OpenAI's Stargate Data Centers Hit by Cost Overruns

The developers building data centers for OpenAI's ambitious Stargate infrastructure project are encountering significantly higher-than-expected construction and operational costs, adding financial pressure to what is already one of the most expensive AI infrastructure bets in history. Stargate — the joint venture backed by OpenAI, SoftBank, and Oracle, with a stated commitment of up to $500 billion over four years — was announced with great fanfare in early 2025, but the ground-level realities of building at that scale are proving costly.

The cost overruns reflect broader challenges in the AI data center buildout race, where surging demand for power infrastructure, specialized cooling systems, and Nvidia GPU clusters has driven up prices across the board. Land, electricity procurement, and skilled construction labor have all become more expensive as hyperscalers and AI startups compete for the same finite resources simultaneously.

The timing is notable given that OpenAI is also reportedly preparing to launch a significant new AI model and expects to pursue an IPO within the next year. A public offering would require OpenAI to present a credible path to profitability — something that becomes more complicated if the capital expenditure side of the equation keeps expanding. SoftBank's Masayoshi Son, who has taken a lead financial role in Stargate, will be watching these cost dynamics closely as the venture tries to deliver on its headline-grabbing promises.

The situation underscores a central tension in the current AI boom: the race to build the most powerful infrastructure is accelerating faster than the economics can be cleanly modeled, leaving investors and partners exposed to cost surprises that could ripple through valuations and timelines.