Project Stargate is a private-sector AI infrastructure initiative announced at the White House on January 21, 2025. OpenAI, SoftBank, Oracle and MGX said they intended to invest up to $500 billion over four years in U.S. AI infrastructure, with an eventual target of 10 gigawatts of capacity.
That headline does not mean $500 billion has already been raised or spent. By August 2026, Stargate had become a portfolio of data-center, cloud, power and financing partnerships. Construction and early operations were advancing at several locations, while individual plans—including a proposed expansion at the flagship Abilene, Texas, campus—had changed.
What Project Stargate is—and is not
The original announcement described Stargate as an initiative intended to deploy up to $500 billion over four years, beginning with an initial $100 billion deployment, to build AI infrastructure in the United States. The announcement also set a goal of 10 gigawatts of U.S. infrastructure by 2029. OpenAI’s announcement and SoftBank’s announcement did not say that the federal government would directly fund the $500 billion.
Stargate is now better understood as an umbrella platform rather than one conventional joint venture building one giant facility. OpenAI’s later description includes Oracle, SoftBank-related entities, CoreWeave, energy companies, data-center developers, chip suppliers and other partners. Different sites can have different owners, landlords, financiers, operators and tenants.
Free tools Windows power users keep installed
One-click scans. No signup required.
- Investment ambition: up to $500 billion over four years, as announced in 2025.
- Capacity ambition: 10 gigawatts of U.S. AI infrastructure, targeted for 2029.
- Platform: a changing portfolio of campuses, cloud contracts, power projects and equipment commitments.
- Operational reality: announced capacity is not the same as energized data centers, installed chips or usable compute.
Who is involved?
| Participant | Role | Important qualification |
|---|---|---|
| OpenAI | Primary AI customer and user of training and inference capacity. | It supplies demand and enters capacity arrangements; it is not the owner of every Stargate facility. |
| SoftBank | Initial equity funder and financial sponsor; Masayoshi Son was a central figure in the launch. | SoftBank-related companies, including SB Energy, also participate in power and site development. |
| Oracle | Oracle Cloud Infrastructure and data-center partner, including work connected with Abilene. | Oracle is not the sole owner or financier of all Stargate sites. See Oracle’s data-center overview. |
| MGX | Abu Dhabi-backed investment firm and one of the original equity funders. | MGX is often omitted from shortened headlines, but was named in the original announcements. |
| Nvidia | Principal accelerator supplier for initial deployments. | Data-center capacity, installed GPUs and available compute are separate measures. |
| Crusoe and Lancium | Development and infrastructure roles at the Abilene campus. | Their participation does not make them owners of the entire Stargate platform. |
| SB Energy | Developer of the Milam County, Texas, project. | OpenAI says the site is designed to minimize water use and add generation; those are company claims. |
| CoreWeave and other providers | Part of OpenAI’s broader compute portfolio. | Not every OpenAI infrastructure contract is a single Stargate joint venture. |
What does the $500 billion figure mean?
The wording matters. The original release said Stargate intends to invest $500 billion over four years. That establishes a long-term target or investment framework, not proof that the money was already deposited, borrowed, spent or contractually committed in full.
A buildout of this scale could combine several forms of capital:
- Equity: sponsor capital from entities such as SoftBank and MGX.
- Debt and project finance: loans, bonds or infrastructure-level borrowing.
- Customer commitments: OpenAI or another tenant agreeing to buy or lease capacity.
- Developer capital: spending on land, buildings, substations and generation.
- Equipment spending: accelerators, networking, storage, cooling and backup systems.
- Operating costs: electricity, maintenance, staff, financing and replacement hardware.
Those categories can overlap in a project’s reported value. Until audited financial disclosures itemize the commitments, the accurate description is: the companies announced an intention to invest up to $500 billion over four years.
How large is the planned infrastructure?
In July 2025, OpenAI and Oracle announced an additional 4.5 gigawatts of capacity and said Stargate capacity under development had passed 5 gigawatts, with more than two million chips planned. That is a forward-looking company estimate, not a count of chips already operating. Read the Oracle partnership announcement.
Rank #2
In September 2025, OpenAI and SoftBank announced five additional sites or expansions that they said could deliver more than 5.5 gigawatts and put the initiative ahead of schedule toward its original commitment. The figure referred to announced or planned capacity, not necessarily commissioned computing capacity. Sources: OpenAI and SoftBank.
These distinctions prevent common errors:
- Power capacity is not actual electricity consumption.
- A powered building is not necessarily filled with GPUs.
- Installed GPUs are not all available to OpenAI at every moment.
- Planned gigawatts are not operational gigawatts.
Where the projects stand
| Site or area | Location | Announced capacity or description | Status and caveat |
|---|---|---|---|
| Abilene flagship | Texas | Figures vary by construction phase and announcement. | Partly operational and under construction; a proposed expansion changed in 2026. |
| Milam County | Texas | 1.2 GW | SB Energy project under development or construction, with initial service expected in 2026 according to OpenAI and SB Energy. |
| Shackelford County | Texas | Announced in the 2025 five-site plan. | Announcement does not establish completed or operational capacity. |
| Doña Ana County | New Mexico | Announced in the 2025 five-site plan. | Water, power and construction status require site-specific qualification. |
| Saline Township campus | Michigan | Multi-billion-dollar campus announced with Oracle, Related Digital, Blackstone and Walbridge. | Construction was celebrated on June 1, 2026. See Oracle’s announcement. |
| Other Midwest and Ohio/Wisconsin projects | United States | Multiple projects and developers. | Development stages and ownership vary; capacity should not be added across overlapping announcements without a dated source. |
The Abilene expansion reversal
Abilene is the most visible Stargate campus, developed through a Crusoe, Lancium and Oracle ecosystem for OpenAI workloads. The existing campus continued, but Bloomberg reported on March 6, 2026, that Oracle and OpenAI ended plans to expand it in the previously discussed form after financing negotiations and changes in OpenAI’s capacity needs. Bloomberg’s report should not be read as saying the entire Texas project was canceled.
On March 27, 2026, the Associated Press reported that Microsoft was taking over construction plans for two nearby buildings and an on-site power plant. That illustrates how a redirected building can remain part of regional data-center capacity even when its original tenant or financing plan changes. Source: AP.
Why Stargate needs so much electricity
AI campuses require continuous, high-density power as well as substations, transmission connections, backup systems and advanced cooling. Developers may build generation specifically for a site or connect it to the wider grid. The answer affects permitting, reliability and who pays for upgrades.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsOpenAI and SB Energy said the Milam County project would minimize water use and add generation to support the campus and protect Texas ratepayers. Those are company statements, not independent proof that ratepayer impacts or water consumption have been eliminated. “Minimize water use” does not mean zero water use.
For any proposed campus, the financially important questions are:
- Who pays for interconnection, substations and transmission?
- Are generation assets dedicated to the campus or shared with the grid?
- Are tax abatements, grants or other public incentives involved?
- How much water is required for cooling and backup systems?
- Will local electricity prices, reliability or infrastructure spending change?
What technology is being built?
The infrastructure stack includes Nvidia accelerators, high-bandwidth networking, distributed storage, liquid cooling, Oracle Cloud Infrastructure and systems optimized for OpenAI model training and inference. Facilities must also accommodate successive accelerator generations; a building designed around one hardware cycle can become uneconomic if demand or chip economics change before it reaches full utilization.
OpenAI’s projection of more than two million chips across over five gigawatts of capacity is therefore best treated as a deployment plan, not an inventory of working systems.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minutePC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Why OpenAI wants its own infrastructure portfolio
Frontier-model training and inference require rapidly increasing compute. OpenAI has historically relied heavily on Microsoft Azure while adding Oracle, CoreWeave, SoftBank-related and other infrastructure relationships.
More dedicated capacity can give OpenAI greater control over hardware schedules, networking, deployment and operating economics, while reducing dependence on one cloud provider. The trade-off is substantial: long-term capacity contracts, leases and hardware commitments can become expensive if model demand, efficiency or product strategy changes.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Financial and commercial risks
Financing risk
A multi-site buildout may depend on a mix of equity, debt, leases, customer contracts and vendor financing. Higher interest costs or stricter lender requirements can delay construction even when demand remains strong.
Hardware obsolescence
Accelerators, networking equipment and cooling designs evolve quickly. A project can be physically complete yet require additional spending before it delivers competitive compute.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Demand and tenant risk
OpenAI’s capacity needs can change as models become more efficient or products grow faster or slower than expected. A campus designed for one tenant may need another customer if contracts are reduced, although data centers can sometimes be repurposed.
Concentration risk
Oracle and other providers may face exposure to a small number of large AI customers. OpenAI, meanwhile, carries long-duration obligations for capacity, power and equipment.
Jobs, incentives and local costs
Construction can create a large but temporary workforce. Permanent data-center employment is usually smaller and more specialized than the construction workforce. Claims about job totals should therefore identify whether they are forecasts, construction roles or permanent operating positions.
Local residents and investors should separately examine tax abatements, land deals, water infrastructure, noise, utility upgrades, backup generation and electricity-rate exposure. A private campus can produce regional economic benefits while still shifting some infrastructure costs to public entities or other utility customers; the answer depends on the specific project’s contracts and incentives.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →A timeline of major announcements
- January 21, 2025: OpenAI, SoftBank, Oracle and MGX announced Stargate, including the up-to-$500 billion, four-year intention and 10-GW target. Sources: OpenAI and SoftBank.
- July 2025: OpenAI and Oracle announced 4.5 GW of additional capacity and more than 5 GW under development.
- September 2025: OpenAI and SoftBank announced five sites or expansions totaling more than 5.5 GW of potential capacity.
- September 2025: SB Energy announced the 1.2-GW Milam County, Texas, partnership.
- March 6, 2026: Bloomberg reported that the proposed Abilene expansion had ended in its prior form.
- March 27, 2026: AP reported Microsoft’s takeover of nearby Abilene construction plans.
- April 29, 2026: OpenAI described Stargate as an infrastructure platform and reiterated a 2029 10-GW U.S. goal in its infrastructure update.
- June 1, 2026: Oracle and partners celebrated construction of the Saline Township, Michigan, campus.
How to judge future Stargate announcements
- Ask whether a dollar figure is targeted, financed, spent or audited.
- Separate announced capacity from energized and operational capacity.
- Identify who owns the land, buildings, GPUs, power assets and cloud contracts.
- Check whether capacity figures overlap with earlier announcements.
- Look for secured power, permits, interconnection and cooling plans.
- Review tenant commitments, debt terms and the ability to serve other customers.
- Separate temporary construction jobs from permanent employment.
- Identify tax abatements, grants and public grid or water costs.
The Bottom Line
Stargate is real, but the $500 billion figure is a long-term private-sector ambition attached to a changing portfolio of projects—not cash already spent on one fully funded data center. The most reliable measures are site-level financing, permits, energized power, installed hardware and workloads actually running.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
