For me, IBM has looked likely to buy Sun for many years, because doing so would give IBM two things that would be very valuable to it

1) Java 2) Sun's Solaris customers.

There are other SPARC vendors (e.g. Fujitsu) that they could sell off to and focus on their own lines (probably with a provision to use the IP they see useful).

Edit: Btw, the ecosystem around SPARC is slowly eroding. Fin Svcs are going x86 more and more. The problem with SPARC is they aren't growing in new markets - it's becoming marginalised.

Just finished up working for IBM at a major ex-state-monopoly Telco, working at the worlds largest hedge fund right now.

Another error might have been overvaluing its vertically integrated SW/HW solution technically and financially, and not commoditizing it quickly enough through open-source and hardware cloning.

Spreading one's wealth goes against one's better judgement often, but the rewards come back in unseen ways.

Sun did a lot of engineering stuff "right" in the late '90s, and again over the past couple years: they were first to market with a mainstream-grade MP kernel and first to market with a mainstream 64 bit OS, and recently they have ZFS and DTrace to talk about.

On the other hand, I think you could have predicted probably starting in 2001 that nobody was going to pay that much for large-scale MP, 64 bits, and advanced filesystems; the entire server market swung to web stacks, which scale horizontally and move most of the systems management burden into a DBMS and out of the Adrian Cockroft and Chris Drake systems engineering realm where Sun has excelled.

I really think that if you compare the systems-level innovation at Apple to the systems-level innovation at Sun, Apple comes out ahead. And systems-level innovation isn't even their core competancy.

Before this, architecture and ISA were absolutely key - RISC was the way of the future. RISC architectures were theoretically running loops around CISC ones.

AMD bolted a CISC onto a RISC core and effectively made this architecture point moot. This approach let AMD and Intel use their momentum and market to catch up and overtake their workstation friends.

And that Sun had one of the first full implementations (I think MIPS got there first).

The difference between microcode and RISC was the amount of synchronisation - RISC generally implied lots of uniform, high performance operations. Microcode could, but didn't necessarily. e.g. Microcode could still be doing funky co-ordination with lots of specialised execution units (and did afaik). This has some similarities to RISC, but RISC is a different philosophy really - namely fewer types, pipelined accesses, uniform instructions, more registers, etc.

(In my view) the big watershed moment was AMD's K5 architecture. This was literally based on one of AMD's pure RISC designs. It wasn't that popular in itself, but it set up a golden age for AMD... Which really only came to an end with Intel's Core.

Just to be pedantic.