Microsoft is in advanced negotiations to lease hundreds of megawatts of data center capacity at a high-profile AI campus in Abilene, Texas, according to two informed sources. This development follows competitor Oracle's recent decision to abandon the same project.
Previous reports indicated that Meta is also in discussions to lease space within the campus. The construction and operation of this project are projected to cost tens of billions of dollars, a significant portion of which is allocated for the Nvidia chips required to power AI servers.
These negotiations are significant because the Abilene campus has been viewed as a bellwether for global AI data center development. Oracle's decision last week to abandon its expansion plans initially sparked investor concerns about a potential cooling in demand for large-scale AI data centers.
The campus is being developed by the startup Crusoe. Currently, Oracle operates a portion of the facilities for its key cloud client, OpenAI.
Spokespeople for Crusoe, Microsoft, and Nvidia declined to comment on the report.
For Microsoft, securing capacity at this campus would support its rapidly growing Azure AI cloud business, which includes providing cloud server leasing services to OpenAI. It remains unclear if Microsoft has identified specific end-customers for the capacity once the campus is completed. Similar to competitors like Google and Amazon, Microsoft executives have stated they are building data centers as fast as possible and indicated that cloud business growth could be even higher if more server capacity were available.
The potential Abilene agreement under consideration by Microsoft and Meta involves approximately 600 megawatts of capacity, which is expected to be brought online in phases. This increase is relatively modest compared to their total server capacity. Most large-scale AI data center projects are planned for 1 gigawatt or more.
Oracle had already leased eight buildings at the campus for OpenAI, intended for deploying computing clusters. This is part of the joint "Stargate" project aimed at constructing a super-sized data center. When fully completed, this phase will have a total power capacity of approximately 1.2 gigawatts—enough to power a major US city—making it one of the largest AI computing facilities in the country.
The campus has future potential for expansion to around 2 gigawatts. Oracle had previously negotiated to provide this additional capacity for OpenAI, but those plans to expand for OpenAI ultimately reached an impasse.
According to reports, OpenAI declined to lease the additional capacity because it prefers to deploy new clusters for Nvidia's next-generation Rubin chips at different locations, rather than mixing multiple generations of hardware within the same campus.
Following OpenAI's withdrawal, Oracle had planned to sign a contract with another customer for the expansion, but those negotiations collapsed last weekend. The specific reasons for Oracle ultimately abandoning the project are not yet clear.
After Oracle dropped the expansion plans, Nvidia attempted to keep the project moving forward. It paid a $150 million deposit to Crusoe and sought new clients for the campus, as the facility was specifically designed for Nvidia's AI chips.
For Crusoe, a startup valued at over $10 billion, a new agreement would ensure the Abilene campus continues construction at full speed following the revised plans. Construction crews have already been working on site for weeks on the expansion.
It has been reported that JPMorgan Chase has agreed to provide more than $9 billion in loans for the construction of the Abilene data center. The construction cost for each of the initial eight buildings leased by Oracle for OpenAI is approximately $1.4 billion, excluding the Nvidia chips which the tenant procures separately. These buildings are still under construction and, once complete, will collectively house around 400,000 Nvidia chips.
To date, two of the eight buildings have been completed.
If customized development for Microsoft proceeds, Crusoe would need to invest billions of dollars more.
Comments