Interesting. My observation on IBM is their entire business model is:
1 - Audit your customers
2 - Buy back shares
3 - Force early retirements
It was easy to see why Watson failed in that environment. The revenue was “We’ll let you out of the $6mm audit bill if you buy $2mm of Watson”. Companies would agree, install better asset management, and never put Watson into production.
I couldn’t imagine Quantum Comouting surviving there. Spinning it off the best play.
That's Kyndryl: They spun it off into it's own entity after "IBM Global Services" had such a (deservedly) poor reputation that they were scraping the bottom of the barrel for clients and employees. Not that Kyndryl is any better, but it's enough of a rebrand that you might fool decision makers for the few minutes that it takes to get them to buy in.
they've just spun off the first quantum computer pure-play, the highest of techs, the greenest of green-fields, and your observation is "IBM doesn't do this"
Seems like tacit acknowledgment that IBM mothership is not the right place for a speculative growth play from both a management and capital perspective.
I’m not IBMologist but I do remember how IBM pushed Watson when it was clear that upper management had no idea what Watson actually was. Regardless of the viability of the underlying technology, it’s best to keep such things away from the consultants.
Also, article is very difficult to read. Bad typeface, spacing, coherence and prose. I found the press release less strained.
It's not that the cheating stayed with IBM: Ray Dalio hired David Ferrucci out of Watson to try to make an AI for Bridgewater. The pitch was to make it a very accurate people assesor, but in practice the goal was to tell you who agreed the most with Ray Dalio. The team spent years of their lives taking Bridgewater's money, building basically linear regression on questionnaires, and calling it advanced AI on interviews. It's all documented in The Fund.
Eh, Watson was a classic open domain QA system originally, no deep learning or much of what we think of in an "AI platform" today. It was one of a bunch of such systems that were built in that early 2000s period. They all failed because the approach fundamentally didn't work very well.
> Seems like tacit acknowledgment that IBM mothership is not the right place for a speculative growth play from both a management and capital perspective.
I'm not understanding your logic, can you explain?
What I see with the program and amounts companies were awarded is some level of acknowledgment of the current state of quantum research (i.e. IBM is generally considered the leader) and their pragmatic approach that piggy-backs on current technologies (for obvious speed+cost benefits).
You must not talk to competent people. IBM is very experienced at this grift. I remember when I used to go to conferences in a different field and IBM would announce "state of the art" results that were very obviously done by cheating (making an ensemble model and tuning the weights on the test set). Everyone doing real work would ignore them, and then they'd go sell to clueless midcap companies on the basis of that announcement.
> I’m not IBMologist but I do remember how IBM pushed Watson when it was clear that upper management had no idea what Watson actually was. Regardless of the viability of the underlying technology.
So pretty much like any other AI company in 2026 hunting for VC money?
IBM still sells extremely POWERful systems, but they don't seem particularly interested in expanding the market.
I once had a conversation with a director of that division about why it wasn't on the market. It basically came down to the existing customers being willing to pay such exorbitant amounts for each system after all the support contracts that "normal" markups like Nvidia and Intel enjoy were too paltry in comparison.
They also charge you for every instruction cycle on the machines (look up MIPS licensing) you own. Imagine if NVIDIA started doing that with their GPUs: spend $2500 on a GPU and then pay NVIDIA a royalty fee for every hour of workload you put on it.
Two different product lines. POWER systems are sold by the socket, on the higher end sometimes you get some license disabled cores but it's usually trivial. Mainframes are the opposite, you usually get the full monty machine that is extremely limited by licensing.
they've also still got their storage stuff. I always wondered why that isn't doing better, it seemed pretty damn good when I've ended up working iwth it.
It massively depends on what type of quantum chip it is, so often the confusion is that there is no one specific method. Especially as we are too early in the development cycle to have a clear winner (although Photonic computing is gaining traction).
I try to follow the motto of my university quantum physics professor: if you feel like quantum mechanics is making you more certain you understand it, you’re probably wrong and need to start over.
But as I understand it, the most "basic" approach now with qubit-based computers is to select the most popular answer across many runs and treat it as the "right" one.
The real story isn't the $2B. It's that the foundry is standalone, so other quantum hardware companies can use it. Shared infrastructure beats nine separate research cleanrooms.
I guess it's a balance. If you think their process makes workable chips for your designs, then you can use it. If you can't adapt your design to what they can build, then you need to build your own foundry. Chances are a reliable supplier will push the market in the direction of their process.
If we had someone making GaAs processors in the 1980s for a price competitive with their silicon counterparts and with a long-term roadmap, we'd have very different computers now. And some extra toxic waste problems.
I've been out of the space for a bit. IBM has been betting on the engineered superconducting approach, which makes sense given their background, but there are other options, often for potentially different problem areas. Need to dive back in.
The superconducting approach is great in principle but has so many issues that need to be solved, from cooling, to energy to cool, to max number of qubits before you can't cool to operating temperature, before optimal connectivity to rest of system and so on. I am of the view photonics is the way forward but as you say, it will depend on the task at hand partially.
There is high agreement on what the real applications of Quantum computing are. Unfortunately these projects are basically useless when it comes to them.
Can you clarify? Do you mean that superconducting qubits are unable to perform the "real applications" theoretically, or that superconducting qubits at the scale this foundry could produce will be unable, or that superconducting qubits that will foundry could produce will still be outperformed by classical techniques?
I mean, we are no where near the scale [qubit count] & quality where the applications apply. Not just this foundry but in general. I suppose the point is to eventually get there, but we are not close yet.
You should still view anything Quantum as early R&D.
I don't have the same level of cynicism with quantum that I had with enterprise blockchain. (Hey, I spent a number of years getting sucked into things that didn't pan out along with some that did in a big way.) I pretty much agree with respect to quantum. Practical value is probably further away than a number of folks were betting on at one point though I still believe it's there.
> You should still view anything Quantum as early R&D.
The good thing is that someone who can make lots of chips can reduce the effort it takes to do R&D. With more people researching possible applications, it's likely we'll progress more quickly.
The most obvious one is SIGINT agencies breaking RSA, DSA, ECDSA, ECDH, etc.
Of course, the plan is by the time quantum computers become capable of breaking those algorithms in practice, the industry will have moved to post-quantum cryptography algorithms.
But there will still be legacy systems which haven't, and also encrypted data recorded in the past in the expectation they'd be able to decrypt it in the future.
There seem to be fixes available for crypto. The issue is getting people to implement those fixes. Which, of course, is the issue with getting people to implement a lot of security fixes more broadly.
Is that actually true in the real world? Or is that some comp sci algorithm dream? I suspect it might be an engineers fallacy where the romantic desire to reduce everything to an algorithm or scalar value that can then be maximized or minimized blinds the engineer to the reality of the situation - the businesses doing route planning already have something thats close enough to optimal so that if the travelling salesman problem was solved, it wouldn't make a material difference to the business.
The algorithm engineer is so in love with the idea that an algorithm is the solution to everyone's problem (its a natural human bias to think the world desires what we have) that they way overweigh the importance of route planning improvements which are incremental or worse - would be thrown away because the practicalities of implementation doesn't warrant the marginal improvements.
Absolutely true in the real world; I was part of a real team that explored quantum optimization algorithms as part of a strategic initiative (my day job is algorithmic optimization on classical computers).
Our problem is similar (but not identical) to the traveling salesman problem. We run on a tight time constraint (measured in days for the complex type and measured in minutes for the simple type).
We're running approximations on classic computers but estimate that we'd save billions if we could reach optimum.
>IBM is developing four custom ASICs — a decoder, a two-qubit gate controller, a single-qubit controller, and an amplifier — designed to handle quantum control at scale, with these circuits expected to converge around 2029 at the point where power consumption becomes manageable at up to 3 megawatts per system.
The current hotness seems to be based on creating pairs of entangled qubits based on what might be realistically achieved with error correction. Shor's requires thousands of entangled qubits (something like 4000 for 2K RSA and 1500 for 256 bit elliptic curves).
So unless someone comes up with a way to break cryptography using pairs of entangled qubits then this probably isn't relevant.
I would like to believe this is a cover story for IBM to make parts for the DoD latest weapon . I would like to imagine its a cover to make parts for some new government super computer. But its IBM its probably nothing even close to that . Its more likely a nice way to get the stock back at $315 and make nothing.
The article talks about IBM spreading bets to other techniques. Reminds me to ponder again. Has Microsoft retracted their sketchy quantum claims about inventing new states of matter in the past year? https://www.theregister.com/on-prem/2025/03/12/microsofts-qu...
I am not surprised, but disappointed, to see something like the CHIPS Act be used for something which is still in ultra-super-unbelievably-early-research-phase. Put more candidly, something not currently useful like Quantum computing.
1 - Audit your customers
2 - Buy back shares
3 - Force early retirements
It was easy to see why Watson failed in that environment. The revenue was “We’ll let you out of the $6mm audit bill if you buy $2mm of Watson”. Companies would agree, install better asset management, and never put Watson into production.
I couldn’t imagine Quantum Comouting surviving there. Spinning it off the best play.