The Stargate Project is a $500 billion, four-year plan to build AI data centers in the U.S. — with $100 billion committed immediately and a target of roughly 10 gigawatts of compute. That's the short answer. The longer answer is more interesting.
It is the largest private infrastructure commitment ever announced in tech — bigger than the Apollo program in inflation-adjusted dollars. But "announced" is doing a lot of work in that sentence. Most of the money does not exist yet, the power to run it has not been secured, and the partners do not fully agree on the timeline. Here is what is real, what is aspirational, and why it matters for anyone tracking where AI capital is going.
What the Stargate Project AI infrastructure deal actually is
The Stargate Project is a joint venture announced in January 2025 to build AI data centers across the United States, committing $500 billion over four years with $100 billion deployed immediately. OpenAI, SoftBank, and Oracle are the founding partners, with MGX as an additional equity backer. The stated goal is roughly 10 gigawatts of AI compute capacity.
In structure, it is a separate company — not a contract, not a fund, but a standalone joint venture with its own balance sheet. SoftBank holds financial responsibility and Masayoshi Son chairs it; OpenAI holds operational responsibility and is the anchor tenant that will actually use the compute. Oracle provides cloud and data-center engineering. The remaining capital — the gap between the $100B committed and the $500B headline — is meant to come from debt, additional equity partners, and project financing raised against the assets as they are built.
Who funds the Stargate Project AI infrastructure buildout
The $500 billion splits across equity funders and technology partners. The numbers below are the publicly stated commitments and roles as of mid-2026 — note how few of them are hard dollar figures versus directional commitments.
| Party | Role | Stated commitment |
|---|---|---|
| SoftBank | Lead financier / chair | Primary financial responsibility |
| OpenAI | Operator / anchor tenant | Operational responsibility |
| Oracle | Cloud & data-center build | Equity + engineering |
| MGX (Abu Dhabi) | Equity investor | Sovereign-fund capital |
| NVIDIA | Chip supplier | GPUs for ~10GW capacity |
| Microsoft | Cloud / Azure partner | Technology partner |
| Arm | Chip architecture | Technology partner |
The tell: only the $100B initial tranche is firmly capitalized. The remaining $400B is a commitment to raise, not committed cash. That distinction is the difference between a press release and a balance sheet — and it is why I treat the headline number as a ceiling, not a forecast. You can track how this lines up against hyperscaler spend on the AI Spending Tracker.
How the Stargate Project AI infrastructure compares to hyperscaler capex
$500B over four years sounds enormous until you stack it against what the big four are already spending every single year. The cloud giants are collectively deploying well over $300B in capex in 2025 alone. Stargate is large, but it is not in a different universe — it is roughly one extra hyperscaler's worth of annual spend, concentrated on a single tenant.
| Program | Capital | Timeframe |
|---|---|---|
| Stargate Project | $500B target | 4 years (2025–2029) |
| Microsoft capex | ~$80B | FY2025 |
| Amazon (AWS) capex | ~$80B+ | 2025 |
| Google (Alphabet) capex | ~$75B | 2025 |
| Meta capex | ~$65B | 2025 |
| Combined Big 4 capex | ~$300B+ | 2025 (annual) |
The Real Constraint Isn't Money — It's Power
The 10-gigawatt target is the number that should worry anyone underwriting this. 10GW is the rough output of 10 large nuclear reactors, or enough electricity to power roughly 7–8 million American homes. The first Abilene, Texas campus alone is designed to draw on the order of a gigawatt. You do not just buy that — you have to generate it, transmit it, and cool it.
U.S. grid interconnection queues already run 3–5 years in many regions. New gas turbines have multi-year lead times. This is why Stargate sites are being chosen for power access first and everything else second — Texas because of its independent ERCOT grid and permissive permitting, not because of talent or tax breaks. The bottleneck for the entire AI buildout has quietly shifted from chips in 2023 to capital in 2024 to power in 2026.
NVIDIA can ship the GPUs. SoftBank can syndicate the debt. But no one can conjure 10GW of firm power on a four-year clock without nuclear, gas, and grid upgrades that are themselves multi-year projects. That is the gap between the announcement and the reality.
Why This Matters for Founders and Investors
Whether or not Stargate hits $500B, the announcement itself reshapes the market. Three things follow directly:
Power is the new venture thesis
Grid, cooling, nuclear SMRs, and energy software just became AI-adjacent categories worth a look.
Compute concentration risk
A single tenant (OpenAI) anchoring a $500B buildout ties model progress to one company's roadmap.
Sovereign capital is in
MGX and SoftBank mean Gulf and Japanese money are now structural funders of U.S. AI infrastructure.
The number is a ceiling
Underwrite to the $100B that's real, not the $500B that's aspirational. Commitments slip.
Stargate is the most ambitious infrastructure bet in tech history — and most of it is still a promise.
$100B is real. $500B is a target. 10GW of power is the question that decides which number wins.
Track AI infrastructure spending and valuations on the AI Spending Tracker and AI Valuations dashboards at Value Add VC. Originally published in the Trace Cohen newsletter.