The reason for IBM's failure is pretty obvious. Their old school bare-metal servers at Softlayer are a good value, but the need for bare metal is decreasing with IO improvements like SR-IOV and real hardware level virtualization (ex HVM on AWS). I've also heard IBM is running that division into the ground.

Their cloud offering sucks. Last time I was curious, I couldn't even figure out how much my server would cost. They're too used to old school contracts when I can easily setup a server with many providers in 5 min with $5 for a month. IBM has never been a commodity business, but nearly everything they sell is becoming a commodity.

GCP's problem, as far as I can tell, is too much "cute" shitty documentation. And their hatred for maintaining services on a reasonable timescale. Everyone I know, even outside of tech, has lost a Google service they loved over the years to neglect. See Golang package management for a notorious example in the devsphere. You can't maintain dependencies on old package versions. Wtf?

It seems to be common knowledge that Google's infrastructure is technically the most solid of any cloud provider. That's just not enough when you need something easy to setup that you can build then forget about for a decade. That's just the reality of how software projects are done for non-technical businesses