A 2.45-GIGAWATT DATA CENTER IS BIG ENOUGH TO BECOME A BUSINESS STORY BY ITSELF
Oracle's Project Jupiter is being built to consume as much electricity as roughly 1.8 million homes at any given moment. At that scale, the data center stops looking like a building and starts looking like an entire industrial business.
THIS IS NOT JUST A DATA CENTER
Project Jupiter is a planned AI data center campus in Doña Ana County, New Mexico, being developed as part of the broader Stargate infrastructure build-out involving Oracle and OpenAI.
The number that makes the project difficult to ignore is 2.45 gigawatts.
That is the planned installed power capacity for the campus after Oracle changed its energy design to use Bloom Energy fuel cells. Bloomberg reported that the 2.45-GW campus would require enough electricity to power roughly 1.8 million homes at any given moment.
The physical project is enormous too.
The campus covers about 1,400 acres and is being developed as a hyperscale AI computing site.
At that point, calling it a “data center” starts to undersell what is actually being built.
It requires:
→ land
→ buildings
→ electricity
→ fuel infrastructure
→ cooling
→ networking
→ financing
→ cloud customers
→ long-term contracts
→ local tax agreements
→ regulatory approvals
The computer servers are only one part of the business.
THE ELECTRICITY BILL BECOMES PART OF THE PRODUCT
Traditional data centers buy electricity from the grid.
AI data centers are changing that relationship.
Large AI clusters need enormous amounts of continuous power, and waiting for a conventional grid connection can become one of the biggest constraints on construction.
Project Jupiter's original energy plan relied on natural-gas turbines and diesel generators.
Oracle later redesigned the project around Bloom Energy fuel cells, with up to 2.45 GW of installed capacity planned for the campus. Oracle said the new design would replace the previously planned turbines and diesel generators with a single onsite microgrid.
That decision illustrates a major change in data-center economics:
Power is no longer just an operating expense.
It is becoming part of the infrastructure that determines whether a computing business can exist in the first place.
AI HAS CREATED A POWER BOTTLENECK
The obvious assumption is:
AI demand rises → companies buy more GPUs → data centers are built.
The less obvious sequence is:
AI demand rises
↓
more computing capacity is required
↓
more data-center buildings are required
↓
more electricity is required
↓
existing grids become constraints
↓
developers search for dedicated power
↓
energy infrastructure becomes part of the AI business
That is why Project Jupiter involves much more than Oracle's cloud business.
The project is effectively combining real estate + energy + computing + finance.
THEN THE PIPELINE BECAME A BUSINESS PROBLEM
The fuel-cell strategy did not eliminate the energy problem.
It changed its shape.
Bloom's fuel cells still require natural gas.
An Energy Transfer pipeline intended to supply the project has been delayed nearly six months, with service now expected on February 1, 2027, after repeated permit problems involving the proposed route.
So the project has a surprisingly old-fashioned dependency underneath its futuristic AI infrastructure:
an enormous supply of natural gas.
The AI may be running on cutting-edge accelerators.
But the infrastructure supporting those accelerators still depends on pipes, permits and physical energy delivery.
THE PROJECT HAS BECOME A FINANCING MACHINE
There is another reason Jupiter is interesting.
The project isn't being paid for like a normal corporate construction project.
About $18 billion of loans were arranged to help finance the campus, with a consortium of roughly 20 banks involved.
That creates an entirely different layer of risk.
The chain looks something like:
OpenAI needs computing
↓
Oracle agrees to provide computing capacity
↓
Oracle needs enormous infrastructure
↓
developers build the campus
↓
banks finance construction
↓
Oracle becomes the major tenant
↓
future computing revenue is expected to support the economics
The data center therefore becomes a financial asset before it ever becomes a fully operational computing machine.
$18 BILLION OF DEBT CAN CARE ABOUT A PERMIT
This is where infrastructure starts behaving differently from software.
If a software company delays a feature by six months, the consequences can be manageable.
If a 2.45-GW data center cannot obtain the infrastructure needed to power itself, billions of dollars of financing can suddenly become more complicated.
Reuters reported last week that loans tied to Project Jupiter were trading around 89–91 cents on the dollar, amid investor concerns surrounding Oracle's borrowing and challenges facing the project.
The underlying issue is not simply whether AI demand exists.
Investors have to ask whether the physical asset can actually be delivered on the expected timeline.
That distinction matters.
AI demand can be enormous.
But demand does not automatically produce electricity.
ORACLE HAS NOW USED A FORCE-MAJEURE NOTICE
On September 24, Oracle sent the project's developer a force-majeure notice.
The move does not mean Oracle is abandoning Jupiter.
According to Bloomberg, the purpose was to preserve Oracle's contractual position if the project fails to meet its planned 2028 delivery date. The company could potentially delay certain payments under specified circumstances.
Oracle said the notice was a way of preserving contractual rights and that it remains committed to the project.
Blue Owl, whose unit is developing the project, similarly said the notice does not change the financial commitments to the multiyear project.
This is a fascinating detail because it shows how AI infrastructure contracts are becoming sophisticated financial instruments.
A data center lease isn't simply:
“We rent this building.”
It can involve:
→ construction deadlines
→ power-delivery conditions
→ financing arrangements
→ rent commencement
→ force-majeure provisions
→ customer commitments
→ infrastructure dependencies
Every delay can therefore move money between different participants.
FORCE MAJEURE IS BECOMING AN AI INFRASTRUCTURE TOOL
Force-majeure clauses are hardly new.
They are common in industries where companies depend on physical events they cannot completely control.
But AI data centers introduce an unusual combination of dependencies.
A project can be delayed by:
One missing component can affect the entire economic chain.
That makes contract language unusually valuable.
A clause that once looked like legal boilerplate can become a financial protection mechanism worth billions.
THE BUILDING ITSELF IS ONLY THE MIDDLE OF THE BUSINESS
Project Jupiter is also interesting because several companies can make money from the same physical site.
Consider the ecosystem:
Oracle
Needs computing capacity and is the planned tenant.
OpenAI
Needs enormous amounts of AI computing capacity.
Blue Owl / Stack Infrastructure
Develops and finances the physical infrastructure.
Bloom Energy
Supplies the fuel-cell technology for the onsite power system.
Energy Transfer
Provides the natural-gas infrastructure needed by the energy system.
Banks
Provide billions of dollars of financing.
Construction companies
Build the campus.
Equipment suppliers
Provide electrical, cooling, networking and other infrastructure.
One AI data center therefore becomes an economic ecosystem.
The server building is simply where all those businesses meet.
THE LOCAL ECONOMY HAS ALREADY STARTED CHANGING
Oracle says Project Jupiter had generated nearly $80 million in state and county tax revenue by July 2026 and had nearly 700 New Mexico residents working onsite.
The company projects more than 7,000 construction jobs and around 1,500 ongoing project-supported jobs once the project is complete.
Local economic-development officials have also described companies moving into nearby industrial space to support the data-center construction boom.
That includes suppliers and contractors involved in HVAC, electrical systems, mechanical construction and data communications.
So another business effect appears:
the data center creates a secondary data-center economy.
A company does not necessarily need to operate the data center to benefit from its construction.
It can sell something to the companies building it.
DATA CENTERS ARE STARTING TO LOOK LIKE INDUSTRIAL PARKS
This is one of the less obvious consequences of AI.
A cloud data center used to be thought of primarily as a technology facility.
The new generation looks more like an industrial project.
It needs:
→ industrial land
→ power plants or dedicated generation
→ water systems
→ roads
→ substations
→ gas pipelines
→ specialized contractors
→ enormous equipment deliveries
→ financing structures
→ local incentives
That means data-center development increasingly competes for the same resources that manufacturing plants, warehouses and other heavy infrastructure projects need.
The difference is what happens inside.
Instead of making physical products, the facility produces something much harder to see:
computing capacity.
THE WATER STORY IS JUST AS IMPORTANT
New Mexico is a particularly interesting location because water scarcity has become one of the central issues surrounding the project.
Oracle changed Jupiter's energy design partly to reduce its water requirements.
The company says the combined cooling and fuel-cell systems are expected to use roughly as much water annually as nine U.S. households, averaged over 15 years, and that the project will not use potable water for cooling or fuel-cell operation.
Oracle has also committed $50 million to local water infrastructure.
But the project has still faced legal and community opposition related to water and environmental concerns.
This creates a fundamental business problem:
A data center may be economically valuable to a region while simultaneously putting pressure on infrastructure that the region considers scarce.
The economic value and the infrastructure cost can exist at the same time.
ORACLE IS SELLING MORE THAN COMPUTE
Oracle's argument for Jupiter is therefore broader than:
“We are building a big computer.”
The company says the project could generate more than $4.7 billion in long-term economic impact for New Mexico through tax revenue, investment and additional economic activity.
Its projected economic effects include construction activity, operational activity, tax revenue, industrial-revenue-bond payments and community investments.
That matters because large data centers increasingly have to justify themselves as infrastructure projects.
The pitch becomes:
AI infrastructure → investment → jobs → tax revenue → local development
rather than simply:
AI infrastructure → servers
THE CUSTOMER IS REALLY BUYING TIME
There is another unusual business concept hiding inside Jupiter.
OpenAI and other AI companies do not necessarily care about owning the physical building.
They care about having enough computing capacity available when they need it.
That means the scarce product is not really square footage.
It is available compute at the right time.
A data center that opens six months late can therefore be economically different from one that opens on schedule.
If demand is waiting for the capacity, every month of delay potentially means computing capacity that cannot yet generate the expected revenue.
This explains why construction dates, power permits and lease clauses can become as important as GPU specifications.
THE DATA CENTER IS BECOMING A LONG-TERM CONTRACT
Project Jupiter is connected to Oracle's broader relationship with OpenAI.
That changes the business model again.
Instead of building an empty facility and hoping customers arrive, hyperscale infrastructure can be built around expected long-term demand.
The logic is:
customer demand
↓
long-term computing agreement
↓
infrastructure financing
↓
data-center construction
↓
power infrastructure
↓
computing capacity
↓
customer revenue
The contract helps justify the infrastructure.
The infrastructure makes the contract possible.
And the financing sits between them.
That is much closer to project finance than traditional software economics.
THE BIGGER THE DATA CENTER, THE MORE THINGS CAN GO WRONG
Scale creates efficiency.
But scale also creates dependencies.
A small server facility might be able to work around a local problem.
A 2.45-GW campus cannot simply plug into another wall outlet.
If its energy supply is delayed, the problem can spread across:
construction
→ financing
→ equipment schedules
→ lease obligations
→ customer commitments
→ revenue expectations
This is why Project Jupiter's recent problems matter beyond one Oracle campus.
It is a case study in what happens when software-scale ambitions collide with physical infrastructure.
AI HAS BECOME A REAL-ESTATE BUSINESS
For years, the biggest AI stories were about models.
Then they became about GPUs.
Now another layer is becoming impossible to ignore:
real estate.
Companies need land on which to put the computers.
They need electricity to operate them.
They need cooling systems.
They need fiber connections.
They need financing.
They need permits.
They need communities willing to host them.
The AI industry is therefore pulling technology companies deeper into businesses they traditionally did not look like.
Oracle isn't just selling cloud software anymore.
Its AI strategy requires something that looks remarkably like an energy and infrastructure company.
AND THAT CHANGES WHO GETS PAID
Every dollar spent building Jupiter can flow through multiple businesses.
A simplified economic chain looks like:
AI customer
↓ pays for computing
Oracle
↓ commits capital to infrastructure
Developer / infrastructure company
↓ pays construction and equipment suppliers
Energy system
↓ purchases fuel and equipment
Banks
↓ finance construction
Local economy
↓ receives wages, taxes and contracts
The AI application might be invisible to someone driving past the site.
The economic footprint is not.
THE STRANGE THING ABOUT JUPITER
The project is supposed to make artificial intelligence faster and more available.
But its biggest problems are extraordinarily physical.
A pipeline.
A permit.
A power system.
A construction schedule.
A lease.
A loan.
A local government.
A community.
This is the part of the AI boom that is easy to miss when attention stays focused on chatbots and models.
AI may be digital.
Its infrastructure isn't.
MAACAT PERSPECTIVE
Project Jupiter shows what happens when computing becomes large enough to behave like heavy industry.
At 2.45 GW, the data center needs its own energy strategy.
At 1,400 acres, it becomes a land-development project.
With $18 billion of debt, it becomes a financing story.
With long-term AI customers, it becomes a contractual business.
With thousands of workers and billions in projected economic activity, it becomes a regional economic story.
And with pipelines, permits and water infrastructure involved, it becomes a political and regulatory story too.
The interesting business equation is no longer simply:
AI → software
It is becoming:
AI demand
↓
compute
↓
data centers
↓
power
↓
infrastructure finance
↓
real estate
↓
local economies
A 2.45-gigawatt data center is therefore not merely a place where computers run.
It is a business ecosystem large enough to become the story itself.
