SoftBank, OpenAI, Oracle and 'Stargate': Will the $500B Plan Actually Happen
Texas hosts a $500B AI infrastructure plan. We audit funding, land, power, and silicon as the four chokepoints in mid-2026.
In January 2025, President Trump made a White House show of unveiling "Stargate" — a joint venture between SoftBank, OpenAI, Oracle, and MGX (UAE) to pour $500 billion into AI infrastructure over four years. Eighteen months in, where does the plan actually stand, and which pieces are stuck? Here's the June 2026 check-in.
The shape of the plan: starting in Abilene, Texas
The initial site is Abilene, Texas. Oracle covers land and buildings, SoftBank (and the Arm group) handles capital and hardware procurement, OpenAI brings software and workloads, and MGX manages Middle East ties.
- Initial commitment: $100B (immediately deployed).
- Total plan: $500B over four years.
- Near-term power demand: multiple 1.2GW-class datacenter buildings, scaling toward 10GW eventually.
- Jobs: 100,000 created per launch announcement, though the real figure deserves scrutiny.
Chokepoint 1: funding
$500B is on the order of several years of total global datacenter capex at 2024 levels. SoftBank has been working through Vision Fund 3-style structures and Arm-share-collateralized financing through 2025-2026, but Arm's stock volatility and SoftBank's group-level leverage cut both ways.
OpenAI is covering its end through tens-of-billions equity rounds and an implicit IPO option, while Oracle leans on long-term customer contracts to back bond issuance. Funding isn't strictly impossible, but hitting the full number assumes markets stay friendly.
Chokepoint 2: land and buildings
The Abilene site is under construction on schedule, with first-phase power-up expected within 2026 (partly speculative). Parallel candidate sites in Texas, Oklahoma, and New Mexico are also in play, so land itself isn't the binding constraint.
The next item is.
Chokepoint 3: power
10GW is equivalent to one to two of Japan's main nuclear reactors. Texas's ERCOT grid can't absorb that alone. Hitting the number will take some combination of new gas plants, SMRs (small modular reactors), solar-plus-storage, and long-term contracts with existing nuclear operators.
OpenAI's investment in Oklo (SMR startup), and Microsoft/AWS/Google locking up existing nuclear plants on long-term contracts in 2024-2025, all sit in the same context. Power doesn't build on a six-month timeline. It's likely Stargate's hardest binding constraint.
Chokepoint 4: silicon
Demand for NVIDIA's Blackwell generation (B200/GB300) and the Rubin generation (volume production starting 2026) remains stretched, with HBM supply from SK hynix, Samsung, and Micron forming the choke. Stargate is racing to lock long-term contracts with NVIDIA and Broadcom (custom ASICs) in parallel, but it's still fully in competition with the other hyperscalers (Microsoft, Google, Meta).
A "Stargate gets specially abundant silicon" scenario isn't realistic. Procurement yields deserve a conservative read.
SoftBank's exposure: balance-sheet risk
With Masayoshi Son out in front, SoftBank Group's consolidated liabilities and commitments are back on the rise. Arm shares have held high through 2025-2026, but sentiment around AI-adjacent equities can shift fast.
The "Stargate succeeds and SoftBank is reborn, fails and SoftBank takes its biggest loss ever" framing gets used a lot. The truer read is that SoftBank's outcome is closely tied to Arm-stock and broader AI-equity sentiment, with the variance dominated by markets.
Will it actually happen? My read
Best guess: "the scale will be discounted, but a substantial version will go up." Even if the full $500B doesn't land, AI infrastructure worth $150-250B materially coming online is fairly likely (speculative).
The biggest risk is the macro environment in late 2026 onward, plus US political events. When political backing softens, project priorities tend to shift.
FAQ
Q. Does this affect SoftBank consumer customers in Japan? Directly, no. Indirectly, the volatility shows up in Arm and SoftBank share prices.
Q. What's the relationship with AWS or Microsoft Azure? Competitive. Which OpenAI workloads stay on Azure versus migrate to Stargate is genuinely fluid.
Q. Could a similar site land in Japan? SoftBank's Hokkaido and Osaka plans are running on a separate track. Whether they fold into Stargate proper is undecided.
Read also
Comments (0)
No comments yet. Be the first to leave one.