Shahin Khan, co-founder of OrionX, joins Yuval Boger to explore the intersection of quantum computing and high-performance computing (HPC). Shahin discusses why HPC is a natural early adopter of quantum technologies, the role of QPUs alongside GPUs and CPUs, and how quantum computing aligns with global megatrends. They delve into scaling challenges, the potential for quantum to revolutionize tensor-based problems, and the broader implications of quantum on energy efficiency and scientific discovery. Shahin also reflects on the industry’s progress, the importance of rational exuberance, and the need to set realistic expectations while maintaining excitement about quantum’s transformative potential.
Transcript
Yuval Boger: Hello Shahin, and thank you for joining me today.
Shahin Khan: It’s great to be here, Yuval. Thank you for the invitation. Honored to be here, really.
Yuval: The honor is mine. So who are you and what do you do?
Shahin: Right, excellent question to start with. So my name is Shahin Khan. I’m part of a very small consultancy called OrionX. You know, Google X, SpaceX, OrionX, why not? And the URL is OrionX.net.
We are an industry analyst consulting firm focused on technology, policy, and actually marketing too, which is, I know, something that you are intimately engaged in and familiar with. And unlike most industry analyst firms that pick a topic and focus on it, because of our presence in Silicon Valley and the variety of companies and startups that we work with, we decided that we’re going to pursue the major mega trends of the day and pick like a half a dozen, you know, a handful of them and drive deep into that.
And that means, right now it means IoT, 5G, HPC, AI, quantum computing, cryptocurrencies, especially Bitcoin, and to some extent a little bit of space tech because it’s kind of coming our way.
Yuval: So let’s talk about quantum and HPC. How quickly do you see HPC centers adopting quantum and what’s stopping them from doing it faster?
Shahin: Right, so as you probably know, I believe, and our roots are in HPC. Most of my colleagues, we all grew up in the HPC world. We worked at companies like Cray Research in the old days or part of the various companies that have been driving it. So maybe it’s a bias in favor of HPC. Maybe it’s just like, you know, deeper knowledge of what’s going on in HPC. But I believe HPC is the main and the first destination for quantum computing. I believe quantum computing is highly aligned with the problems that HPC solves, the mindset that the HPC community has, and the kind of technologies that can come to bear. So I think it’s going quite fast, in fact, within the HPC world.
Yuval: Where do you see that adoption? I mean, we see a lot of government programs, of course. We see some of the national labs get into quantum, but do you see that in corporate HPC centers?
Shahin: I do. And, you know, most of the HPC community really is driven by public investment. And the private investment has more of a stringent ROI attached to it, where the problem needs to be identified already before the technology gets employed to address the problem. But where research and development reigns, and that’s the government labs, national centers, academic centers that are notable, that stuff is, that area is really going very fast. So within the US, you’ve got the Oak Ridge, Argonne, Lawrence Berkeley, Sandia, and, you know, Brookhaven, all of them have non-trivial efforts to advance quantum technologies in general and quantum computing in particular.
US cloud providers are probably the best example of commercial HPC centers that have an increasing emphasis on HPC in general and quantum computing as a way of both differentiating what they offer, but also enable companies to get started with quantum computing and perhaps even run some applications.
Yuval: But quantum computers today are, on one hand, very valuable, people pay a lot of money for them, but at the same time, quite useless. I mean, there’s not a lot of problems that you can solve on a quantum computer you cannot solve classically today.So shouldn’t an HPC center be just waiting and saying, well, show me a problem that you can really do on a quantum computer and then I’ll get one?
Shahin: You’re sounding like Jensen. Well, I think when you say useful, you have to say useful for who and useful for what. I don’t think anybody in the quantum computing world is claiming that quantum computers are ready for an IT data center production environment for some bank doing transaction processing, right? That isn’t the reality today. Maybe it will be as things get infused with other sorts of tasks.
But for those who want to advance the technology, for those who want to participate and not miss out, because when advances happen, they can happen rapidly, for those who want to start building applications and be prepared, and indeed, as you know, and as you get into the quantum world, sometimes you come up with quantum-inspired algorithms that are better than classical algorithms and actually advance the nature of science.
The other aspect of it is that quantum computing generally in my view has three potential benefits. One is speed, where we all focus on, but it’s also accuracy and it’s also energy savings. And as GPUs, which are like the main alternative right now, become hotter and hotter and hotter and don’t look like they’re stopping, there comes the point where you say, you know what? If it’s half the performance, but 20 times less energy, maybe I’ll take it and I’ll just like, you know, double up the resources. So that potential is also there.
And I think all of that comes together to say that, okay, this is a technology that we must invest in, we must stay on top of, the progress is significant and real, what we need to do is also significant and real, but it makes sense to participate.
Yuval: A few months ago, I wrote an article, and the title of the article was that the QPU, the quantum processing unit, the QPU was the next GPU. The next day I get these calls from my friends at Nvidia, they say, what do you mean the QPU is the next GPU? No, we think they work together and so on. And next time, maybe you should phrase it differently unless you believe that QPUs are going to replace GPUs.
We’re still good friends, so no problem. But do you think that QPUs are going to take away some of the calculations, some of the computations that are done on GPUs, or they’ll just enable completely new type of problem solving?
Shahin: Yeah, well, I think the moment where we have real quantum advantage is the moment when nothing else is going to just compete at all. But while we’re waiting for that, GPUs, VPUs for vector processing units, and CPUs are all alternatives. And really, if you look at the history of HPC, you could say that the formulation of the problems that are being solved in HPC started out with scalar. Let’s just program it. And that’s like the old Fortran codes. And that all ran on CPUs. Then with Seymour Cray and CDC and Convex and Alliance and FPS, the mini supercomputer era, and led by Cray as a big supercomputer and CDC as a second there. And also, Hitachi, Fujitsu, NEC, Siemens in Europe, many others, those were all after vector processing, where you went from scalar to vector. Right around, I would say in the early 1990s, we started seeing matrix co-processing, MCP, matrix co-processing was a thing. And it was really focused on solving matrices. At the same time, graphics co-processors were necessarily doing a lot of four by four matrix multiplications to do the transformations, the scale, translate, rotate, transformations that you need to do in the graphics world. So that was the beginning of GPUs.
So you could say that GPUs are really optimized for matrix processing, but it’s not like vectors can’t do matrix, and it’s not like matrix can’t do tensor. So as you go from matrix to tensor processing, you can have GPUs, you can have GPU extensions to CPUs, like CPUs already have an MX for matrix extensions, they have VX for vector extensions, but if QPU shows up as a legitimate and effective solution for tensor math, that is really going to differentiate it from matrix algebra moving to tensor algebra. And I think it will occupy a prominent place.
I think it’s going to be coexistence, an ideal scenario is that you just apply the right accelerator to the right portion of your code, but it’s coming, and I think that’s what we should expect.
Yuval: So would it be correct to say that people who are promoting QPUs should actually look for these problems that are essentially tensor problems instead of trying to reconfigure QPUs to do what GPUs or CPUs are doing?
Shahin: Well, I think GPUs are already doing some of that tensor math, and of course, they have tensor cores now, and Google’s GPU is called TPU for tensor processing unit.
So there’s a recognition that there is a class of nonlinear problems that require tensor math, and that it can be reduced to matrix algebra, but if you don’t have to reduce it, it could be more efficient, and I think quantum machine learning is after that particular base. But also, you have to look at what quantum transformations are good at. If quantum transformations are formulated in tensor math naturally, which they are, then they lend themselves to that.
I mean, really, if you take Shor’s algorithm aside, which is prime factorization, and it has spawned its own kind of mini industry, if you look at the other applications, both optimization as well as quantum physics, quantum chemistry, all of them really are aligned with that statement, is that, hey, we are really, at the end, looking at Hamiltonians and Hilbert spaces, and that’s fundamentally an energy optimization formulated as tensor algebra, which can be reduced to matrix algebra, which can be linearized and solved, all of that, so QPUs can play a very prominent role there, and I think that’s something that we should expect.
Yuval: You mentioned Cray and a bunch of other companies that don’t exist today. I mean, they do exist, but they exist sometimes within other companies. Cray, famously, is owned by HPE, I believe.
Shahin: Yes.
Yuval: Do you think that quantum computing companies are going to stay independent, or do you think they’ll be swallowed by cloud or HPC or GPU providers?
Shahin: Yes, yes, well, for those of you who attend the supercomputing conference every year, and it’s going to be in St. Louis next year, the Sunday before the show, I have an informal gathering with old friends that we lovingly call the dead architecture society, because, in fact, of all the mini supercomputers and supercomputer companies that are no longer there.
But look at Nvidia. Nvidia is a GPU company. There were other accelerator companies that didn’t pan out, and, you know, Nvidia is now one of the top three valuation companies in the world, going from strength to strength. So I believe it really depends on how the game is played. I think good strategy plays a part. I think luck plays a part, too. And who knows? I think it’s entirely, you know, conceivable. I think it is also very attractive for a small company to take a really good deal and be sold, because why not? You kind of cash out and that’s a good exit for you.
But there are also examples where that doesn’t happen. I think Facebook is a good example of not selling when people made them a very attractive offer. But Groupon is another example of an offer maybe they should have taken, right? So it’s hard to tell. But I think that’s possible very much.
Yuval: Earlier this week in the White House, there was a $500 billion program announced for AI. Does that take away budgets from quantum computing, or does that actually provide a chance that a small 10% of that, only 50 billion, would be allocated to quantum?
Shahin: I would like to believe it’s the latter, that if you have that much money going into something, and if the something that you’re investing it in is not flawless, I think the energy requirements, the power cooling, power transmission, I mean, power provisioning, power transmission, taking heat out, transmitting the heat out, those are all absolutely non-trivial problems within the AI world.
And if quantum computing can offer you an alternative that is significantly less energy reliant, you have to look at that. And I think investing in that makes perfect sense. We saw that in the CHIPS Act, the Chips and Science Act that was put into law in 2022 was at the high level about chip manufacturing. But if you double clicked on it a couple of times, it included quantum science and investment in quantum science. Some of it translates into semiconductors as well, but some of it does not. So I believe that the big companies with really serious big dollars would do well to hedge their bets, continue to look at the entire spectrum of science. Obviously, neuromorphic chips are a way of getting to lower energy, but they really haven’t panned out. Quantum computing does that plus more. So my expectation is that investment in quantum computing will carry on. Certainly for governments, there’s really no downside. I mean, even if for the sake of argument, it doesn’t work at all. Just the fact that you gathered all this massive talent in one place is a good thing for society. It will do other good things, right? And then other technologies come out of it. And in fact, the advances in the technology are very real and very significant and very impressive.
Yuval: You mentioned massive talent in one place and when people were writing about this $500 billion proposed investment, they were comparing it to the Manhattan Project, which I think in inflation adjusted terms was only a mere $30 billion. So the question begs, do you think there is a need or an opportunity to create a quantum equivalent of a Manhattan Project to create some quantum supercomputer?
Shahin: Yeah, I don’t think the analogy is really valid beyond a meme really. If you want to put a label on something big that you’re doing, Manhattan Project is a very attractive verbiage, but it’s not the same. The beauty of computing is that it’s a meta technology. It can be applied to every other technology. It’s not a bomb. I mean, the bomb led to other advances in an indirect way, but itself, it really was just a bomb that you hope to never use, right? And fingers crossed that will remain the same.
But when you look at computing, it’s a different ballgame. And I think it’s in some ways bigger than a Manhattan Project because it is really setting the tone for the economic foundation at a global level for the next many, many decades. And that of course leads to geopolitics. We coined the term technopolitics and we’ve seen the early examples of techno-politicians because that’s where all the advances are. So I think that it’s a big thing. And I think that quantum computing plays a significant role in this portfolio. It’s not the only thing, but I think it’s a very important pillar.
Yuval: Listeners of this podcast cannot see you on video, but they should trust me that I think you’re above the drinking age. And I hope you’re not offended, but
Shahin: – Not at all. I take it as the compliment that it is.
Yuval: So what have you learned about the quantum industry in the last year that you didn’t know before?
Shahin: Oh, right, right. Well, I don’t think I’ve seen anything that has changed my perspective from before that. I’ve been sort of consistent for the past several years that when you deal with a technology where somebody like a Richard Feynman says, “I’m having difficulty understanding it. Like what chance do I have?” Right?
So there continues to be a very big amount of complexity in the technology. We’re still in a superposition of research and development, as somebody said on LinkedIn, which I very much liked. We are still in the zone where, again, somebody else said that it is simultaneously overhyped and underestimated, that also continues to be true. My joke is that for all questions about quantum computing, the answer is yes and no, right? So there’s all of that.
But at the same time, the progress that’s been made in error correction and you and QuEra were a big part of that just about a year and change ago, and additional advances have been made since then, that was significant. I think the various modalities that are being pursued are all valid and important. And I think I echo those who think we need to invest in basic research and continue to invest in basic research. I think that’s a major challenge in science in general.
But a market exists, right? I mean, various estimates have the quantum computing market as about a billion dollar market. That’s not trivial. And people who are spending that money are getting uses for it. They know what they’re getting into. So in general, I think the challenge for this industry is to continue to make real advances and stop at the passion and exuberance and not cross that line. I think you do need passion. You need to be committed. You need to be excited. You need to never say no, never say never. We’re going to make it happen. That sort of commitment is important, but it’s also important to set expectations properly, to recognize who the audience is, to not throw things out without guardrails. I think some of that has happened over the years, and that’s unfortunate because I think it backfires. But in general, the market, you couldn’t ask for a better market with the science and technology and the depth. I mean, you were on stage at the Q2B conference in December 2024. Where else are you going to get that sort of confluence of massive talent, as I’ve said before, that is trying to solve a really big problem? I think that’s just wonderful for humanity.
Yuval: So rational exuberance, that’s the key.
Shahin: Well said, well said.
Yuval: If we look at the progression of other technologies, computers started as the mainframes, and then they went out to the edge more and more and more. We see AI moving to the edge now. How long do you think quantum computing is going to stay at the center, as opposed to moving it to the edge? And I’m ignoring quantum random generator chips on phones and so on.
Shahin: Well, first of all, I’m in agreement with you. I think we’ve been witnessing what I call the big bang in computing, that we had the mainframe and a dumb terminal, and then it went to client-server, then it went to mobile cloud.
And for like five seconds, we thought you have an iPhone at one end, a cloud at the other end, and you’re done, but you are not done. So now what you’re getting is essentially a digital fabric where there are nodes in the fabric everywhere and computing of different strengths. And it’s true that the majority of data is going to be created and used outside of the data center. But there is also a data gravity that pulls some data into the data center to be held there forever. But the answer to the question of quantum computing, I would like to propose is to say, where do we see quantum effects in nature? And the answer is everywhere.
So really the question is, what can we harness? What can we manipulate in a controllable programmable way? Or even if it is not in a programmable way, in a way that has a major application that justifies a larger value. So I think that’s going to happen down the line as we understand how to manipulate it and build it in the data center in a more controlled environment and get those applications going. So I think it might very well follow what we’ve seen in computing, where it was just in the data center before it gradually left the premises, right? To the point where on-prem and off-prem became a thing. But I do think that that’s like a ways out.
I kind of generally agree with what was said at Q2B as well, that the roadmaps started looking really interesting by several vendors and IBM gets credit for publishing theirs and compelling others to do so. Like between 2030 and 2035 was when the roadmaps become really, really interesting.
Yuval: Do you get more questions from your clients about the PQC than about quantum computing? Do they worry more about the security or are they focused more on the computing part?
Shahin: The HPC world is really focused on the computing part. I think businesses and investors are attracted to quantum-safe cryptography and post-quantum cryptography because they believe there is a market segment there and there is. But I call that sort of the un-quantum market because besides quantum random number generators and key distribution, and maybe there are a couple of algorithms that actually could use a quantum computer for encryption, the rest of the market is afraid of quantum computers. You’re trying to avoid it rather than use it. So I think it’s a good motivator, but I don’t think that’s ultimately where the action is going to be.
Yuval: You mentioned Q2B a couple of times. In my presentation at least, I was trying to make the case that once upon a time, people didn’t know if quantum computers could work at all. If you can create a qubit or two qubits and entangle them and do something with them, and then it became, okay, we can do that. Could we actually detect and correct errors? And there’s been tremendous, tremendous progress in that from many companies and academic institutions like Harvard and MIT over the past year. So I think that’s, it’s not done, but I think it’s a given that errors could be detected and corrected. And then the next challenge that I was trying to present is about scaling. Well, can we make it large enough to be useful? And my question to you is one, do you agree? And two, what other problems do you see as those that remain to be solved?
Shahin: Right, so in general, I propose that we classify problems into linear and nonlinear. The linear problems are easy, solved already, and we’re done, thank you. The nonlinear parts become harder because they are high dimensional, they could be spatial, temporal, nonlinear, multi-scale physics. What are the patterns in data? It could get really complicated, but you could crudely, because it’s convenient, distinguish nonlinear-easy and nonlinear-hard. I think nonlinear-hard is where all the action is going to be and where all the interesting stuff is going to be. Everything in life is going to end up being a 3D partial differential equation. How are we going to deal with that? Everything in life is going to have some kind of quantum effects manifesting themselves as you get into more precision and more accuracy and higher optimization. So that is a wealth of problems. I mean, as humanity, there is so little we know about so much. It’s just a fertile ground. There’s just, we don’t know, I mean, just ask any question and like two levels deeper and we don’t know. So the answer to those has to be science, it has to be HPC, it has to be quantum computing, it has to be GPU computing, and we’re not going to have a shortage of that.
Yuval: Finally, now that we’ve established that you can drink, we need to get you some food. So if you can have dinner with one of the quantum greats dead or alive, who would that person be?
Shahin: Oh, you hosted a panel at Q2B at the end, which I thought was excellent. And as usual, you did a fantastic job as did your panelists. But for me to just be in that room at that time was quite a reward. So yeah, if you guys go for a drink, I’d love to join. And that was, you know, Preskill and Aronson.
Yuval: Yes, and as I wrote, I felt very small sitting next to these…
Shahin: No, no, no, no, you’re very nice to say that, but the whole exchange was wonderful.
Yuval: Well, Shahin, thank you so much for joining me today.
Shahin: My pleasure, thank you for having me, and thank you all for listening.