Stargate reality check: only the first $100B is real, Microsoft is the biggest loser

Summary

The $500 billion Stargate headline is misleading. Only the first $100 billion represents committed capital with secured power and data center sites ready to deploy. The remaining $400 billion is speculative, contingent on future funding rounds and site development.

What's actually happening

The $500 billion and $100 billion figures measure total cost of ownership (TCO)—not capital expenditures. TCO includes data center construction, servers, networking, power infrastructure, operations, and financing costs. This matters because it inflates the apparent commitment. The first phase, $100 billion, is genuinely allocated to the Abilene, Texas campus already under construction, which will house 100,000 NVIDIA GB200 GPUs across 20 separate 500,000-square-foot buildings. Ten buildings are currently underway, with 180 megawatts of power capacity today and a full gigawatt targeted by 2025.

The Abilene campus itself represents the Oracle-Crusoe data center deal announced last year, being built by Lansium, Crusoe, and Oracle. This was repackaged into the Stargate joint venture as one component of a larger infrastructure build-out. The structure matters: different equity holders own different pieces. Oracle owns its data center. OpenAI will own the GPUs housed there and run the cluster management software. SoftBank, Oracle, MGX, and OpenAI are the equity founders, though funding is incomplete and the mix of equity versus debt remains undefined.

Funding and the capital shortfall

SoftBank has under $10 billion in immediately available capital, despite being listed as a $19 billion equity contributor on paper. Elon Musk's public criticism of SoftBank's liquidity was correct. The company will likely fund its stake by selling a portion of its ARM Holdings position. OpenAI similarly lacks the $19 billion in committed capital but can raise it once O3 is released and commercialized. Project financing—debt layered on top of equity—will cover the gap.

The structure insulates parent companies from balance sheet risk. If OpenAI or SoftBank cannot meet rental payments on the cluster in 2026 and beyond as costs scale, the risk sits with the joint venture entity, not the parents. This is deliberate financial engineering.

Winners and losers

NVIDIA is the unambiguous winner. It supplies all GPUs and will see massive orders as the cluster expands from 100,000 GPUs to 1 million to 10 million GPUs across successive phases. ARM benefits tactically because it supplies the CPUs that coordinate alongside every NVIDIA GPU, but the upside is offset by SoftBank's need to sell ARM shares to fund Stargate—a fact most investors are missing.

Microsoft is the biggest loser. It sits on the sidelines as OpenAI pairs with Oracle for infrastructure instead. Microsoft negotiated a right of first refusal (ROFR) but feared the investment risk and declined to write the check. Oracle carries the burden of managing a 1.8-gigawatt cluster and procuring all hardware while OpenAI controls operations.

The real capital question

The gap between announced commitments and actual available capital is real. SoftBank and OpenAI together account for $38 billion in claimed equity but have access to perhaps $15 billion today. Oracle and MGX (Abu Dhabi's new $100 billion technology fund) can absorb significant portions through their balance sheets, but the ultimate funding mechanism—particularly for phases two through five totaling $400 billion—remains unresolved. Debt will carry more weight than the headline suggests, and that debt will have covenants tied to OpenAI's revenue and profitability, neither of which is guaranteed to scale fast enough to service the obligation.

The first $100 billion is plausible. Everything beyond it depends on OpenAI's ability to commercialize its technology at a rate that justifies the spending. That's the real bet underneath the cluster names and the power specs.