Does AI computing power have a shelf life too? Microsoft doesn’t want to keep renting old GPUs, leaving Oracle in an awkward spot!

Your phone may still be good for watching videos after four or five years, but chips in the AI industry don’t get such a generous upgrade cycle.
One key reason Microsoft is gradually ending its computing-power rental partnership with Oracle is that the existing hardware uses an older generation of NVIDIA GPUs, which no longer quite fits Microsoft’s changing needs.
This is a wake-up call for the entire AI infrastructure industry: computing assets also face the risk of technological depreciation.
Today’s highly valuable GPUs may face competitive pressure in the future as new generations of chips emerge.
Worse still, data centers must also bear the costs of electricity, maintenance, networking, and financing. If equipment utilization declines, returns on investment may be affected.
Of course, older-generation GPUs are not entirely without value. Oracle has also said there is still demand for its AI infrastructure, including older-generation GPUs.
But how much customers are willing to pay is the real question.
In the AI era, the biggest fear isn’t having no chips—it’s having chips that are still there after customers have moved on to the next generation!#微软终止甲骨文算力租赁并退出3个项目