the revenues.
Why “Open-Weight” Doesn’t Mean “No Business Moat”
Many people upon seeing
open weights
first think,
“If weights are given away, how is there money to be made?”
The answer might be
scale.
Small teams can
download for experiments.
But large enterprises need
reliable service,
security,
global deployment,
SLAs,
billing,
monitoring,
massive computational power.
All these
still require commercial platforms.
Models can be open,
but enterprise-level operation may not be.
Alibaba Is Also Taking a Similar Path
SasaDaily wrote on August 7 that
Alibaba plans to share revenue with some large Qwen commercial users
once significant revenue is generated.
The key point then was that
open-weight AI is entering an era of
free model availability paired with revenue sharing upon scale.
Moonshot has taken this a step further.
It is no longer just negotiating with large direct customers,
but with
the world’s largest cloud platforms.
This makes the business model of open-weight AI more complete.
Free Access to Models First, Monetize at Scale
This model can be imagined as:
Model weights
reduce initial usage barriers.
↓
Developers adopt and experiment.
↓
Model popularity increases.
↓
Enterprises need large-scale deployment.
↓
Clouds provide real execution capability.
↓
Enterprises pay per usage.
↓
Clouds and model companies
share revenue.
This contrasts with
fully closed models,
where every token must be paid for directly through the model company's API.
So Why Does Moonshot Need an IPO?
Because even if the business model works,
AI remains capital intensive.
Reuters reports that
Moonshot raised over
$2 billion
in May this year.
Total funding has exceeded
$5.5 billion.
Investors include
Alibaba,
Tencent,
IDG Capital,
HSG,
and other major Chinese institutions.
But as Kimi K3 demand grows,
more computing power is needed.
Developing new models also requires more capital.
So
private funding,
cloud partnerships,
and IPOs
are emerging simultaneously.
Moonshot’s Latest Valuation Is Reportedly $50 Billion
Reuters quotes two insiders stating that Moonshot’s ongoing latest funding round values the company at approximately
$50 billion.
This is a huge number,
but among China's cutting-edge AI firms,
it's not the largest.
Sources say DeepSeek's valuation is around
$74 billion,
and listed Z.AI is valued at about
$66 billion.
The market is beginning to establish
real capital market pricing for Chinese AI model companies.
This Reflects a Major Shift in the AI Industry in 2026
In previous years, comparisons focused on:
benchmarks,
context windows,
parameters,
and token prices.
Now there is another leaderboard:
valuations,
revenues,
fundraising amounts,
cloud contracts,
and the ability to go public.
AI competition has officially shifted from
research capabilities
to
capital capabilities.
Larger models require staggering capital to survive to the next generation.
Hong Kong Is Becoming a Key Capital Outlet for Chinese AI Companies
Moonshot is not the first.
This year,
Z.AI
and MiniMax
have already listed in Hong Kong.
Reuters cites LSEG data saying that as of mid-August,
Hong Kong IPO fundraising this year has reached
$41.2 billion,
an increase of
142% year-over-year.
Chinese tech companies account for a large portion.
If Moonshot completes the IPO,
it will join not only
the AI IPO wave,
but also
a trend of Chinese tech firms leveraging Hong Kong’s capital markets again.
Moonshot Is Also Restructuring Before Listing in Hong Kong
Reuters reveals another detail:
To obtain Chinese regulatory approval,
Moonshot must adjust its original
Red-chip structure,
meaning the original offshore company setup.
Two insiders say the company needs to reorganize into
a China-registered entity
before IPO application.
This reminds us that for AI companies to go public,
it’s not just about
strong models,
sufficient revenue,
and investor interest,
but also handling
regulation,
corporate structure,
cross-border capital flows,
data governance,
and international politics.
Moonshot Faces Another Big Uncertainty: US-China AI Politics
This issue might directly impact
Microsoft,
Amazon,
and Google’s ability to cooperate smoothly with Moonshot.
Moonshot has faced criticism from some US officials in recent months.
US Treasury Secretary Scott Bessent reportedly said
Moonshot might be added to the
trade blacklist.
US officials have accused Moonshot of
illegally obtaining restricted Nvidia chips
and using Anthropic’s models for distillation.
Moonshot denies that Kimi K3’s capabilities come from distillation,
claiming the model improvements derive from its own architectural innovations.
These controversies remain unresolved.
Negotiations with Microsoft, AWS, and Google Are Very Sensitive
If it were only
a Chinese model company cooperating with Chinese cloud services,
it would be less surprising.
Now, one of China’s most closely watched large models
wants to enter
the three most important US cloud infrastructures.
On one side are
business demands,
enterprises wanting to use the model,
and clouds wanting to expand their model offerings.
On the other side are
US-China chip restrictions,
national security concerns,
model origins,
data governance,
and political issues.
So this cooperation,
while commercially sensible,
is politically very complex.
This Suggests the AI Model Market Won’t Be a “Winner Takes All” Like Operating Systems
If enterprises go to Azure,
they won’t be limited to using only
Microsoft or OpenAI models.
If they go to AWS,
they won’t have to use just Amazon’s models.
The true cloud marketplace could include
OpenAI,
Anthropic,
Google,
Meta,
Qwen,
Kimi,
and other open models
all accessible.
Model company competition would then focus on:
who can get more clouds to offer their models,
and
how much revenue share they get per usage.
What Does This Mean for Regular Users?
You might not buy AI stocks
or care about IPOs.
But this will influence
how AI is priced in the future.
If large open models become easier to use through
Azure,
AWS,
and Google Cloud,
enterprises won’t need to overhaul their entire infrastructure for switching models.
Today it’s Kimi,
tomorrow Qwen,
and the day after, other models
all accessed through the same cloud gateway.
AI models may increasingly become
interchangeable engines within the cloud.
This will increase price competition among models.
But For Model Companies, Competition Will Be Even More Intense
If models are easy to swap,
easy to download,
and easy to use via cloud,
the real customer retention factors won’t be
“I was number one in benchmarks,”
but rather
quality,
cost,
stability,
enterprise support,
inference efficiency,
tool ecosystems,
and
cloud distribution capabilities.
In other words,
the model itself is only part of the product.
Moonshot Is At a Key Turning Point
July:
Kimi K3’s story was:
Model too popular, insufficient computing power.
August:
The story became:
Starting revenue-sharing talks with top three US cloud providers.
September 3:
The story moves forward:
Reportedly secretly filed for a Hong Kong IPO aiming to raise $3 billion.
Connecting these events reveals a clear trajectory.
Model demand
leads to computing power shortages,
which creates the need for cloud,
cloud commercialization requires revenue sharing,
and global expansion demands more capital,
ultimately leading to
an IPO.
The Real Takeaway Tonight Isn’t Whether Moonshot Is Worth $50 Billion
That’s a matter for the investment market.
More importantly,
an open-weight AI company is trying to prove
open models do not prevent building large-scale business models.
Weights allow broader adoption,
cloud provides execution,
enterprises pay per use,
model companies draw revenue,
and capital markets provide funding for the next compute cycle.
If this cycle works,
the future competition in open-weight AI won’t be just
“who is more open,”
but
who can convert openness into scalable revenues.
A Key Quote to Remember Tonight
AI models may be free to download,
but
that doesn’t mean AI companies have no business model.
The truly costly parts usually lie in:
deployment,
compute power,
stable services,
and enterprise distribution.
Moonshot is now trying to turn
Kimi K3’s model popularity
into
cloud revenue,
and then cloud revenue
into
IPO capital.
This is the real turning point tonight:
open-weight AI models are no longer just competing with closed models on capabilities, but also on who can build bigger business cycles.
Today, progress a bit with AI.
Learn one AI skill a day.
Save a bit of time every day.
Improve a bit every day.
SasaDaily, growing with you.
