• SoftBank's $16B loan talks in March 2025 were the financing mechanism for Stargate.
  • A $500B AI infrastructure buildout now spanning 4 continents. Here's what happened.

SoftBank Needed $16 Billion to Fund a Bet That Would Reshape Global AI Infrastructure

The March 2025 report about SoftBank seeking a $16 billion loan looked, at the time, like another chapter in Masayoshi Son's well-documented history of ambitious leverage. But the loan was not a speculative bet — it was a financing mechanism for a commitment that Son had made two months earlier, standing alongside Donald Trump, Sam Altman, and Larry Ellison at the White House to announce the single largest AI infrastructure project in history.

The Stargate Project, formally announced on January 21, 2025, committed up to $500 billion over four years to build AI infrastructure for OpenAI across the United States, with $100 billion planned for immediate deployment. SoftBank holds financial responsibility; OpenAI holds operational responsibility. Masayoshi Son serves as chairman.

The $16 billion loan talks were how SoftBank planned to fund its share of that commitment. What followed was an 18-month buildout that now spans four continents and involves more capital than any technology infrastructure project in history.

What Stargate Actually Is

Before the scale, the structure matters. Stargate LLC is incorporated in Delaware as a joint venture between OpenAI, SoftBank, Oracle, and investment firm MGX. SoftBank and OpenAI each contributed $19 billion and hold 40% ownership. Oracle and MGX each contributed $7 billion. The remaining capital comes from limited partners, debt financing, and equipment partnerships.

The technology partnership layer is equally significant. Arm, Microsoft, NVIDIA, and Oracle are the key initial technology partners. In September 2025, OpenAI announced that NVIDIA had made a $100 billion investment to purchase AI processors for the data centers.

AMD agreed in October 2025 to supply GPUs for 6 gigawatts of future deployment capacity. OpenAI is simultaneously developing its own custom AI chip — codenamed "Titan" — with Broadcom, to be fabricated on TSMC's 3nm process, targeted for mass production in the second half of 2026.

The Buildout in Numbers

The flagship site in Abilene, Texas defines the scale of what Stargate is physically constructing. The Abilene campus will draw 1.2 gigawatts of power once all eight AI factory halls come online — enough electricity to power roughly one million homes. Oracle will deploy over 450,000 NVIDIA GB200 GPUs at this single campus under a 15-year lease agreement. The first two buildings became operational in September 2025. By late 2025, the Abilene campus already employed 6,400 construction workers.

The expansion beyond Abilene moved faster than the partners had projected. By September 2025, five new US sites had been announced, bringing Stargate to nearly 7 gigawatts of planned capacity and over $400 billion in investment — ahead of the original schedule. The full roster of US sites includes locations in Texas, New Mexico, Ohio, Wisconsin, and Michigan. Stargate Michigan, announced in October 2025, is a $7 billion, 1 gigawatt facility in Saline Township — described by Michigan's governor as "the largest economic project in the state's history."

By early 2026, the project had reached 8 gigawatts of planned capacity and over $450 billion in committed investment, ahead of the original schedule.

The International Expansion

What began as a US infrastructure commitment rapidly became a global one. Stargate UAE, the first international project, was announced in May 2025 — a 1 gigawatt AI infrastructure cluster in Abu Dhabi, developed through a partnership between NVIDIA, Cisco, OpenAI, G42, Oracle, and SoftBank, expected to open in 2026.

Stargate Norway launched in July 2025 as the first European deployment, featuring AI data centre infrastructure powered by renewable hydropower. Stargate Argentina was announced in October 2025, representing an estimated $25 billion investment in Patagonia — the first Latin American site, with up to 500 megawatts of capacity, making it the largest AI data centre in Latin America.

The international expansion reflects something important about Stargate's strategic purpose: it is not only about US compute dominance. It is about establishing OpenAI's infrastructure presence in markets where sovereign governments are willing to co-invest in AI capacity.

The UAE's G42 partnership, Norway's renewable energy advantage, and Patagonia's cooling climate and available land are each different answers to the same question: where can large-scale AI compute be built fastest?

SoftBank's Full Commitment — and the Complications

The $16 billion loan in March 2025 was one part of SoftBank's financing strategy. The fuller picture emerged through 2025. SoftBank completed a $22.5 billion second closing of its OpenAI investment in December 2025, bringing its total investment to approximately $41 billion when including $11 billion from third-party coinvestors, and giving SoftBank approximately 11% ownership in OpenAI.

To fund this commitment, Son made a decision that signals the depth of the bet: SoftBank sold its entire stake in NVIDIA, worth approximately $5.83 billion, liquidating what had been one of the most valuable AI-adjacent positions in its portfolio to direct the capital toward OpenAI. The company that had made its reputation on bold technology bets — WeWork, Uber, Arm — was concentrating its next phase on a single thesis: that whoever controls the AI compute layer controls the next technological era.

The journey was not without turbulence. In August 2025, Bloomberg reported that the project had not started and no funds had been raised to meet the initial $500 billion budget, citing market uncertainty, American trade policy, and AI hardware valuations as contributing factors.

The report created significant uncertainty before OpenAI and SoftBank's subsequent announcements — the five new sites in September, the Michigan campus in October, the operational Abilene buildings — demonstrated that physical construction was, in fact, underway at scale.

What This Means for India and Emerging Markets

The Stargate model — sovereign co-investment in AI compute infrastructure — is a template that emerging market governments are now studying and replicating. The UAE Stargate is the clearest example: a Gulf state government using national wealth and a domestic AI company (G42) to secure frontier AI compute on its own soil, anchored to a US partner who provides the technology.

For India, which has articulated its own AI sovereignty ambitions through the IndiaAI Mission and the planned creation of domestic GPU computing capacity, the question is whether a similar anchor partnership is possible. SoftBank itself has deep India ties through its Jio investment, and the Stargate playbook — government land and energy commitments, foreign partner compute and model access — is structurally applicable to the Indian market.

Bottom Line

As of early 2026, Stargate is a sprawling, multi-continental initiative to build roughly 10 gigawatts of AI data centre capacity underwritten by an unprecedented $500 billion investment. The $16 billion loan that SoftBank was seeking in March 2025 was the first piece of a financing structure that ultimately committed $41 billion to OpenAI alone.

Son's bet — and the leverage he was willing to take on to make it — is the clearest signal in the market about where sovereign capital believes AI infrastructure value will accrete over the next decade. Whether the bet proves as prescient as SoftBank's early Alibaba investment, or as costly as WeWork, will be answered by whether the compute demand that Stargate is building for continues to materialise at the pace that OpenAI's $47 billion revenue run rate currently suggests it will.

Edited by Nabarun.