Stargate is a $500 billion, four-year bet to build OpenAI's compute backbone on American soil — but only $100 billion of it is actually committed today. That's the short answer. The longer answer is more interesting.
Announced from the White House on January 21, 2025, Stargate was framed as the largest private AI infrastructure project in history. SoftBank and OpenAI lead it, Oracle builds it, and Abu Dhabi's MGX helps fund it. The number that got the headlines — half a trillion dollars — is a target, not a wire transfer. Understanding the gap between the $500B announcement and the $100B commitment is the whole story.
For context, $500 billion over four years is roughly $125 billion a year — on par with a single hyperscaler's annual AI capex. What makes Stargate different is that it's a standalone, OpenAI-anchored vehicle, not a line on Microsoft's or Google's balance sheet.
What the SoftBank–OpenAI Stargate Deal Actually Is
The SoftBank–OpenAI Stargate deal is a joint venture announced in January 2025 to invest up to $500 billion over four years building AI data centers across the United States, with $100 billion deploying immediately. SoftBank holds financial responsibility, OpenAI holds operational responsibility, and Oracle and Abu Dhabi's MGX are the other initial equity backers. Masayoshi Son chairs it.
The structure matters. This is not OpenAI buying servers. It's a separate company — capitalized by outside investors — that builds and owns the infrastructure, then contracts capacity primarily back to OpenAI. That lets OpenAI lock in years of compute without putting hundreds of billions of capex on its own books, and it lets SoftBank deploy capital into the single hottest asset class on earth with a built-in anchor tenant.
Who Pays for Stargate: The Backers and Their Commitments
The equity and technology partners split into two groups: those writing checks and those supplying chips, cloud, and software. Here's how the announced roles break down.
| Partner | Role | Reported Commitment / Contribution |
|---|---|---|
| SoftBank | Lead — financial responsibility | ~$18–19B initial equity; Masayoshi Son chairs the JV |
| OpenAI | Lead — operational responsibility | ~$18–19B initial equity; anchor tenant for capacity |
| Oracle | Equity + primary data center builder | Develops and operates campuses; signed multi-year cloud deals |
| MGX (Abu Dhabi) | Equity backer | Sovereign wealth capital from UAE's AI investment fund |
| NVIDIA | Technology partner | GPUs and networking for the compute clusters |
| Microsoft | Technology partner | Azure relationship; right of first refusal historically |
| Arm | Technology partner | CPU architecture; SoftBank-owned chip designer |
| Crusoe Energy | Site developer | Builds the flagship Abilene, Texas campus |
Sources: White House announcement (Jan 2025), OpenAI and SoftBank statements, Oracle disclosures. Initial equity figures are widely reported estimates, not audited filings.
The $500B vs $100B Gap: What's Committed in the Stargate Deal
The most important sentence in the entire announcement was the smallest number. Of the $500 billion target, only $100 billion is committed for immediate deployment. The remaining $400 billion is expected to materialize over four years as data centers come online and as the venture raises additional equity and debt against contracted demand.
This is standard for mega-scale infrastructure — you don't pre-fund a four-year buildout on day one — but it's also where the skepticism lives. Within hours of the announcement, Elon Musk publicly claimed the backers "don't actually have the money," and that SoftBank had "well under $10B secured." OpenAI's Sam Altman pushed back directly. The truth sits in between: the $100B tranche is real and deploying, and the path to $500B depends on project financing that hasn't all been raised yet.
- →Committed now: ~$100B, anchored by SoftBank and OpenAI equity of roughly $18–19B each, plus Oracle and MGX participation.
- →Target by ~2029: $500B total, requiring an additional ~$400B from new equity, debt, and project-level financing.
- →The collateral: long-term capacity contracts, primarily from OpenAI — which is itself a pre-profit company. The financing is only as solid as OpenAI's ability to pay.
Track how this flows through private AI valuations on the AI Valuations Dashboard, and see OpenAI Valuation 2026 for the anchor tenant's own math.
Where Stargate Is Being Built and How Much Power It Needs
The buildout targets roughly 10 gigawatts of compute capacity — an almost incomprehensible amount of power. For scale, 10GW is enough to power roughly 7–8 million homes, and it exceeds the entire generation capacity of many US states. The flagship site in Abilene, Texas, developed by Crusoe Energy, is designed to house hundreds of thousands of NVIDIA GPUs across multiple buildings.
Crusoe-developed campus; first buildings energized in 2025; designed for 1GW+ and hundreds of thousands of GPUs
OpenAI, Oracle, and SoftBank named additional US locations across Texas, New Mexico, and the Midwest, pushing planned capacity past the initial target
~10GW total target; sites co-located with or near new generation, including natural gas and grid-scale connections
The announcement projected 100,000+ jobs created — a figure that includes construction and is debated by analysts
The power requirement is the real bottleneck — not capital or chips. See the AI Spending Dashboard for how energy is reshaping the entire AI capex cycle.
How Stargate Compares to Big Tech AI Capex
Stargate sounds like it dwarfs everything, but spread over four years it's roughly $125B a year — squarely in hyperscaler territory. The difference is concentration: this entire vehicle exists to feed one model company.
Four-year $500B target; OpenAI-anchored JV
Calendar 2026 capex guidance
2025 AI and cloud capex
2025 AI capex
2025 capex, ~39% of revenue
Microsoft + Amazon + Google + Meta
Compare the public-company numbers in real time on the Big Tech Earnings Dashboard.
Why SoftBank Made This Bet
For Masayoshi Son, Stargate is the comeback narrative. After the Vision Fund's WeWork-era wreckage and tens of billions in writedowns, SoftBank had been searching for the next defining bet. Son has called AI the realization of his life's thesis, and SoftBank also owns Arm — whose chip architecture sits underneath much of the AI hardware stack — and took a separate multi-billion-dollar position in OpenAI equity in 2025.
So Stargate isn't an isolated infrastructure play for SoftBank. It's vertically aligned: equity in OpenAI (the tenant), ownership of Arm (the architecture), and now a financing lead role in the data centers themselves. If the AI buildout pays off, SoftBank captures value at three layers. If it doesn't, the concentration cuts the other way — and SoftBank has been here before.
The Bull and Bear Case on Stargate
Bull Case
- ✓ OpenAI demand for compute is genuinely outrunning available capacity
- ✓ Dedicated infrastructure removes OpenAI's dependence on Azure alone
- ✓ Government backing and US-soil siting de-risk permitting and politics
- ✓ SoftBank + Arm + OpenAI alignment captures value across the stack
- ✓ $100B is real and already deploying in Abilene
Bear Case
- ✕ $400B of the $500B is unfunded and depends on future raises
- ✕ The anchor tenant, OpenAI, is still pre-profit and cash-burning
- ✕ ~10GW of power is a brutal physical and grid constraint
- ✕ Inference costs are falling fast, compressing revenue per GPU
- ✕ SoftBank's track record on mega-bets is uneven at best
The $500 billion number was never the point.
Stargate is a bet that whoever controls AI compute controls the AI economy — and that the $100B already in the ground will pull the other $400B in behind it.
Track real-time AI infrastructure and capex data on the AI Spending Dashboard, AI Valuations, and Big Tech Earnings Dashboard at Value Add VC. See also: Microsoft $80B AI Capex and The Microsoft–OpenAI Deal Explained.