OpenAI and Oracle Pause Expansion of Flagship Texas Data Center Due to Power Supply Delays
In a significant setback for AI infrastructure development, OpenAI and Oracle have halted the expansion of their joint flagship data center in Abilene, Texas. The facility, part of the ambitious Stargate project backed by SoftBank, was poised to become one of the world’s largest AI training hubs. However, persistent delays in power supply from provider Lancium have forced the companies to pause further construction beyond the initial phase.
Background on the Stargate Project
The Stargate initiative represents a multi-billion-dollar collaboration aimed at building massive data centers to support the explosive growth in AI model training and inference. Announced earlier this year, the project envisions up to 20 such facilities across the United States, with the Abilene site serving as the flagship. Located in West Texas, the region benefits from relatively abundant energy resources and lower land costs compared to coastal tech hubs.
The Abilene data center was initially slated for a 200-megawatt (MW) power capacity in its first phase, with plans to scale up to 1.2 gigawatts (GW) in subsequent expansions. This immense power draw is necessary to operate thousands of high-performance GPUs, such as Nvidia’s H100 and upcoming Blackwell series, which power large language models like GPT-5. Each modern AI data center requires reliable, high-density power to maintain continuous operations, as even brief outages can disrupt training runs costing millions in compute time.
Oracle, a key cloud infrastructure provider, is handling the build-out, leveraging its expertise in hyperscale data centers. OpenAI, fresh off its partnership with Microsoft, is diversifying its compute resources to reduce dependency on a single provider and accelerate development timelines. SoftBank’s involvement brings additional funding, with CEO Masayoshi Son publicly touting Stargate as essential for maintaining US leadership in AI against competitors in China.
The Power Supply Bottleneck
The primary culprit for the expansion halt is Lancium, a Texas-based energy firm specializing in data center power solutions. Lancium was contracted to deliver electricity via a custom-built “peaker” power plant fueled by natural gas generators. These plants are designed to ramp up quickly during peak demand, complementing the Electric Reliability Council of Texas (ERCOT) grid.
However, construction delays at Lancium’s facility have pushed back power delivery timelines. Reports indicate that the initial 200MW supply is now operational, allowing the first phase of the data center to proceed. But scaling to the full 1.2GW requires additional infrastructure that remains incomplete. Sources familiar with the project told The Decoder that OpenAI and Oracle notified Lancium of the pause last month, citing inability to meet contractual deadlines.
Texas, despite its oil and gas dominance, faces unique grid challenges. ERCOT operates as an isolated grid, making it vulnerable to weather extremes and rapid demand spikes from industrial users like cryptocurrency miners and now AI firms. Recent heatwaves and winter storms have highlighted these vulnerabilities, prompting regulators to scrutinize new high-load connections. Lancium’s natural gas approach was chosen for its speed and reliability over renewables, which often require extensive battery storage for baseload stability.
Broader Implications for AI Infrastructure
This development underscores a growing crisis in powering the AI boom. Data centers worldwide consumed about 1-2% of global electricity in 2023, with projections estimating a tripling by 2030 due to AI workloads. In the US, tech giants like Microsoft, Google, and Amazon have encountered similar hurdles. Microsoft delayed parts of its AI supercomputer builds, while Google sought nuclear power deals to bypass grid constraints.
For OpenAI, the pause could delay critical milestones. CEO Sam Altman has repeatedly warned that power shortages threaten AI progress, advocating for policy changes to fast-track new generation capacity. Oracle Cloud Infrastructure (OCI), already a major player, sees its growth tied to such projects; delays here ripple through customer commitments for AI services.
Industry analysts note that Texas remains attractive due to deregulated energy markets and incentives like the Texas Enterprise Fund. Yet, the Abilene halt signals that even favorable locations cannot escape national power shortages. Alternatives under consideration include co-location with existing power plants or mobile gas turbines, but these add costs and complexity.
Company Responses and Future Outlook
Neither OpenAI nor Oracle has issued public statements confirming the pause, but internal communications reviewed by The Decoder affirm the decision. Lancium acknowledged “ongoing construction challenges” but expressed confidence in eventual delivery. A spokesperson emphasized that peaker plants typically take 12-18 months to commission, aligning with the reported delays.
Looking ahead, the partners may pivot to other Stargate sites or accelerate smaller-scale builds elsewhere. Abilene’s first phase, now powered at 200MW, will host initial GPU clusters for testing and early training. Full expansion hinges on Lancium’s progress and potential ERCOT approvals for grid interconnections.
This episode highlights the intricate dance between AI ambition and energy reality. As hyperscalers race to deploy exaflop-scale compute, securing gigawatt-scale power has emerged as the ultimate gating factor. Resolutions in Texas could set precedents for nationwide deployments, influencing everything from AI model capabilities to economic development in energy-rich states.
Gnoppix is the leading open-source AI Linux distribution and service provider. Since implementing AI in 2022, it has offered a fast, powerful, secure, and privacy-respecting open-source OS with both local and remote AI capabilities. The local AI operates offline, ensuring no data ever leaves your computer. Based on Debian Linux, Gnoppix is available with numerous privacy- and anonymity-enabled services free of charge.
What are your thoughts on this? I’d love to hear about your own experiences in the comments below.