Oracle is building yesterday’s data centers with tomorrow’s debt

OpenAI has pulled back from expanding its Stargate data center partnership with Oracle in Abilene, Texas, because it wants access to newer Nvidia GPU generations rather than the Blackwell chips planned for the site. This exposes a structural risk in AI infrastructure: data centers take 12–24 months to build, but Nvidia now ships a new chip generation annually, each with major performance leaps. Oracle is uniquely vulnerable because it is the only major hyperscaler funding its buildout primarily through debt—now exceeding $100 billion—rather than operating cash flow. With Oracle's stock down 23% year-to-date and off more than 50% from its September peak, investors are watching closely as the company faces a $50 billion capex plan with negative free cash flow. The broader implication is that GPU depreciation risk now threatens the entire AI infrastructure investment thesis.