Apple's cloud software engineering muscle is fairly weak and providing an IaaS might be "easy" for the hardware side of the company, but very difficult for the software side.
My last job was at a tiny startup for 4 software engineers and we picked AWS because of that. As the people using the platform, we hated AWS after they consistently messed up our account in every possible way despite paying $700/month for support... but they routinely took our sales team (???) out for dinner, so we never switched to GCP. (I used GCP at my next/current job and they are ... weird. But we used their managed K8s service in a very weird way and they were very supportive when we escalated things.)
No, the support situation is weird. You have to have a contract with some third-party company, in our case, SADA. You then send all your support requests to them, and if they deem it necessary, they escalate it to Google. I feel like we always stumped them and then a person with a google.com email would be cc'd and instantly give us a good answer. SADA was very chill, though. Pretty competent in general except for their weird account enrollment system (which I had to deal with any time someone joined my team).
The weird thing that we were doing was creating 100s of managed Kubernetes clusters a day. We make an open-source app that runs on Kubernetes and tried to have a cloud offering, and this provided the best isolation between customers. People would sign up, and then we built all the infrastructure for them from scratch. (The AWS way is to have the customer give you their AWS account ID, and then they would pay for all the infrastructure themselves. But for whatever reason, GCP didn't seem to have that option, and honestly our best customers were all too "new" to the cloud to have a routine AWS bill. So it worked for us, and whenever cluster creation was glitchy, GCP was happy to help us. What we were doing was unusual, but since we always paid on time, I guess, they were not mad about it or anything. I did offer to CC them on our cluster-creation monitoring alerts and they declined ;)
Anyway, what I thought was weird was that with AWS, AWS is your support contract. With Google, it's some other company entirely. That'd kind of weird, right? You buy something from Vendor X and then when the product doesn't work, you talk to Vendor X. But with Google, we were too small potatoes for them to be worth talking to us for. But if something got escalated to them, they were very nice about it.
(What's extra crazy to me is that I used to work on Google Fiber, and we had great support for people paying them much less than we were paying. Want to call someone and talk about how you're scared about viruses for 45 minutes? Google Fiber support would listen and assuage your fears, every single time. They were great people consistently doing a great job. Not sure why GCP didn't steal them!)
The weird thing here is that you dealt directly with SADA and knew that it was a third party. I actually see that as a good thing as it is more transparent.
Yeah, now that you mention it, the whole SADA thing is pretty transparent. I don't have any major complaints; it just feels weird to me is all. Despite what HN thinks, Google can offer good support. I've seen it done! But it is apparently not their favorite thing.
Apple has already proven that they can build a server product, they just haven't figured out the "why" yet. Cloud customers are trying to buy low-margin compute with good support for their software. Apple is trying to sell high-margin compute with good support for their software. The indifference towards either party's needs damned Xserve from the start, and Apple would have to take a radically different approach if they want a shot at succeeding nowadays. Their hardware would be competing with Nvidia's enormous Grace servers and Ampere's dirt-cheap ARM vCPUs. Their software would probably be ignored unless it was Open Source.
And all that begs the question - why bother? If Apple won't be making iPhone-sized hardware margins or getting Mac-level brand loyalty, they'd be spending money to undermine their core products. Comparatively, a VR headset that retreads the last decade of HMD tech a-la iPhone is a pretty safe bet.
So, not exactly something that I would consider competitive against the likes of Azure or AWS. In fact, AWS now sells services where you can bring your VMware internal clouds over to be officially supported by VMware on AWS.
if for not other reason than that being perhaps the most likely way to actually compel apple to do it themselves, hahaha! :)
I’ve always been suspicious that some of the cheaper “rent a Mac mini VM” may be doing trickery with a Linux virtual machine host.
To get plausible Mac mini benchmarks on the fastest x86 processor, you'd need overhead to be less than 25%. Plus, you'd be buying a $650 x86 processor instead of spending $600 on a Mac mini. You'd also need to think about power. That x86 processor draws 150W base and 253W peak compared to an M1 Mac mini which hits a peak of 39W. Even if you're getting pretty cheap power (5-cents per kWh), that's still $44/year in extra costs.
I've used QEMU on Github's x86 runners to run Linux-ARM and it's painfully slow. It seems like it would be hard to make the economics work out well while still offering a plausible experience speed-wise.
And who would your customers be? You won't get US or EU companies wanting to use your Hackintosh Cloud. They know the jeopardy they'd get in if they were building Hackintoshes. Renting them probably isn't much better. What happens when Apple figures out that a company has been building their iOS apps on the Hackintosh Cloud?
Plus, if you were to create a Hackintosh Cloud that got any traction, it would be a constant moving target as Apple tried to foil your plans. What happens when the next security update also has something that makes your machines not boot? Sure, you find a workaround over the next month and in the meantime your customers find that your service isn't worth it. Companies pay a fortune for things to be reliable.
Finally, how much could you really undercut genuine M1 pricing? Scaleway will rent me an M1 Mac mini for €80/mo (€0.11/hour) or an M2 Pro for €173. You're going to need to pay really smart people to keep breaking Apple's attempts against you. Scaleway just buys some Mac minis. Plus, the ARM processors you're buying won't have the performance of Apple's chips and might even have slightly different behavior for things like memory ordering. Maybe that won't matter most of the time, but if I'm paying for a build cloud, I probably don't want to be second guessing it.
When it comes to cloud services, I think buying genuine Macs isn't really the issue. I mean, Github is charging $350/mo for mediocre 2-core x86 runners. They're charging nearly $7,000/mo for an M1 runner. If you wanted to compete with Github's runners, you could just rent Scaleway servers at €80/mo and still hugely undercut Github's pricing as long as you got a bit of usage. At half of Github's price, you'd need about 17 hours of usage per month to cover Scaleway's price - less than 2.5% utilization.
Apple knows that people use them as servers.