Re: "servers have a useful life of five years"
CPU and architecture progress is anything but a crawl, if you've looked at the enterprise space lately. It was pretty much crawling until AMD came along and threw an EPYC party, blasted past Intel, and now Intel is madly scrambling to catch up.
Then along came servers with PCIe 4.0 (thanks, AMD), and now along come servers with PCIe 5.0 (thanks, Intel), and here comes the NVMe that can take advantage of it (thanks, NAND vendors!) and finally ease THAT horrible bottleneck. Then we have the new SSD form factors coming, which have been talked about for years but are finally making it into real mainstream products.
Your spinning disks may have a 10-yr life span (assuming you got HGST/WD drives, and not Seagate), but even there, every few years you can get a drive that's twice the size, while using the same or less power - not that people really use spinning rust in the datacenter much anymore.
For enterprise/datacenter SSDs the advances are even more notable - not that long ago, a 12TB SSD was exciting (except the only company that had one made it out of a bunch of eMMC glued together with silly putty, and the performance was terrible). Now we can get an off-the-shelf 30TB SSD from a couple different vendors, and the performance is actually good. In the near future, hopefully I can get a 60TB SSD from the same vendors, reducing my part count, cost, and overall power draw.
Getting rid of the servers that are 3-5 years old is not so much because they're "worn out" and non-functional, but because it doesn't make economic sense to keep running them when they're so far behind the things I can buy now. When I can wheel in one rack of hardware to replace and outperform 3, that's a win - less datacenter space to buy.
So, the limit is more about what makes sense from a performance/scale perspective, and not about the actual operable life of the hardware.