AWS claims customers are packing bags and heading back on-prem Cloud behemoth AWS says it is facing stiff competition from on-premises infrastructure, which is a turnaround from its once-proud boast that all workloads would eventually move to the cloud. In a summary of evidence given to UK watchdog, the Competition and Markets Authority (CMA), AWS denies that customers face any difficulty …
On-prem or cloud, horses-for-courses, some stuff is great in cloud and other stuff is bloody awful. The problem is that middle-management are always after a single perfect solution and as anyone who's been in IT for more then 5 mins will know, there ain't no perfect solution to anything in this game!
The other thing I've seen is "cloud migrations" where they just lift and shift on-prem into the cloud and ram it until it fits. No thoughts to redesign or rebuilding around cloud components or paradigms, just spin up a load of old school server instances.
That's probably not how it's supposed to be.
It would seem (to me at least) that from a customer point of view, the best option would be to keep cloud migrations easy. While I can understand that the likes of AWS, Google and Microsoft want to lock you in as much as possible, that's good for them. Not you. If you have your entire infrastructure hosted in (say) Google, and they suddenly announce they are closing the particular service you use, you are going to need to move your infrastructure quickly if your business is going to survive. I picked Google because they do have a habit of dropping services with little notice, although I would hope they at least honour any contracts they have.
Have you tried to use Appengine ?
There are many APIs that are completely non standard like their emailer or http fetcher that dont exist on other platforms, so theres no way you could ever take that code and put it anywhere else even on your own local machine. Most people dont create abstractions and if they did they are extremely leaky, so there goes a rewrite for that.
"If you have your entire infrastructure hosted in (say) Google, and they suddenly announce they are closing the particular service you use, you are going to need to move your infrastructure quickly if your business is going to survive."
It gets even worse if you have set up your workflow around a service that is unique to Google so there isn't a way to migrate over to another provider. They might just decide to triple the cost of the functionality you are using if they spot that there's no way out in the short term.
True, but that is what lot of people want. They are drinking the Cloud SaaS KoolAid that the sales people are constantly pushing. A lot of this is driven by Developers who never see any of the underlying issues, they just connect to a database the created and stuff security, backups, DR, etc. They simply don't care.
It is going to get worse before (if...) we see any change on this because there are so many now who don't know anything better.
It is not helped by outfits constantly going into C-Suite meetings pushing anything that is not "Cloud-Native" as "Legacy".
A lot of this is driven by Developers who never see any of the underlying issues, they just connect to a database the created and stuff security, backups, DR, etc. They simply don't care.
That's because developers write the applications, DBAs deal with the databases, and Ops should take care of server security, backups, and DR. The server should be a secure, backed up, and recoverable no matter if there's one app, two apps, or ten apps, and it's not the dev's job to do this unless their job description says that they're working in some multi-hat devops circle of hell.
"The server should be a secure, backed up, and recoverable no matter if there's one app, two apps, or ten apps"
If there is ever an issue, there needs to be a way to throw resources (money) at the problem to get it going again. With outsourced services, you might be lucky to get somebody on the phone.
Too true. What cloud resources can offer for scaling often requires certain design and development approaches.
Putting an access app on a terminal server in the cloud doesn't do anyone much good at all. I've literally seen this.
Similar for many rdbms apps. You doing get magical scaling by putting it in a cloud server, but keep the queries with dozens of joins and trying to scale to thousands of users.
Exactly. We're hybrid, we're now, stupidly, moving to fully cloud. In our every few weeks meetings I'd point out with our supplier and management that were on the call it would be more expensive. I was shot down every time "No it won't. You'll be getting rid of your old kit, so in the end, it will be cheaper", so I stopped bringing it up as no one was listening, or didn't want to hear the truth.
Roll on to the day we're almost fully cloud and we get asked from middle-management "We've just seen the cloud bill. Wasn't this move supposed to save us money? Why is it actually costing us more money". The next call we're on with the supplier, they finally admit "Yes, of course it will be more expensive". I smirk and just keep my mouth shut.
Cocks.
"Why is it actually costing us more money".
The problem with cloud solutions is exactly this. The cost isn't known (or fixed) in any meaningful way. It's literally "we'll bill you whatever this month" and it may be wildly different the next. Forever.
Most cloud providers have calculators which offer estimates of the cost of various solutions. Using them gives you a figure that's not any guarantee of what you actually end up paying.
We still use a few dedicated servers that have a cost of £x/year on a 3 year contract. We know exactly what those will cost and for a good period of time. That in itself is a big saving in terms of the time people don't have to spend figuring out what the cost will be!
In theory you're time sharing those staff, so you'd be sharing the cost with others, and in theory, only paying for when you need them to do something rather than paying them in case you need them to do something: That's why it's claimed there is salary savings with cloud.
However, that also means you don't know what skills they have, or how good they are: They're hired by the cloud provider to do a job, and that's what they do.
You're also bound to their SLA: You want quick response, you'll pay premium for it.
That's how they make their money: That's why it's not as cheap as on prem. You hire your staff: They work to your rules. You go cloud and rely on the supplier's staff: They work to the supplier's rules and to the suppliers priorities.
So you're right, but... that's now how things are presented by the sales-droids who tout cloud as the best solution, ever.
"However, that also means you don't know what skills they have, or how good they are"
You also have no control over the staffing level and what they are spending their time on. While it makes sense that plenty of what they'll be doing winds up amortized over a bunch of different clients saving you money, the rest of the time that staff isn't dedicated to your needs and working on solutions that fit your company. Saving money gets too much attention to the detriment of making a better produce or providing a better service.
"Forget the hosting costs, the people costs [of cloud] is also signifiantly greater than on prem".
The biggest problem I've seen at several employers is the overly complex permissions model of cloud systems. Days wasted while ops try and figure out why one system or application can't talk to another.
I had a presales tech from Microsoft tell me off the record that the only way to truly price up the cost of running an app in public cloud is to move it and see what the bills are... The calculators can give you a guide regarding some components but pricing up a complex tiered app - no chance...
Same thing happening where I work. They are stupidly moving "everything" to cloud and bills keep rising. Prior to the move, the nameless, faceless decision makers, the know-it-all types, used way too underrated infrastructure requirements and based the cost savings on those. As things are moving along, these nameless, faceless have already quit and now the remaining are finding out that infrastructure is far from adequate. As they increase the CPUs, add memory, move data to higher performance disks, the bills keep going up. You want a FW, you pay by the GB for all data entering and exiting. So now, everyone has given up all hopes of any cost savings. It is not even production yet. When production load hits, the s**t will hit the fan...I am just waiting and watching. Schadenfreude...
Regardless of that, I don't consider the fact that AWS is failing to deliver as promised in a lot of cases to be "competition." I consider it a failure to provide a useful service that the market actually wants, and that is an entirely different thing, and due to nothing other than Amazon's own failure to listen instead of tell.
Solutions like Nutanix offer the best of both worlds. Cloud like solutions on prem and reduction from cloud lock-in by running on bare metal in AWS and AZURE. True lift and shift whilst retaining the ability to be hybrid. They have services like DBaaS and S3/Objects and K8s with data. The perfect all round fit in my opinion.
“building a datacenter requires significant effort, so the fact that customers are doing it highlights” the fact the Cloud was severely oversold.
Verily, verily, I say unto you: In the realm of the cloud, many stewards found their costs beyond measure and sought refuge in the steadfast on-premises domain. They desired prudent management, secure protection, and control over their labors and local decrees. Thus, they returned to the peace and sovereignty of the haven they had once forsaken.
and yet they still saved money......
https://world.hey.com/dhh/our-cloud-exit-has-already-yielded-1m-year-in-savings-db358dea
Cloud exit suggests that they needed to get something to run the workloads on, rather than using kit they'd just had hanging around, so they've already budgeted the cost of one lot of hardware and it makes commercial sense over a period.
If the period is > 5 yrs, you're correct, there will likely be another hardware refresh, which no doubt will still leave them better off than having rented in the cloud for the same time period they're planning for.
I heard that people had to simplify (get rid of the old stuff they do not need/use) to move to cloud. If they had done the same work and stayed on prem it would have been win win.
Just like moving house to get more space!
Much like many other solutions that are now subscription based, whether on-prem software or a cloud solution.
This has been driven by the obsession in business that recurrent expenditure is better than capital. They don't like capital because it is lumpy and requires planning more than a month in advance.
Most of society is now modelled on monthly payments,
Phones
Cars
Entertainment
Energy
The list goes on.
"Most of society is now modelled on monthly payments,"
And that's exactly the problem that society should be wary of. "This entertainment subscription is only $9.99/month.... and this one, and this one". At the end of the month there's sofa change left in the checking account and if the company wasn't auto-deducting money to go in a retirement account, plenty would have nothing and no place to live come retirement.
If times are tight, a company could put off or scale back IT upgrades and make do with the on-prem they have which might be 100% owned. There's some equity in that gear to use as a cushion. Outsourcing leads to the "own nothing and be happy" paradigm. If you don't pay your monthly invoice, that's your business shut off. Don't upgrade your own gear and it's just an annoyance when you consider how much better things could be running, but the business is still open.
Exchange is a right pig to administer. Paying Microsoft for Office 365 is genuinely a cheaper & less stressful way to consume Exchange.
And before any smart arse says "Well don't use Exchange" let me introduce you to enterprise integrations that only work with Exchange. I live & work in the real work and that means having to have Exchange.
Paying Microsoft for Office 365 is genuinely a cheaper & less stressful way to consume Exchange.
AWS Lambda (and its equivalents at other cloud providers) is horrible. Event driven architecture even when it's not appropriate, unpredictable warm up times, ridiculous limits on execution time. A supplier of meal boxes went all in on this architecture and managed to create a system no one understands and results in incredible wastage from race conditions and no single source of truth for any data. The term "bulkheading", where data is cached for each sub-system to have any chance of performing adequately, still brings me out in hives.
It’s the same as any other service.
You will get someone in that writes a business case that proves you will save millions by outsourcing.
A couple of years later, after you have, someone else comes in that writes a business that proves you will save millions by bringing it back in house.
A couple of years later, after you have, someone else comes in that writes a business that proves you will save millions by outsourcing again.
If you actually made 1/10 of the savings, after a few iterations they would actually be paying you.
Similar to the constant activity involving between merging and taking over businesses then splitting or selling them off. Whether it actually benefits the companies involved or not, management types need to be seen to be doing something to justify their salary and bonuses, and this gives them something to crow about.
By the time it's clear that the outcome didn't meet the original promises, those who sold or pushed whatever it was on that basis and the management types who bought into the hype will have moved on elsewhere- fat bonus intact- no longer likely to be in a position to be held accountable, let alone actually likely to be called out on it.
for many of us [cough][cough] long in the tooth commentards around here.
We have been saying for years that going all in 'With the off-prem Cloud' was a bad idea. Sure, it made sense at the start because the costs were very appealing to the Bean-counters. Now? The exit costs could make the difference between the CTO keeping their job or getting fired.
A mix of on prem and off prem is probably the best mix. Systems that are critical to the operation of the biz should be 'on-prem'. Other, less critical stuff can be out in the cloud.
But I'm not a penny pinching beancounter so my opinion is irrelevant.
Often it's a question of capex v opex. Opex, cloud is more expensive so those pushing cloud focus on the capex. They also don't admit that the initial costs are the lure and the prices will ramp up quickly once that honeymoon period is over and you don't have the specialist staff available any more who supported on-prem.
It's why, if you look back to the early adopters, they moved to hybrid or pure on-prem within a few years, but they were touted as examples of how big business was embracing the future, and the future is Cloud.
Just ignore the reliance on someone else's hardware, the quality of their staff, your connection to their servers, their security, the potential they can get in and view your data/copy it/sell it et al: They won't because the reputational damage would put them out of business.
If they're caught, that is.
But they have the techs and you don't so how would you know? Okay, that's conspiracy territory: You will have techs to maintain what IT remains on your prem. The cloud supplier only supplies techs to support their infrastructure, not what you're putting on the VM's you've built. You don't save so much on staff as you think because you still need people who know what they're doing to implement the systems needed to support your business, you just don't need the hardware specialists you never really had unless you had a lot of on prem hardware 'cause supporting the servers was only a small part of your IT tech staff's duties rather than a full time job.
And all the while the costs go up 'cause you're using more and more of the cloud services available and aren't cleaning up after you 'cause it's not on your servers, and it's easy to forget what you've got.
Oh, and the supplier isn't going to tell you that you're paying for stuff you're not using: That's pure profit for them.
Yes, we've just been through that latter one: Found there were orphaned (unattached) disk we were paying for that we didn't need. Also found we had servers that were supposed to be used for short term testing that were sitting there, unused, but still being paid for (Business needs changed and demands shifted so people got distracted and simply forgot).
I think it largely comes down to AWS pricing not looking up with 4+ generations of performance and memory uplift. AWS pricing per core should have come down over time as many core processors became more common at lower process relatively speaking.
There are now 128-192 core monsters at or less than the 8-16 core options from a decade ago. Pricing in the cloud hadn't kept pace at all. Similar for memory and storage.
It's now crossed a point that the convenience of cloud isn't worth the higher margin costs to the customer. Combined with the fact that horizontal scaling is far less of a requirement than on hardware a decade or more ago.
Lastly, leveraging tools like Kubernetes is pretty much the same in prem or cloud.
That doesn't go into a lot of the not complex cloud configs either. Especially for internal apps.
It will get more interesting for SaaS providers though.
Oh the timing of this article is quite amusing. Didn’t Amazon CEO Andy Jassy (the former head of AWS) mandate a full time return to office? Maybe IT infrastructure can benefit from that on-prem collaboration too…
His boomer management insecurities are devaluing the whole premise of AWS.
(Also RTO nothing more than illicit layoffs, knowing a certain percentage of remote employees will quit and give up severance/unemployment benefits)
Plenty of organisations are still proudly boasting of their 'cloud first' strategies. I've always warned against that as it's more of an ideology than a strategy. 'Cloud where appropriate' was always my suggestion, but I'm not in marketing so I didn't get very far with it.
I am surprised by a few things, 1 that Amazon are in any way admitting to this practice, my inner cynic thinks they must see a tax break in claiming AWS isn't doing as well;
2 that returning to on prem is in anyway economically feasible. Most IT directors burned that bridge to get savings. To have to restart physical services plus recruit all the skills you paid off 10yr ago isn't going to be cheap. Possibly just goes to show how badly the cloud slingers are milking customers.
NEVER Ever use a cloud vendors sw, Lock in is lock in. Once you are locked in you have no negotiation ability.
Systems are cheaper than ever. You can afford your local critical infrastructure.
you can even aford a remote hot site. AWS... get the hw cheaper but do not pass on the savings.
You can keep test ,,, and research in the cloud.
The cloud can only increase your security attack surface and offers many more attack vectors to security.
Maintain local backups that are offline to mitigate ransomware attacks. Backup the backups to the cloud.
<quote>NEVER Ever use a cloud vendors sw, Lock in is lock in. Once you are locked in you have no negotiation ability</quote>
You can use the cloud to your benefits, however, treat cloud in code as on-premise, do not use their shiny expensive features that are subpar anyway, build your own, pay once for the effort, and it's yours!
On-site is always better, but when manglement wants cloud, make your cloud infra cloud-agnostic.
A half decent Sales team is good at flogging anything to anyone, once they've worked out your "Pain Point" (insert sales bullshit bingo phrases to suit) and your weakest link, they'll move in for the kill.
Vendor lock in, being sold as "look how much you save by using us for everything" is to benefit the vendor, as the OP already said.
Businesses will work out for themselves, in due, course, whether cloud is the right solution in it's entirety or in part.
I've been in tech long enough to see fad computing trends come and go, whether cloud is a long term part of IT, well time will tell.
Some organisations I've worked for really would benefit from taking the time to work out the right solution, not rush to the latest tech craze to be seen to be keeping up with the Jones' in their respective industries.
1. Management are persuaded to go cloud
2. Techies point out that it will only be cheaper if the always on, overnight batch processing crap is rearchitected
3. Management faint at the cost/time of rearchitecting
4. Lift & Shift "to achieve the objective"
5. Start getting billed.
What about the exposure to UK if all the major banks move to AWS and despite its good backup architecture, it gets taken out? With all the UK's "eggs in one basket" how long would we last. At least with each bank running old fashioned data centres also with good backup, the UK keeps some resilience.
This is what makes me twitch about their "making tax digital" nonsense. Given the way government IT projects* tend to end up after a few years, I can foresee a time of great wailing and gnashing of teeth and etc., when masses of people miss the self assessment deadline because HMRC's systems go pear shaped and they insist on charging late fees anyway.
* I've been involved in a few. Some even got mentioned here. Which is why I'm posting anonymously.
One of my customers was spending £40m per year on cloud hosting costs, they have now decided to "repatriate" as much as possible, they reckon they can host much of it themselves for significantly less and have built their own private cloud environment based on vSphere and NSX. However, moving to cloud did mean they were able to empty one of their data centres and sell off the land, they are now using Equinix to host their network kit and a couple rented data centres for all the server workloads.
" they are now using Equinix to host their network kit and a couple rented data centres for all the server workloads."
For the basics such as a roof over the gear, battery back up and a fat pipe to the internet, it can make good sense to rent some cabinets to house your gear. I happen to be a fan of owning hard assets such as land and buildings, but the corporate fad is to rent everything for tax reasons. As long as there's physical access to the hardware for hands on requirements, most of the time the servers will be administered remotely. It's no big deal if you have to pay $100 for somebody on-site with the premises provider to do a hard reset if that doesn't happen very often. It might be a good idea to make some direct connections with people there and prime them with cash money. It doesn't take much off-the-books cash to make friends. It can also be cheaper than sending your own personnel just to power cycle a box. Any logged work might be better to do officially.
Why aren't people looking at private cloud or CoLo or a mix of both? I have Managed service providers that will stick our VMs on their kit (Nutanix or equivalent) water and feed the hardware in their tier 3 datacenter, provide BCP/DR, backups networking etc etc all cheaper than running our own hardware on prem in a room in the building. Built in metrics, SLAs etc. for a fixed price per VM.
I get fixed cost for 3 or 5 years, bench strength if something goes wrong at 3am, my contract stipulates security requirements, 3rd party audits etc. etc. Best of both worlds and as long as we manage the provider, ensure they adhere to the contract we achieve safe secure reliable infrastructure for the same or less than it costs to run it ourselves.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
