https://www.ibm.com/investor/att/pdf/IBM-2Q23-Earnings-Press... IBM is getting killed in infrastructure, their margins look to be growing but losing ~15% of their revenue y/y. This seems like them just gouging the last of the customers who are either unwilling or unable to leave.
for context during the same period AWS grew 12% y/y.
I often wonder is AWS success due to paying top dollar to staff where as IBM seems to have gone down the cost cutting route and low salaries. I'm sure that isn't the only reason, but IBM's fall from the top has been pretty drastic over the last 30 years.
I guess it also could just be due to companies eventually get too big and bloated which causes them to innovate slower.
>I often wonder is AWS success due to paying top dollar to staff where as IBM seems to have gone down the cost cutting route and low salaries.
I think that's almost certain. Even if IBM finds and rewards the rockstars they have and promotes them quickly and gives them special treatment, i'm guessing a lot of them would prefer to work at Amazon or Google. Not only would they make good money, most of their coworkers would be of higher quality as well.
I bet IBM beats AWS handily on work-life balance, pressure from management, and most of the metrics of work satisfaction other than compensation and maybe office facilities
IBM probably doesn’t use Chime for meetings, either.
I always assumed a good chunk of developers derive satisfaction from either making software that is useful and used, or compensation. I don't see how IBM wins on either of those fronts.
I suppose for a narrow group of people might prefer IBM, but certainly not the majority of talent.
AWS is also charging a premium for the DDR5 hardware they just launched. The m6 family was the same price as the m5 equivalents. The m7 series is 5% more expensive on Intel, and while the AMD machines used to be 10% cheaper than Intel, they’re now 20% more expensive.
Funny, I was just looking at where to put a VPS and what is the most cost effective way if I don't care about getting the best performance.
Oracle has an always free tier, but it is hard to build much on 1 vcpu and 1 gb of ram. They offer $300 free credits so you can use some services for free until those credits run out, but their servers are at least double the price of others like Hertzner or Vultr and there are even cheaper ones if you are willing to take the tradeoffs.
I thought, why not check out IBM and see if they offer a free tier or credits? And, then was triggered by the "Free credits for Watson!" I don't know if they have a free tier or not, but that ad alone had me walk away. IBM's reputation over the last few years carries forth in my imagination.
> are at least double the price of others like Hertzner or Vultr
Unless you specifically need a lot of bandwidth then no.
Vultr has EPYC 3rd gen behind a more expensive paywall and don't have 4th gen announced yet either.
Hetzner is better, but again 3rd / 4th gen are only available for dedicated CPU plans. It's a bit of a lottery and there aren't many 4th gens in the mix. There is like a 40% performance difference.
So whereas with Oracle cloud you can specifically pick the latest generation and with preemptible instances it's not as bad as it looks. Definitely not 2x.
IBM had quite a good free tier on their Bluemix platform (which got ingested by Cloud), but in the end it was discontinued. A few years later I started getting bills for predicted server costs (which actually were never incurred) arriving at my parents' home that I had given as a sign-up address, but wasn't able to log into my account to close it altogether having altogether forgotten about my account, let alone the login details. And it was using my university email address anyway which of course I had no access to. In the end I pestered someone in the company enough that they just went in and closed the account for me.
Google Cloud also has a free tier (2vCPU / 1GB RAM). I use one to host Vaultwarden and Syncthing.. works great once you find your way around their console :P
Any bets on whether IBM will be able to keep afloat until 2035?
They're still big but they keep making one unforced error after another. One surefire way to get your customers to research the competition is by such immense price changes. This sort of thing is done gradually unless you are trying to get rid of a certain segment of your customers. Which may well be the case, but if it isn't they have a problem, and regardless you can't really predict how many of the customers you wanted to keep will leave as well. And that's before we get into reputation damage.
For me, IBM is a quantum play supported by legacy mainframe business. Probably not long for this world, but if their quantum system 2 announcement actually goes well, hype may serve diligent investors well.
Turning quantum into a business model that is not only self-sustaining but at the level of profitability of IBM's legacy business... that's a big ask for even the most agile company with the best technologists in the world. And if they internally knew their Quantum System 2 was game-changing - why do this desperate move in their cloud business?
If practical quantum does pan out, it will make this current AI situation look insignificant by comparison. It may also become extremely synergistic with the current AI movement. Quantum search will allow for us to seek out global minima across gigantic parameter sets.
> why do this desperate move in their cloud business?
Everyone is jacking up their prices. Microsoft would have to triple our cost before we'd think about tearing up our tech stack and moving again.
If the floor is the surface you are interested in and if you're not in space...
It's gravity that does the sorting for you, if you want to do this in space at a minimum you will have to have a spinning container and the surface you're interested in had better be the floor.
There's a significant gap of engineering effort between practical quantum at the scale of "rent time on the small handful of systems in the world that can do this" vs. at the scale of "anyone can rent or buy hardware to do this at scale."
IBM may possibly be able to recoup their investment in quantum by whale-hunting with the former (and certainly some of the crypto-breaking characteristics will make that quite lucrative to sell to state-level agencies), but it will likely be decades before every company in the world is asking the question "how can quantum help us" the way they are with generative AI.
Meh, if you've been following their day to day business, mainframes are just one part.
IBM now is primarily a big IT consulting business (think Java, not zOS), one of those semi garbage shops like HP Consulting, Oracle Consulting, TCS, etc.
They're a 2nd or 3rd tier company as far as job quality and compensation is concerned.
I wouldn't worry about them, these companies will probably outlive the both of us.
The company will be around, but it might only be in the form of a patent portfolio and government contractor. No telling if the government contractor will have their own data centers or resell other clouds.
They have ironclad contracts with the DoD and other government organizations, and with the way those processes work if every person from the president to the peon decided they wanted to move everything off of IBM and were going to move heaven and earth to do it. They wouldn't even be able to start entertaining bids from other competitors until 2035.
Beyond this IBM's core competency is winning government contracts, a niche very few other people have the desire or the resources, or the proper understanding of the necronomicon to enter.
You are not forced to use the "poison" - you can get them as LinuxONE machines and run everything on top of a very thin hypervisor (which is necessary, since nothing runs on bare metal on the newer machines).
AFAIK, the LinuxONE boxes come without lock-in - what you pay for is what you get and there aren't any monthly subscriptions as there are with z/OS and the various pay-as-you-go features (besides, of course, the maintenance contract you WILL want to have for a multi-million dollar machine that uses mostly bespoke hardware).
Don’t dismiss the possibility of IBM getting bought out for those easy cash flow government contracts. Someone like Amazon or Oracle could potentially goggle them up in 5-7 years when their stock price grows and IBM’s stagnates or drops.
I have a feeling US regulators wouldn't look too favorably on any IaaS mergers. Not to say it couldn't be done, but there would most likely be pretty high regulatory hurdles if any 2 major public cloud providers propose a merger or buy out
Full disclosure, IBM Systems employee here. I honestly think the server and mainframe development will remain regardless of what else happens. There's a niche between linux x86 box and supercomputer that won't likely disappear.
I think it's not accurate to call their architecture a cluster of machines. Either way I'm not referring to Linux as an OS - it is really the only choice for enterprise scale computing.
Regardless of what architectures supercomputers use, their use cases are different from that of server/mainframe high performance compute. 8 cores with on-chip AI acceleration running at 5GHz sharing virtual cache, using ~256GB-1TB of of RAM at ~76-304Gbps transfer speed is something a major bank would use. It won't cut it for SoTA atmospheric modeling, molecular modeling, or something like rapid computational chemistry search space exploration. What it excels at is the ability to handle a blazing number of "ordinary" and AI inferenced transactions per second. Then there's the SAP HANA where entire workloads have data fully stored in the gargantuan RAM.
Could I be neglecting to foresee certain trends in technology or industry? Perhaps. What I'm doing here is highlighting that the current niche is well served by POWER and Z. I don't see the current users switching entirely to Cloud - that ship sailed 5 years ago.
Good points. Essentially Linux is the delivery mechanism for the power potential of the mainframes so that Linux is a convenience and not a sign that the tables are turning. It also helps to be able to run a mountain of software without having to develop it first for some proprietary OS. This is where the big difference between transaction processing software and massive parallel problems such as simulations lies. A bank doesn't need a supercomputer, it needs a reliable and scalable transaction processor.
Off topic, and I'm sure you are not in a position to comment, but I find the 'Watson' marketing ploy a massive turn-off, it is so tone-deaf that I really don't understand why they keep pushing it so hard.
I'll be honest I haven't seeked out how Watson X is being projected externally, and don't know much about the details internally - it's being driven by a different division of the company. So even if I were in the position to comment, I don't believe I have anything meaningful to say there.
You touched on a point I meant to add - reliability. Something that sets the Z mainframe apart from the Power server is the unyielding quest for absolutely no downtime (in exchange for compromise in raw compute/memory per drawer vs Power). I believe the quoted number is 3 seconds per year maximum. Because having huge throughput is well and good for business, but lost transactions could really strike at the bottom line with an unpredictable ripple effect. And in the case that a network of major hospitals is using the mainframe - no downtime could potentially save lives.
Exactly. I worked for a bank (using IBM hardware, as well as Sperry) in the good old days and back then the reliability factor is what drove each and every major purchase. Tandem made their whole business case around that. When the computers and the IT department surrounding them are a small rounding error on the total budget of your organization it makes sense to go for a solution like that and I guess there will always be customers that need that. Consider me convinced: IBM will still be around in 2035, or at least, the mainframe division will be. It also puts their RedHat acquisition in a different light, so thank you for that bit of insight.
Not IBM employee nor Shareholder but Agree. And even if they may disappear, it is obviously not before 2035. I am even willing to bet not even 2045.
People asking these questions obviously have zero idea of IBM's mainframe and server business. May be it is worth mentioning, as little as 5 - 6 years ago more than half of HN have never heard of SAP.
Like Lowes. Home Depot might go out of business if the market crashes hard enough, but Lowes is protected as they're a GSA contractor. The Federal government will never let them fail. Too essential.
They will be, their mainframe business is still pretty central to all things banking related. And I'm pretty airlines use them for critical services as well.
From a financial perspective IBM's model of buying up customer's by taking over smaller successfull outfits, maximizing profit's while the quality goes down the drain and using that profit to restart the process have actually been successful enough over the last 2 decades since they switched from being an engineering company to an capital fund with a tech focus.
This is probably just the same strategy ie IBM might have realized that their cloud department is as popular as it will ever be and is now entering the squaze phase of it's lifecycle.
Without a problem, because HNers keep ignoring the whole set of services, patent records per year, R&D being done, its domain across Java, mainframe, micros, being the only UNIX that survived Linux, being one of the major Linux backers, being the main GCC driver while clang lags behind while Apple/Google backed away and everyone else enjoys its license, the xl toolchain used in some supercomputers...
Starting in 2019 or so, they dropped any ability to manage cloud resources using the Softlayer-branded portal and forced everyone into using the IBM Cloud API/Console.
Why do you people think that anything that isn't discussed constantly on HN is dying or dead? Their cloud revenue was $22 billion last year and climbing.
> Why do you people think that anything that isn't discussed constantly on HN is dying or dead? Their cloud revenue was $22 billion last year and climbing
Because so many vendors decided to jump into this space, and there's a track record of using accounting tricks as a substitute for market success. Azure, for example, is believed to be obscured[1].
Even if their cloud revenue were half of what they reported, their offering would still be several times larger than DigitalOcean, Vultr, Linode, and several others combined. No one is claiming that IBM is the largest cloud provider, but come on.
(Disclaimer, I'm an IBM employee but not in cloud)
This feels like kind of a clickbait article. It looks like single-digit percentage price increases for non-US international customers using IAAS. It's the 25% and 26% increase in prices for flavors of archive Cloud Object Storage.
I used to work at an AIX shop. I gotta say, AIX is pretty solid. Back in the late 90's, it had features like volume management long before LVM was available in Linux. They also had a great virtualization story early on (LPAR.)
I don't think a portable (or luggable) POWER10 machine is in the cards.
OTOH, a single SMT8 core (or two SMT4, firmware-defined) consumes about 20W, so it could be possible to manufacture a few-core module that could fit inside a sensible laptop.
My Dells with Linux are quite reliable though, even if not as fast or good looking as the Apples...
I worked on an initiative within IBM that was forced to use IBM Cloud. Not only would they not be my first choice, they wouldn’t be in my top 10 choices. And I can’t even name 10 cloud providers.
They offer some interesting products that are completely unique to them. If you need to run on IBMi, z/OS and AIX, they are your only option that doesn't require you to upfront buy hardware.
Their offerings could be a little bit better and more orthogonal though - they could offer Linux on POWER as VMs the same way they do with Linux on Z, as well as AIX and IBMi.
94 comments
[ 3.0 ms ] story [ 172 ms ] threadfor context during the same period AWS grew 12% y/y.
This course of action is not surprising looking at their net income and profit margin trends:
https://www.macrotrends.net/stocks/charts/IBM/ibm/net-income
https://www.macrotrends.net/stocks/charts/IBM/ibm/profit-mar...
I guess it also could just be due to companies eventually get too big and bloated which causes them to innovate slower.
I think that's almost certain. Even if IBM finds and rewards the rockstars they have and promotes them quickly and gives them special treatment, i'm guessing a lot of them would prefer to work at Amazon or Google. Not only would they make good money, most of their coworkers would be of higher quality as well.
IBM probably doesn’t use Chime for meetings, either.
I suppose for a narrow group of people might prefer IBM, but certainly not the majority of talent.
https://www.prnewswire.com/news-releases/aws-and-azure-cloud...
Oracle has an always free tier, but it is hard to build much on 1 vcpu and 1 gb of ram. They offer $300 free credits so you can use some services for free until those credits run out, but their servers are at least double the price of others like Hertzner or Vultr and there are even cheaper ones if you are willing to take the tradeoffs.
I thought, why not check out IBM and see if they offer a free tier or credits? And, then was triggered by the "Free credits for Watson!" I don't know if they have a free tier or not, but that ad alone had me walk away. IBM's reputation over the last few years carries forth in my imagination.
Unless you specifically need a lot of bandwidth then no.
Vultr has EPYC 3rd gen behind a more expensive paywall and don't have 4th gen announced yet either.
Hetzner is better, but again 3rd / 4th gen are only available for dedicated CPU plans. It's a bit of a lottery and there aren't many 4th gens in the mix. There is like a 40% performance difference.
So whereas with Oracle cloud you can specifically pick the latest generation and with preemptible instances it's not as bad as it looks. Definitely not 2x.
Oracle will throw away your running VM on free tier no warning.
Mine got thrown away so hard entire account was fucked and I didn't even had an option to upgrade to paid account...
It is still pricing that VM at about $9 per month. I'll play with it and see.
They're still big but they keep making one unforced error after another. One surefire way to get your customers to research the competition is by such immense price changes. This sort of thing is done gradually unless you are trying to get rid of a certain segment of your customers. Which may well be the case, but if it isn't they have a problem, and regardless you can't really predict how many of the customers you wanted to keep will leave as well. And that's before we get into reputation damage.
If practical quantum does pan out, it will make this current AI situation look insignificant by comparison. It may also become extremely synergistic with the current AI movement. Quantum search will allow for us to seek out global minima across gigantic parameter sets.
> why do this desperate move in their cloud business?
Everyone is jacking up their prices. Microsoft would have to triple our cost before we'd think about tearing up our tech stack and moving again.
Just like pouring water on the floor will find the global minima on it. Just wait until there is one puddle left. O(1).
I am yet to be convinced quantum computers are real at all and not just some slow convoluted variant of an analogue computer.
It's gravity that does the sorting for you, if you want to do this in space at a minimum you will have to have a spinning container and the surface you're interested in had better be the floor.
IBM may possibly be able to recoup their investment in quantum by whale-hunting with the former (and certainly some of the crypto-breaking characteristics will make that quite lucrative to sell to state-level agencies), but it will likely be decades before every company in the world is asking the question "how can quantum help us" the way they are with generative AI.
IBM now is primarily a big IT consulting business (think Java, not zOS), one of those semi garbage shops like HP Consulting, Oracle Consulting, TCS, etc.
They're a 2nd or 3rd tier company as far as job quality and compensation is concerned.
I wouldn't worry about them, these companies will probably outlive the both of us.
Beyond this IBM's core competency is winning government contracts, a niche very few other people have the desire or the resources, or the proper understanding of the necronomicon to enter.
Shit man, IBM will live on for centuries (assuming we get that far)
The poison I'm referring to is the IBM bean counters and lock-in not the software
Am I wrong?
Regardless of what architectures supercomputers use, their use cases are different from that of server/mainframe high performance compute. 8 cores with on-chip AI acceleration running at 5GHz sharing virtual cache, using ~256GB-1TB of of RAM at ~76-304Gbps transfer speed is something a major bank would use. It won't cut it for SoTA atmospheric modeling, molecular modeling, or something like rapid computational chemistry search space exploration. What it excels at is the ability to handle a blazing number of "ordinary" and AI inferenced transactions per second. Then there's the SAP HANA where entire workloads have data fully stored in the gargantuan RAM.
Could I be neglecting to foresee certain trends in technology or industry? Perhaps. What I'm doing here is highlighting that the current niche is well served by POWER and Z. I don't see the current users switching entirely to Cloud - that ship sailed 5 years ago.
Off topic, and I'm sure you are not in a position to comment, but I find the 'Watson' marketing ploy a massive turn-off, it is so tone-deaf that I really don't understand why they keep pushing it so hard.
You touched on a point I meant to add - reliability. Something that sets the Z mainframe apart from the Power server is the unyielding quest for absolutely no downtime (in exchange for compromise in raw compute/memory per drawer vs Power). I believe the quoted number is 3 seconds per year maximum. Because having huge throughput is well and good for business, but lost transactions could really strike at the bottom line with an unpredictable ripple effect. And in the case that a network of major hospitals is using the mainframe - no downtime could potentially save lives.
People asking these questions obviously have zero idea of IBM's mainframe and server business. May be it is worth mentioning, as little as 5 - 6 years ago more than half of HN have never heard of SAP.
Like Lowes. Home Depot might go out of business if the market crashes hard enough, but Lowes is protected as they're a GSA contractor. The Federal government will never let them fail. Too essential.
This is probably just the same strategy ie IBM might have realized that their cloud department is as popular as it will ever be and is now entering the squaze phase of it's lifecycle.
Softlayer is for example, owned by IBM
Starting in 2019 or so, they dropped any ability to manage cloud resources using the Softlayer-branded portal and forced everyone into using the IBM Cloud API/Console.
Because so many vendors decided to jump into this space, and there's a track record of using accounting tricks as a substitute for market success. Azure, for example, is believed to be obscured[1].
[1]: https://www.bigtechwire.com/2023/06/30/microsoft-azure-gener...
This feels like kind of a clickbait article. It looks like single-digit percentage price increases for non-US international customers using IAAS. It's the 25% and 26% increase in prices for flavors of archive Cloud Object Storage.
Which is rather odd - they are increasing prices on the most commoditized products they have.
I want a solid UNIX ecosystem with power10 to compete with macOS + Arm. I’m tired of there being only one real option for a reliable unix laptop.
There is a reason why it is the biggest survivor from Linux conquest on the server room.
IBM managed to succeed where Sun, HP and SGI, among others, failed.
Ah, I see you like pain.
OTOH, a single SMT8 core (or two SMT4, firmware-defined) consumes about 20W, so it could be possible to manufacture a few-core module that could fit inside a sensible laptop.
My Dells with Linux are quite reliable though, even if not as fast or good looking as the Apples...
Personally, they wouldn’t be my first choice. Only if I had a gun to my head
Their offerings could be a little bit better and more orthogonal though - they could offer Linux on POWER as VMs the same way they do with Linux on Z, as well as AIX and IBMi.
They built notebooks and pcs, but sold that to Lenovo years ago, IBM is still a billion dollar company, but I have no Idea what they do.