IBM to its credit has always plowed back the huge profits it makes in the Mainframe world into R&D for the mainframe, and that technology waterfalls down to the Power architecture and then eventually to their Intel chipsets. Along the way, they have also made money licensing or selling patents or selling excess fab capacity to other companies, true.
However the reality is that as a percentage of revenue and revenue to the company, the Systems group has always been tiny compared to its sister Software and Services department.
The bigger problem is that it's not clear the mainframe "scale up" model is the wave of the future. If you start using "scale out" computing, then you need a very large number of cheap computers --- and IBM doesn't do low margin very well. Historically any time a product enters the commodity (high volume, low margin) space, IBM sells the division: example: printers, hard drives, laptops, etc.
Maybe IBM could sell software to control the large scale out, "big data" wave of the future. But the problem there is that IBM software has always optimized for big machines, where it's acceptable to take the overhead hit of having dozen or more Java Virtual Machines chewing up a large amount of CPU and Memory. The argument IBM has always used is that programmers are expensive, and memory and CPU is cheap in comparison. That maybe true with a few dozen machines. But if you now have several thousand, or tens of thousands, of machines in a big "scale out" cluster, the overhead of putting a half a dozen or a dozen JVM's on each server starts taking its toll. Originally IBM was pitching using its Tivoli management software and WAS, which used JVM's and which had a heavy memory and CPU footprint for its cloud initiative. Now it's looking at some of the open source cloud management software, which is (a) not written in Java, and (b) open source, which means a big hit to IBM's software margins, at least insofar as it pursues that market.
So I recognize IBM's contributions on the hardware side, and that's all good. But in terms of long term future where the a large part of the industry is going (huge cloud deployments with small mobile devices as the access points), it's not clear to me IBM can follow. Sure, there will always be a half dozen big Oracle database machines running SAP or Oracle Financials, and maybe those will be on a big Power Iron. But if the rest of the data center is populated with thousands and thousands of cheap Intel servers, I don't see where the next stage of IBM's growth is going to come from.