Stargate Expansion Stalls Oracle and OpenAI Scale Back Texas AI Mega-Center.
Oracle and OpenAI have reportedly scrapped plans to expand their flagship AI data center project, part of the "Stargate" initiative in Abilene, Texas. Sources indicate that the decision stemmed from a combination of financial hurdles, concerns over data center power stability, and frequent shifts in OpenAI’s infrastructure requirements.
The Crusoe and NVIDIA Connection
The Abilene site is being developed by Crusoe Energy Systems, a firm backed by NVIDIA. Following the withdrawal of the expansion plan, Crusoe and NVIDIA have reportedly begun scouting for new partners to occupy the additional capacity. Meta has surfaced as a potential candidate; however, industry insiders remain skeptical as Meta is currently prioritizing AMD GPUs for its AI clusters a move that conflicts with Crusoe’s NVIDIA-centric infrastructure offerings.
The Scope of the Cancellation
It is important to note that this move does not derail the entire Oracle-OpenAI partnership. The massive 4.5-gigawatt (GW) data center agreement remains on track. The cancellation specifically targets a previous expansion proposal intended to boost a separate site's capacity from 1.2 GW to 2.0 GW.
Sachin Katti, OpenAI’s Infrastructure Lead, clarified that while the broader Stargate vision remains active, the company is actively evaluating alternative geographical locations for future build-outs.
Expanding from 1.2 GW to 2.0 GW isn't just about adding servers; it's about using power equivalent to a large city. The "stability" issue mentioned in the news may mean that the Texas electrical grid (ERCOT) cannot handle the simultaneous power surges from a massive number of AI chips, potentially causing brownouts.
Meta's potential rejection of the project due to its use of AMD chips (Instinct MI300/MI325X) reflects the polarization of the data center market. Platforms designed for NVIDIA (such as Crusoe's) have different cooling and networking architectures than AMD, making tenant transfer more difficult than anticipated.
OpenAI's frequent shifts in requirements may stem from their shift towards in-house chip design and smaller, more efficient modeling (Small Language Models), reducing the need for massive, centralized data center space.
OpenAI's search for alternative locations might mean diversifying into states with cleaner energy sources (nuclear or geothermal) to avoid carbon taxes and the instability of fossil fuels.
Strait of Hormuz Alert Brent Crude Surges as War Drums Drown Out Economic Stability.
Source: Bloomberg
Comments
Post a Comment