TL;DR
- The gist: OpenAI and SoftBank Group have invested $1 billion in SB Energy to build a 1.2 GW data center in Milam County, Texas.
- Key details: The deal includes acquiring construction firm Studio 151 and marks the first physical deployment of the $500 billion “Stargate” AI infrastructure initiative.
- Why it matters: By bringing construction in-house, the partners aim to bypass utility delays and secure sovereign-scale compute capacity faster than standard leasing models allow.
- Context: This move coincides with Meta’s recent 6.6 GW nuclear investment, confirming an intense industry-wide competition to lock down power for next-generation AI models.
Moving to execute their infrastructure ambitions, OpenAI and SoftBank Group have committed $1 billion to SB Energy. The strategic partnership secures a 1.2 gigawatt (GW) data center lease in Texas, marking the first concrete deployment of the $500 billion “Stargate” initiative.
Beyond purchasing power, the partners are verticalizing the supply chain by acquiring Studio 151, a specialist construction firm. The acquisition brings engineering capabilities in-house as the companies make an effort to build sovereign-scale compute facilities.
Competition for grid capacity is intensifying. Just hours earlier, Meta announced a separate nuclear agreement for 6.6 GW, confirming an industry-wide competition to lock down firm power for next-generation artificial intelligence (AI) models.
Promo
The Execution Phase: From White House Promises to Texas Soil
Formally committing to the venture, OpenAI and SoftBank Group invested $1 billion in equity to SB Energy. Funds are split evenly, with $500 million from each partner directly backing the development of a 1.2 GW data center in Milam County, Texas.
Representing the first tangible asset of the $500 billion “Stargate” initiative announced in January 2025, the Milam County lease moves the project from planning to physical construction. While the initial announcement outlined a broad vision for US-based AI infrastructure, concrete deployment has now begun.
Greg Brockman, President of OpenAI, outlined the strategic logic behind the joint venture, emphasizing how it merges two distinct technical disciplines. On one side, SB Energy provides the necessary physical backbone through its experience in large-scale infrastructure and energy development.
On the other, OpenAI contributes its proprietary knowledge regarding the specific engineering needs of AI workloads. By fusing these capabilities, the partners aim to bypass standard bottlenecks, creating a streamlined pathway to deploy massive, highly optimized compute clusters faster and more reliably than traditional methods would allow.
Milam County is one of five planned sites for the broader initiative. Geographically, the plan aims to tap into diverse energy markets, with other locations targeting Abilene, Texas, as well as sites in Ohio and New Mexico.
Vertical Integration: Buying the Builders
In a parallel move to control the supply chain, SB Energy announced the acquisition of Studio 151. Bringing a track record of 20 completed campuses to the partnership, the specialized data center construction firm adds immediate execution capability.
Acquiring the firm signals a strategic shift for OpenAI. Rather than operating solely as a tenant leasing capacity from third-party developers, the company is becoming an active participant in the development process.
In their announcement, the partners detailed a new framework for infrastructure delivery that moves beyond standard developer-tenant relationships. Through a non-exclusive “preferred” alliance, the companies are integrating their workflows to create a specialized model for constructing AI facilities.
Under this arrangement, OpenAI contributes its proprietary “first-party” architectural designs—blueprints specifically tailored for the unique thermal and power density requirements of AI models. SB Energy complements this by applying its operational expertise in rapid deployment, strict cost management, and complex energy grid integration.
The goal of fusing these distinct capabilities is to move away from generic data center builds and instead deliver purpose-built, gigawatt-scale infrastructure faster and more efficiently than the current market standard.
Owning the construction management layer allows the partners to control costs and timelines more tightly than standard hyperscale leases allow. Vertical integration mirrors strategies seen in the semiconductor industry, where design and manufacturing are increasingly coupled.
The Energy Competition: Stargate vs. Prometheus
Scale remains the defining feature of the Milam County project. At 1.2 GW, the single site exceeds the output of a standard nuclear reactor, which typically generates around 1 GW.
Competition for such capacity is high. Simultaneously, Meta announced a 6.6 GW nuclear power deal for its “Prometheus” supercluster. These parallel moves highlight the direct competition between the two AI giants for grid access.
Joel Kaplan, Policy Chief at Meta, linked infrastructure directly to geopolitical standing. “State-of-the-art data centers and AI infrastructure are essential to securing America’s position as a global leader in AI.”
Strategies differ distinctly between the two companies. Meta is locking in firm nuclear baseload through its partnership with Vistra and SMR developers. SB Energy, by contrast, relies on a renewable-heavy portfolio to power its facilities.
Google is also active in this space. Recently, the search giant acquired Intersect Power for $4.75 billion, a move designed to bypass interconnection queues by owning the generation assets directly. The acquisition allows Alphabet to control the power generation source.
Financial Alchemy: The SoftBank Loop
Funding its portion of the deal, SoftBank liquidated $5.8 billion in Nvidia stock. The sale of Nvidia shares represents a rotation from liquid public equity into illiquid private infrastructure.
These flows create a circular financial structure. SoftBank sells external assets to fund OpenAI, which in turn pays SB Energy, a SoftBank subsidiary, for power and infrastructure.
Analysts have warned of “financial alchemy” in such arrangements. Reliance on continued valuation growth of OpenAI to support the debt and equity commitments involved increases systemic risk if revenue growth stalls.
Projections indicate a projected deficit of $14 billion in 2026, making efficient capital deployment necessary for long-term solvency.
Feasibility & The Grid Reality
Broadly, the Stargate vision targets 5 GW per site, totaling nearly 25 GW of capacity. Energy experts have questioned the physical possibility of adding this amount of power by 2030.
To put the figure in context, 25 GW is roughly equivalent to the entire installed capacity of countries like Ireland or Portugal. Adding such load to the US grid faces physical and regulatory hurdles.
Grid interconnection queues in the US currently average more than four years for new projects. Located within the ERCOT market, the Milam County site benefits from Texas’s independent grid operation.
ERCOT operates independently of federal FERC regulation, potentially allowing for faster deployment than in other regions. Regulatory arbitrage appears to be a key component of the site selection strategy for the initial Stargate facilities.
