Meta Platforms is pursuing a roughly 2-gigawatt artificial intelligence data center in Texas, targeting the same Abilene site that OpenAI and Oracle just abandoned. The move comes after the two original partners scrapped plans to expand what had been billed as one of the highest-profile AI facilities ever announced. If Meta closes a deal for the Crusoe Energy-owned campus, it would inherit a project whose power demands alone could serve hundreds of thousands of homes, and it would reshape the competitive math behind the largest AI training clusters in the country.
OpenAI and Oracle Walk Away From Abilene
Oracle Corp. and OpenAI have scrapped plans to expand the flagship AI data center that both companies had treated as the crown jewel of the Stargate initiative. The Abilene facility, owned by Crusoe Energy, had been positioned as the anchor of a broader buildout that included five additional data centers planned with Oracle and SoftBank. The decision to pull out leaves a massive campus without its two most prominent tenants at a time when demand for AI compute capacity is only accelerating.
The withdrawal was separately confirmed through reporting that cited the end of the Texas expansion plan. Neither Oracle nor OpenAI has issued a detailed public explanation, and the sourcing on the reasons remains limited to people familiar with the private talks. That opacity matters: without a clear account of what went wrong, the industry is left to read between the lines on whether the issue was cost, power availability, timeline friction, or something else entirely.
How the Stargate Flagship Took Shape
The Abilene site gained national attention when OpenAI leaders showed off the Stargate data center at a public event in Texas, laying out an ambitious vision for AI infrastructure at a scale few companies had attempted. That event established the public narrative around Stargate’s flagship campus and set the timeline for what was supposed to be a rapid construction push. OpenAI also announced plans for five more data centers elsewhere with Oracle and SoftBank, framing the Abilene site as the first domino in a global chain of AI supercomputing hubs.
Oracle had been rapidly filling buildings on the site, according to Bloomberg reporting on the facility’s status. The Crusoe-owned data center in Abilene ranked among the highest-profile projects yet announced in the AI infrastructure race. That profile is precisely what makes the collapse of the expansion so striking: this was not a speculative greenfield proposal but an active campus with real hardware being installed.
Why Meta Sees Opportunity in Abandonment
For Meta, the Abilene campus represents something rare in the current AI infrastructure market: a site that already has physical buildings, power interconnection work in progress, and local permitting largely resolved. Building a 2-gigawatt data center from scratch can take years of environmental review, utility negotiation, and construction. Stepping into a facility where much of that groundwork is done could shave significant time off Meta’s own AI training timeline.
The strategic logic runs deeper than convenience. Big tech companies are locked in a race to secure enough electrical capacity to train and run ever-larger AI models. Power is the binding constraint. A 2-gigawatt campus would consume electricity on a scale comparable to a mid-sized city, and sites with that kind of grid access are not easy to find, especially in regions where utilities are already stretched thin by residential and industrial growth. Texas, with its deregulated energy market and relatively abundant generation capacity, has become a magnet for data center developers, but even there, grid operators are warning about tightening supply.
Meta’s interest also reflects a broader consolidation pattern. When one major player abandons a large infrastructure project, the sunk costs and partially completed work create a discount for the next buyer. The permitting, land acquisition, and early construction represent real value that does not disappear just because the original tenants left. For a company with Meta’s balance sheet, acquiring that value at a lower price than building fresh is a straightforward financial calculation.
What the Exit Signals About AI Buildout Risk
The collapse of the Oracle-OpenAI expansion plan at Abilene challenges a popular assumption in the tech sector: that every announced AI megaproject will reach completion. Over the past two years, dozens of companies have announced data center plans measured in gigawatts, often backed by splashy press events and political endorsements. The Abilene exit is a concrete example of how those announcements can fall apart when the details of cost, power delivery, and partnership terms get tested against reality.
Most coverage of AI infrastructure treats every new announcement as a done deal. That framing misses the friction that sits between a press conference and a functioning data center. Power purchase agreements can stall. Construction costs can escalate. Partners can disagree on cost-sharing or technical specifications. The Abilene situation suggests that even the most prominent projects, backed by some of the best-capitalized companies in the world, are not immune to these pressures.
The fact that high-profile geopolitical stories can share space with opaque infrastructure negotiations in the same news cycle also underscores how little visibility the public has into the mechanics of AI buildouts. Investors, local governments, and grid operators all make decisions based on announced capacity plans. When those plans quietly dissolve, the downstream effects on power planning and community expectations can be significant. Abilene, a city that likely anticipated jobs and tax revenue from a massive data center expansion, now faces uncertainty about whether a new tenant will deliver on similar promises.
Texas as the Default Destination
The Abilene saga highlights how Texas has become a default destination for energy-intensive computing projects. The state’s lightly regulated power market, ample land, and history of courting large industrial users have drawn in crypto miners, cloud providers, and AI startups alike. For companies like Meta, Texas offers both political support and a grid that, at least on paper, can be expanded to accommodate multi-gigawatt loads.
Yet the same characteristics that attract data centers also create vulnerabilities. The state’s grid has faced scrutiny since the 2021 winter storm, and large new loads can amplify stress during peak demand. Regulators and utilities must balance the promise of high-tech investment with the risk of volatility for residential and small-business customers. As more hyperscale campuses cluster in places like Abilene, those trade-offs become harder to ignore.
Local officials often welcome the tax base and construction jobs associated with such projects, but the long-term employment footprint of modern data centers is relatively small compared with their physical and electrical footprint. Communities betting on AI infrastructure as an economic engine may discover that property and sales taxes, rather than permanent headcount, are the main local benefit.
Financing, Policy, and the Cost of Delay
Behind the scenes, the economics of AI infrastructure are tightening. Building and powering a 2-gigawatt campus requires billions of dollars in capital, long-term power contracts, and confidence that AI demand will remain strong enough to justify the expense. Recent moves in other corners of finance, such as a large private credit fund limiting withdrawals, illustrate how even deep-pocketed investors are becoming more cautious about liquidity and long-duration bets.
Policy uncertainty adds another layer of risk. Trade and tariff disputes can affect the cost and timing of key hardware components, from advanced chips to cooling systems. In a separate context, the U.S. government’s stance in a tariff-related court battle shows how contested and slow-moving trade rules can be. For data center builders, shifting import duties or export controls on semiconductors can quickly alter project budgets or delay deployments.
When projects stall or partners walk away, the cost of delay is measured not only in lost time but also in opportunity. AI model development cycles move fast; a year-long slip in infrastructure readiness can translate into missed product launches or competitive disadvantages. That urgency helps explain why Meta would consider stepping into a half-built campus rather than waiting for a new site to clear every regulatory hurdle from scratch.
What Comes Next for Abilene
If Meta ultimately takes over the Abilene campus, the city could still realize many of the benefits it expected from the original Stargate plan, albeit under a different corporate banner. Construction activity would likely resume, local suppliers would see renewed demand, and the region’s profile as an AI hub would be reinforced. The details, however, will matter: commitments on hiring, community investment, and grid upgrades will shape how residents perceive the trade-offs.
For the broader industry, Abilene will serve as a case study in how fast the landscape can shift. A project once touted as a flagship for one alliance may become a cornerstone for a rival. The episode underscores that in the race to build AI infrastructure, control over physical sites and power capacity can be as decisive as breakthroughs in algorithms or chips. As companies recalibrate their plans, the winners may be those willing to move quickly when others step aside, and to absorb the political and financial risks that come with inheriting someone else’s unfinished megaproject.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
