Oracle is turning massive capital spending into a revenue windfall: the company expects to recognize $76.6 billion（約12兆円） in incremental AI infrastructure revenue in fiscal 2027 alone, with another $219.6 billion（約35兆円） over the following two years, as five new datacenters come online.

What happened: Oracle fired up 1.2 gigawatts of datacenter capacity in fiscal 2026 and will add another 1 gigawatt in Q1 fiscal 2027. The company has contracted five major datacenters (Stargate in Texas, Shackleford in Texas, Dona Ana in New Mexico, Saline in Michigan, and Port Washington in Wisconsin) totaling 5.5 gigawatts of capacity. In the quarter ended May, Cloud Infrastructure revenue grew 93.2 percent to $5.79 billion（約9300億円）, with CPU and GPU rentals doubling at 2.2X growth. Oracle's capex exploded from $6.9 billion（約1.1兆円） in fiscal 2024 to $166.8 billion（約27兆円） in fiscal 2026.

Why it matters: Oracle has bet $291.6 billion（約47兆円） over three years building AI infrastructure, but has a revenue backlog of $638 billion（約100兆円） to draw on—meaning the company will likely recover its investment and then some. With OpenAI as a major customer anchoring the backlog, Oracle is positioned to capture a large share of AI compute demand. Even if OpenAI were to default, the article notes there is a long list of other customers waiting to rent GPU capacity, and Oracle could potentially earn more selling that capacity on the open market.

What to watch: Oracle will recognize 12 percent of its revenue backlog ($76.6 billion（約12兆円）) in fiscal 2027 and another 34 percent ($219.6 billion（約35兆円）) across fiscal 2028 and 2029. The Stargate datacenter is at 42 percent capacity with another 35 percent coming in the next three months. GPU utilization across Oracle's entire fleet is at 97.5 percent. These milestones will show whether the company can execute the buildout and convert the backlog into realized profit.