A big problem I've always had is getting hardware to improve productivity. Most of the time the management chain can't grasp why it's worth a $3k machine and maintenance costs to save each dev an hour a week in development time.
I think this goes to just how often people I management like the title and the power but don't understand that they're supposed to be in charge of strategy. I don't think most of them would know strategy if it walked up and bit them.
For example, if I can spend a day tweaking my project's build scripts and CI workflows to get it from taking 30 minutes to build/test/deploy down to 30 seconds, I've dramatically altered my team's workflow.
Also, there are very few reasons to buy your own hardware anymore -- cloud services are better at almost everything unless it's a low-latency operation (or you're in Australia - the speed of light is still too slow, sorry mate). Obviously there are exceptions, but with the scale of GPGPU build-out all the big public clouds are deploying, the benefits of anything else are rapidly disappearing.
Most managers aren't swayed by the argument that 1 hour = ($annual_salary / 2000). They know that if you give people an extra 5 minutes a day, they'll probably just spend it drinking coffee or socializing. Those certainly aren't bad things for employees to do, but unless your team is already running well beyond capacity, giving them more time to do things doesn't really improve productivity. Improving process (by, say, removing a manual step through automation) really gives you more bang for your buck.
Cost. I've yet to see a cloud service that get anywhere near as cheap as owning your own hardware (fully loaded with financing costs, full staff costs etc.) for anything that runs more than ~8 hours a day.
Cloud services have plenty of uses, especially for brief capacity spikes. Need to run an experiment for a few days that requires 20 extra servers? Go nuts. But even with reserved instances, if most of your servers run 24/7, you're throwing money out the window.
(And you don't need to own hardware - you can go for rented dedicated managed servers and still get ~80% of the savings)
And keep in mind salaries are just a fraction of the costs of staff time. Payroll taxes plus benefits can add up to almost 20% of base salary and that's not counting occupancy and equipment costs.
Bottom line: people cost a lot. It's smart to think carefully through the ramifications of planning decisions.
What I found bizarre was that each time I've had to explain why I needed the numbers, and people were surprised at receiving a report of what the staff budget went. You'd think that would be expected, not something unusual. But permanent staff is not treated as a cost during day to day work, even though their time is a limited resource.
It's quite fascinating to see how quickly priorities change when the cost breakdowns make the rounds and someone higher up starts asking critical questions about why we spent $50k of staff time on feature X that no customer had asked for.