OpenAI, Oracle, and Meta Navigate Shifting AI Infrastructure Landscape

A planned expansion of the Stargate datacenter, a key facility for OpenAI’s compute needs, has reportedly stalled due to financing challenges and uncertainty surrounding OpenAI’s demand forecasting. This development has opened the door for Meta, potentially leasing unbuilt capacity at the Abilene, Texas site, brokered by Nvidia.

Stargate Expansion Plans Derail

Just over a year after announcing the $500 billion Stargate initiative, OpenAI and its compute partner Oracle are reportedly abandoning plans to expand their flagship datacenter in Abilene, Texas. Bloomberg reported that the expansion, initially intended to reach up to 2 gigawatts of capacity (up from the current 1.2 gigawatt campus under development), was hindered by financing issues and OpenAI’s difficulty in accurately predicting future compute demand [The Register].

Meta Steps In, Nvidia Facilitates Deal

With the expansion in question, Meta is now reportedly in discussions to lease the unused capacity from Crusoe, the site’s developer. Nvidia played a key role in facilitating this potential deal, reportedly putting down a $150 million deposit on future capacity before approaching Meta [The Register].

Oracle’s Financial Strain and Layoffs

The situation comes as Oracle faces a significant cash crunch stemming from its massive investment in AI data center expansion, including the $300 billion partnership with OpenAI. The company is planning to cut around 30,000 jobs across various divisions to address these financial pressures [Economic Times]. Some of these cuts are targeted at roles expected to become redundant due to advancements in AI [Economic Times]. Oracle announced plans to raise an additional $50 billion in debt and equity to finance its datacenter aspirations, but it appears this funding may not arrive in time to salvage the Abilene expansion [The Register].

Hyperscaler Investment in Compute Capacity

The demand for datacenter capacity is immense. The eight largest hyperscalers – Google, Amazon, Meta, Microsoft, Oracle, Tencent, Alibaba, and Baidu – are collectively projected to spend $710 billion in 2026, largely on data centers and the GPUs that power them [The Register]. Meta, in particular, has signaled its willingness to invest heavily, allocating up to $135 billion for capital expenditures focused on GPU compute capacity.

OpenAI’s Broader Compute Needs

Despite the setback with the Stargate expansion, OpenAI’s existing $300 billion contract with Oracle for 4.5 gigawatts of compute capacity remains on track [The Register]. The company, along with partners Nvidia and SoftBank, is likewise pursuing a $110 billion funding round, with a goal of securing 5 gigawatts worth of GPUs.