AI Morning Briefing｜2026/09/04: Crusoe Reportedly Secures $13B Cloud Contract with Jane Street, Snowflake AI Drives…｜Article

Three AI news stories today answer one question: Is the AI boom turning into real business? Crusoe reportedly signed a $13B five-year AI cloud contract with Jane Street; Snowflake’s Q2 product revenue rose 37%, with AI fueling half its recent growth acceleration; meanwhile, Texas halts some new…

Over the past two years,

the most common question in the AI industry has been:

“Everyone’s spending so much—but is anyone really using it?”

Data centers keep expanding.

More GPUs are being bought.

AI company valuations rise higher and higher.

But the real test remains the same:

Are customers really willing to pay?

Today’s three news stories

break down the answer into three layers.

First:

A financial company reportedly willing to spend about $13 billion

on AI Cloud over five years.

Second:

Snowflake starts to see AI driving growth in its core platform in official earnings.

Third:

Demand is so great that

Texas is now asking,

“Can our power grid, water supply, and residents really handle so many data centers?”

So the true common theme today isn’t just:

AI is still growing.

But rather:

The AI boom is now simultaneously undergoing testing for “commercial demand” and “physical costs.”

First｜Crusoe Reportedly Signs $13 Billion AI Cloud Contract with Jane Street

The first news item

turns AI computing demand into a big number.

Reuters on September 3, citing Bloomberg News, reported

that AI infrastructure company:

Crusoe

has reportedly signed a

five-year AI Cloud contract

with financial trading firm:

Jane Street Group.

The contract is valued at around:

$13 billion.

Two words to note:

“Reportedly.”

This contract information currently comes from Bloomberg citing insiders,

and Reuters further reported on it.

At the time of Reuters' release,

Crusoe did not immediately respond to comment requests.

So we can’t say:

Both parties have officially publicly disclosed all contract details.

What exactly does Crusoe do?

It’s not an AI app like ChatGPT aimed at general users.

Crusoe is more like an:

AI infrastructure provider.

It supplies enterprises with:

data centers,

GPUs,

AI Cloud,

and the compute power needed to train and run large AI models.

Companies like these are now often called:

Neocloud providers.

You can understand it as:

a new type of Cloud Provider born specifically for AI high-performance computing.

Unlike AWS, Azure, or Google Cloud that offer various services,

Crusoe focuses more on:

GPUs,

high-speed networking,

AI training,

inference,

and large compute clusters.

Crusoe wasn’t originally focused on AI

The company was founded in 2018,

initially involved in crypto-related businesses.

It later gradually shifted towards:

AI data centers

and AI Cloud.

This shift perfectly reflects industry changes over recent years.

Earlier, the focus was on:

crypto mining.

Now, the scarcest compute demand is for:

AI.

So those who originally understood:

electricity,

data centers,

and high-density computing

have mostly moved into AI infrastructure.

How big is $13 billion?

If Bloomberg’s report holds,

this isn’t:

Jane Street trialing an AI SaaS.

Nor is it:

buying thousands of chatbot accounts.

It represents a

five-year:

massive computing capacity purchase.

Spread evenly,

$13 billion ÷ 5 years,

means:

about $2.6 billion per year.

Of course,

actual payment may not be evenly distributed.

The contract might also be structured around capacity, equipment, build progress, and other terms.

So $2.6 billion isn’t necessarily the annual billing.

But it at least signals to the market that

AI infrastructure has started to see:

long-term enterprise customers worth billions of dollars.

Jane Street is not an AI model company

That’s even more noteworthy.

Jane Street is a large financial trading firm.

Meaning:

AI computing demand no longer comes only from:

OpenAI,

Anthropic,

Meta,

Google—

companies that train large models themselves.

Non-AI-native enterprises may also need very large amounts of:

compute capacity,

data processing,

model development,

research,

and internal AI systems.

But bear in mind,

public reports don’t specify

how Jane Street’s $13 billion compute will be used—

which trading models,

financial decisions, or strategies.

So it’s incorrect to say:

“Jane Street spent $13 billion on fully automated AI trading.”

What we truly know is:

it reportedly became a very important large cloud customer for Crusoe.

Crusoe also raised more than $3 billion

The same report included another number.

According to Bloomberg,

Crusoe’s latest funding round raised:

over $3 billion

with a post-money valuation of around:

$30 billion.

“Post-money” means

company valuation after the new investment.

This Jane Street mega contract

was reportedly a key factor attracting investors in this round.

This is easy to understand.

Investors’ biggest worry in AI infrastructure is:

spending billions to build data centers,

only to find

no one actually rents capacity.

If you already have large, long-term customers,

the financing story changes completely.

Crusoe announced an even bigger figure in June

In June, Crusoe officially said

contracted AI infrastructure capacity was nearly:

4.9 GW

meaning 4.9 gigawatts.

Its total project pipeline, including:

contracts signed,

tenants in negotiation,

and advanced development phase projects,

exceeds:

40 GW.

This doesn’t mean:

all 40 GW capacity is built now.

Pipeline is not formal revenue either.

But it shows you

that AI infrastructure companies now think in units of:

billions of watts of power and data center capacity

rather than just number of GPUs.

What really matters about the first story isn’t how rich Jane Street is

It’s that AI infrastructure is starting to resemble

traditional heavy industry.

For instance, airlines sign

long-term aircraft orders.

Manufacturers sign

long-term energy contracts.

Large companies sign

long-term cloud commitments.

AI is now entering

five-year, tens of billions of dollars compute procurements.

This moves AI infrastructure from:

“Everyone says demand will be huge someday”

to:

“There really are signed long-term contracts.”

Second｜Snowflake: AI is not just a feature but driving overall platform usage

The second story

takes us back from data centers to software.

After Snowflake released its Q2 earnings on September 2,

its stock surged nearly

25%

on September 3.

What really excited the market wasn’t:

another AI chatbot launch.

But that:

the company connected AI to actual

platform usage growth.

Snowflake Q2 product revenue grew 37% year-over-year

Snowflake’s Q2

product revenue increased by

37%

year-over-year.

The company also raised its FY2027 product revenue forecast from

$5.84 billion

$6.07 billion.

What’s really notable is CEO Sridhar Ramaswamy’s statement.

He said:

AI offerings contributed about

half of the recent growth acceleration.

Note the exact meaning here.

It’s not that

half of Snowflake’s revenue solely comes from AI,

nor that the entire 37% growth was AI-driven.

Rather, the company believes:

AI accounts for about half of the recent speed-up in growth.

This distinction is important.

Why is Snowflake’s case more meaningful than “AI apps selling well”?

Because Snowflake isn’t purely a generative AI company.

It’s core business

is enterprise data cloud.

Enterprises put their

data,

analytics,

applications,

and workloads

on Snowflake.

With AI emerging,

if companies start to:

use more data agents,

write code with coding agents,

let employees query enterprise knowledge,

and generate many more queries via AI,

then AI-driven revenue might not only appear as

“AI product subscription fees”

but also increase

overall platform consumption.

This is the kind of result enterprise software companies want to see.

Cortex Code surpasses 9,100 accounts

Snowflake’s AI coding assistant,

Cortex Code,

has more than

9,100 accounts,

adding over

2,000 new users in one quarter.

Another enterprise knowledge work agent,

CoWork,

has expanded to

5,800 accounts.

The real significance of these numbers isn’t

“how many have tried it,”

but that AI features are increasingly integrating

more daily work into the Snowflake platform.

The true business value of enterprise AI might not be just another “AI fee”

This is a challenge many SaaS companies face.

Suppose you originally sell:

databases,

CRM,

collaboration tools,

or design software.

After adding AI,

there are two ways to monetize:

First:

Charge extra AI fees.

For example:

a monthly per-user AI charge.

Second:

Let AI drive users to use more of the core service.

For instance:

query more data,

run more queries,

build more applications,

and perform more computations.

Snowflake seems to see both possibilities,

but what markets should note is that

AI is starting to drive core platform consumption.

This contrasts with skepticism from a few years ago

Investors used to ask enterprise software companies,

“If you add AI,

does it only increase GPU costs?”

“Every query calls an external model,”

“gross margins decline,”

“but customers are reluctant to pay more.”

If that’s the case,

AI might just:

eat into software margins.

Snowflake now presents a different possibility:

If AI drives customers to use

more data,

more cloud,

and more core functions,

then AI isn’t just a cost—it can become a

usage growth engine.

But we can’t yet say “enterprise AI ROI is fully proven”

Snowflake’s performance reflects one company’s results.

The AI contribution is management’s view.

Different enterprise software firms have

different pricing,

customers,

AI costs,

and usage rates.

So we can’t generalize from Snowflake’s strong quarter

to all SaaS companies making more money post-AI.

But at least we have one important new piece of evidence now:

AI is appearing not only in demos and product announcements but also in explanations of official revenue growth.

Third｜Texas once aimed to be an AI hub but now slows data center growth

The first two news items tell us:

Demand is real.

Customers are willing to pay.

Software companies start to see revenue.

What happens next?

The third story answers that:

The real world begins to bear these demands.

Reuters reported on September 3 that

less than a year ago,

Texas Governor Greg Abbott called Texas

an important AI development hub.

Now he advocates for:

more restrictions on big data centers,

tax incentive cuts,

and limits on some rural area development.

And Texas has effectively paused new grid connection approvals for data centers

Abbott launched a

state audit on August 3.

Reuters noted this review has

effectively suspended new data center grid access approvals.

This does not mean:

Texas bans all data centers,

nor that existing centers shut down.

The real issue:

New large projects seeking

grid connection

must now verify that:

the demand is real,

where power comes from,

who pays for infrastructure,

and whether other residents will be affected.

Why does Texas, usually pro-energy and pro-business, take this stance?

Because data center demand is growing too fast.

Residents now see firsthand:

electric bills,

water usage,

land use,

nighttime lighting,

and infrastructure impacts,

all external costs.

Reuters quoted local opponents saying

large data centers may

drive up electricity and water costs.

Meanwhile, local governments provide massive incentives,

but actual long-term jobs and tax revenue created

may not meet residents’ expectations.

So the debate has shifted from

“Should we develop AI?”

“Who should bear AI’s costs?”

Only 30% of Texans support building data centers in their communities

A Texas Politics Project August survey shows

only

30%

support building data centers in their own communities.

This is why the issue is rapidly politicizing,

and it’s not just one party.

Reuters points out that Democrats in many US states have taken tougher stances locally against data centers,

and some Republicans are now following suit.

This turns data center policies into

a new political battleground for AI expansion in the US.

Texas even diverges from federal policy and Trump’s approach

President Donald Trump’s policy direction has broadly emphasized

accelerating AI infrastructure

to keep the US ahead in AI.

But some Texas Republicans now argue

local costs must be controlled as well.

This creates a classic contradiction.

At the national level, the goal is

to build more

and faster

to compete with China.

But locally, residents ask:

“Who pays the electricity bill?”

“Is there enough water?”

“What do we get out of it?”

Neither side is entirely wrong;

they simply represent

different costs and interests.

And here’s another problem: some data center demand might be fake

Another Reuters investigation on September 1 mentioned that

US utilities have received data center grid connection requests totaling

more than 700 GW.

This is an extremely high figure,

far exceeding the actual electricity use of all US data centers.

One issue:

the same development project may simultaneously submit

requests to different locations

and utilities,

but only build one.

This is called:

Ghost Demand.

False or phantom demand.

Why is Ghost Demand troublesome?

Assuming a utility receives 10 data center applications,

each claiming they need

1 GW.

The utility sees a total of 10 GW new demand,

and may begin planning

power plants,

substations,

and transmission lines.

If in the end, only

2 centers are built and

8 disappear,

the planned infrastructure may be

excessive,

and these costs may be

passed to customers via

electricity rates and infrastructure investments.

So Texas’ current audit isn’t just

“We don’t like AI.”

It’s also

“Prove your data center will really be built and can pay the bills.”

This makes today’s first story even more interesting

Crusoe said it has

4.9 GW contracted capacity,

and more than

40 GW in the pipeline.

Today there’s news of

Jane Street’s reported $13 billion long-term cloud contract.

Why are contracts like this important?

Because when the data center industry starts facing

ghost demand and skepticism,

simply saying “I want to build a huge data center”

isn’t enough.

Governments,

banks,

utilities,

and investors

all begin to ask:

“Who is the customer?”

“Where’s the contract?”

“Who will pay the power bill?”

“Is this real demand or just hold a spot?”

Large long-term contracts become crucial

as evidence for infrastructure financing.

Therefore, these three stories form a complete AI business chain

The first element:

Real enterprise demand.

Jane Street reportedly willing to sign a large five-year computing contract.

↓

The second element:

Actual software usage and revenue.

Snowflake shows AI is increasing core platform consumption.

↓

The third element:

Real physical costs.

Data centers need:

power,

water,

land,

transmission,

and community acceptance.

Together they create a concrete industrial chain:

AI usage

↓

Cloud usage

↓

Data centers

↓

Electricity

↓

Infrastructure

↓

Resident costs and politics

This is what AI must face after

evolving from

an internet product

real infrastructure.

The problem of AI commercialization has moved beyond “Will anyone pay?”

Early on, people asked:

Can AI really make money?

Today we actually have some partial answers.

Some are willing to sign billion-dollar cloud contracts.

Snowflake says AI is driving platform growth.

But the next question immediately emerges:

If every company really starts using AI, do we have enough compute and energy to support it?

Thus, AI’s biggest risk may no longer be

insufficient demand,

but

too much demand too fast.

What does this mean for the average person?

You probably won’t buy Snowflake stock.

Or rent $13 billion worth of cloud capacity.

But these three stories ultimately affect the

cost of your AI usage.

If sufficient compute is available,

data centers get built smoothly,

and software companies have plenty of paying customers,

AI is generally more likely to

reduce prices

and enhance features.

If instead,

power is tight,

local limits increase,

build costs rise,

and compute demand explodes,

then these costs may reflect in

cloud fees,

API prices,

enterprise software costs,

or even electricity and public infrastructure expenses.

AI isn’t some magic floating in the cloud.

Under the cloud are still land, chips, data centers, and power.

What really matters to companies: don’t just focus on “Do we have AI?”

Today’s Snowflake news

offers a simple reminder for all smaller companies.

Don’t just count:

how many AI queries employees made this month,

or how many agents were launched,

or how many logged in.

Start tracking whether this AI usage

delivers more completed work,

more customers,

less manual labor time,

lower costs,

higher usage volume,

and real revenue.

Because now, even large publicly listed companies

need to use

actual revenue

to validate their AI initiatives.

Small companies shouldn’t rely only on

“Everyone says it’s convenient.”

The same goes for AI infrastructure investment

Seeing numbers like

40 GW pipeline,

hundreds of billions in investments,

and mega data centers,

it’s easy to think:

The bigger the number,

the stronger the company.

But what really matters is:

How much is already contracted?

Who are the customers?

Can customers keep paying?

Can the power actually be connected?

Is land and water available?

Will the government approve?

If it can’t be powered up in the end,

no matter how nice the data center renderings,

they’re just

pictures.

Today’s three stories bring AI back to a traditional business problem

In the end, a business needs three things to succeed:

Someone to buy.

Crusoe’s large contracts,

if confirmed,

prove demand.

Someone to keep using.

Snowflake’s usage and revenue growth provide proof of ongoing usage.

Capacity to supply.

Texas data center controversies remind us that

AI isn’t infinitely supplyable.

So a healthy AI industry

needs more than just

demand growth.

It also needs:

supply,

infrastructure,

and economics

to keep up.

The takeaway phrase today

Whether the AI bubble is truly a bubble can’t be judged just by

company valuations

GPU orders.

You have to look deeper at

whether real customers sign real contracts,

whether AI truly drives revenue,

and whether the power and infrastructure can really support it.

Today’s three stories

happen to line up those three layers.

AI is moving from

“believing the future will be big”

“now please show contracts, revenue, and power.”

Today, make a little progress with AI.

Learn one AI skill each day.

Save a little time every day.

Build your abilities bit by bit.

SasaDaily, growing with you.