Microsoft is winding down Oracle deal to rent AI cloud capacity

Commenters interpreted Microsoft’s decision to wind down its Oracle capacity deal in sharply different ways. One view treated it as a routine shift away from older A100 GPUs used for Bing toward newer chips Microsoft can run itself; Oracle could reportedly rent the older capacity to other customers. Others argued that demand does not settle the central question: whether renting GPUs can earn a return after providers repeatedly invest in newer, more expensive generations. They questioned whether customers will keep paying for older hardware and warned that the buildout could outrun profitability before a market downturn. One commenter stressed that utilization alone means little without knowing the rental price.

The debate turned on how durable and broad AI demand is. A commenter cited roughly $2.3 trillion in hyperscaler service backlogs, including about $1 trillion tied to OpenAI and Anthropic, to argue that commitments from other customers also support the investment. Skeptics focused on the concentration of demand in frontier labs, uncertain returns for smaller AI firms, and how long creditors will fund companies that have yet to show a return. Another commenter called the financing circular, while a different view held that bubble predictions can be wrong for a long time. Political and local opposition also entered the discussion: one commenter argued that weaker prospects for federal support and anti-datacenter measures could make the buildout harder to sustain.