My first job in 2008 was working on a backup product that ran on a ton of different OSes and architectures (HP-UX on Itanium and PA-RISC, AIX on PowerPC, Solaris on SPARC and 32+64-bit X86, SCO Unix and Unixware, Linux, Windows, FreeBSD, MacOS, Novell Netware, Linuxes running on IBM zSystems, probably a couple more that I'm forgetting). On most of those systems, we basically made a headless backup daemon. It was only really Windows and MacOS where we made a
I was there for about 10 years, and over that time, the number of customers using non-x86 machines plummeted, and the ones hanging on were frequently on ancient versions of their respective OSes.
My impression was that even when I started, those generally weren't new deployments. We were supporting customers who'd been sold the hardware years before, and were just trying to wring use out of them before moving over to a more "standard" setup, i.e. Linux on 64-bit x86. I infer that to be the case because our support matrix got smaller over time, dropping platforms that customers weren't paying us to support anymore. Development moved further and further from the aging Unix machines in the lab, and they started feeling like "special cases", while development continued on x86-based VMs for the OSes that supported it.
I don't have an exact date or anything but feeling was with Google in the early 00's. While my "big" first gig was managing Linux and FreeBSD in the data center about the same time, we still had a lot of SGI, IBM/AIX, IBM VM/ESA, AS400, SunOS/Solaris and Windows. Our main driver for getting rid of the big iron was the industry was turning towards cheap commodity systems running Linux and saving the yearly support costs was a big win. Google doing Linux at scale was helpful for the management team to "risk" moving away from tried and true vendors.
My question is, when did this fact become clear?
In my experience there was not a specific year. Smaller businesses started requesting Linux in the early 00's and the popularity started working its way in with bigger companies. With time more Linux monitoring tools were being developed and that really helped the adoption as customers had SLA requirements that required standardized monitoring tools. More high availability tools and daemons also started being ported from BSD which also helped adoption. More and more enterprise capabilities meant more customers requested it. Eventually our customers had enough developers using Linux that it was just expected to be an option. There was more to it but this would turn into a book.
Oh but I should add one important driver ... cost. HP-UX, AIX were insanely expensive and Linux was free to our customers since we had a lot of Linux experience in house. Commodity hardware was significantly more cost effective if apps could scale horizontally. Solaris and Windows held a position for some time and then after Oracle roflstomped Solaris it was just Linux and Windows by the time I left that company. This was when commodity hardware and horizontal scaling become more common. Everything I described mostly transitioned from around 2000 to 2003.