The paradigm has shifted quite a bit since the hey-day of AIX systems. Back then, you would custom configure a substantial AIX system, and require premium service from IBM, with fast response from them. RS/6000s were renowned for reliability, though, with very few core system problems (most problems happening in costly adapter cards). With "commodity" type Intel hardware, reliability is acceptably "okay", DoAs not uncommon, and redundancy realistic in the form of adjoining rack mount X-series or similar boxes, or even blades. Most software products have accommodated the paradigm shift, and are sold for Linux as well now.
The major challenge these days is not computer cost but the cost of supporting complexity, as many companies are finding out the hard way. Disk arrays with multi-level RAID, SANs, and VMs make for an amalgam that is becoming impossible for humans to understand. Downtimes and service-level problems are a big factor in implementations. (Ugh - a RAID array is rebuilding and will entail hours of impaired performance... Sure hope another disk doesn't fail before that completes or we may lose everything.) And good luck getting vendor support: they will know their little piece, but recoil if you try to involve them in the larger picture. This is making for a substantial market for technology consultants, who may be your only hope of dealing with the complexity; and that adds substantial cost to implementations. Oh, the days of simple computer systems... Richard Sims