The Six Five On the Road at AWS re:Invent 2022. Patrick Moorhead and Daniel Newman sit down with Richard Moulds, GM of Amazon Bracket, AWS. Their discussion covers:
- What is Amazon Bracket?
- How Amazon Bracket is driving education and adoption within Quantum computing
- The role of classical computing in accelerating the development of Quantum
- The future of Amazon Bracket within the Quantum community
Be sure to subscribe to The Six Five Webcast so you never miss an episode.
You can watch the full video here:
You can listen to the conversation here:
Disclaimer: The Six Five On the Road is for information and entertainment purposes only. Over the course of this webcast, we may talk about companies that are publicly traded, and we may even reference that fact and their equity share price, but please do not take anything that we say as a recommendation about what you should do with your investment dollars. We are not investment advisors, and we do not ask that you treat us as such.
Patrick Moorhead: Hi. This is Pat Moorhead and we are live at AWS re:Invent 2022 in Las Vegas. The Six Five is back and we are rocking it with awesome tech conversations and great people. Daniel, how are you?
Daniel Newman: It is good to be here. It’s good to be back. This week has been compelling, let’s just say, from day one. It doesn’t really matter if we’re talking about practical implementations of the cloud all the way to ambitions and sustainability and who knows, the future of data and even quantum.
Patrick Moorhead: No, I know, and that’s great to lead in. Let’s introduce our guest. Richard, how are you?
Richard Moulds: Well, it’s Wednesday, so I still got a voice, but it’s an amazing show. Two and a half days in, wow. A lot of people, a lot of conversations.
Patrick Moorhead: No, it’s great. In some ways, industry analysts are professional event attenders, right?
Richard Moulds: Right.
Patrick Moorhead: Literally, we’re like a pack of clowns going from show to show. One of the ways that I gauge a show is the amount of people per square foot who’s walking around in the hallways. If I go to the Venetian, a little bit of it here at the Wynn, a little bit calmer, but this show is absolutely rocking and totally understandable. AWS is at the epicenter of so many things that essentially invented IAS 15 years ago. One of the things in our research and we talk to your customers that they do so well is you have so many variations of compute and the ways of doing things. Also, some of the more future forward ones are looking at what’s next, and quantum computing is absolutely what’s next. I don’t think there’s anybody debating, is it going to be big? The only thing we’re debating is when.
Richard Moulds: Right.
Patrick Moorhead: Maybe a good place to start to talk is what do you do for AWS?
Richard Moulds: Sometimes I think I have the best job in AWS. I’m the general manager of Amazon Braket. Amazon Braket is the quantum computing service of AWS. We launched a service three years ago at re:Invent and we’ve just gone through our second anniversary as a GA service. So mainstream service, making it easy for customers to knock down the barriers to adopting and learning about quantum computing. So yeah, we figure out the roadmap. We manage the decisions around which quantum hardware we think is interesting to customers.
Patrick Moorhead: Right.
Richard Moulds: I’m sure we’ll get on to talk about it, try to span a massive range of diversity in terms of experience and focus. Some people are just looking for a quick win in this industry. Some people are thinking 20 years out as to where this technology’s going to expand. This is a builder’s conference. It’s not really a marketing schmaltzy conference.
Patrick Moorhead: No.
Richard Moulds: It’s all about building. What’s great about re:Invent is it’s training people. It’s getting people up the curve, getting people to adopt technology, serves a super, super curious crowd, this show, which is obviously perfect for quantum.
Daniel Newman: We were talking to one of his peers, Raj on the semiconductor side and I think we had a comment that 75% of the attendees of AWS re:Invent are geeks.
Richard Moulds: Absolutely.
Daniel Newman: The other 25% are wannabe geeks. So it really is for builders. It is [inaudible 00:03:32].
Richard Moulds: Absolutely. And that’s why I think it’s so valuable because it feels like a very practical event.
Daniel Newman: But there might be nothing geekier than quantum.
Patrick Moorhead: It really hits new levels.
Daniel Newman: It does. And we’ve been following it. We’ve been close. You can follow our work. We’ve done a lot of episodes, done a lot of research, published a lot of opinion on it, talked to a lot. By the way, it is a mostly PhD field, mostly. It is mostly a group of very academic. Most of the conversation’s been very academic, but one of the things that I’ve been really longing for is some people that can help democratize it. Let’s start making it make sense. Let’s start talking to the audience and just figure out how to take the cloud at scale and build really impressive stuff and do the same with quantum. So that, to me, is a lot of what Braket was all about, was starting to experiment with all the different companies that are building different quantum computers, using different quantum technologies, super conducting, ion trapping, atom splitting. So talk a little bit about the experimentation, the process that you’re going through, deciding who you’re partnering with and what Braket’s really setting out to do for quantum.
Richard Moulds: Yeah. Economy has been on the radar of a lot of companies obviously for 20 years. It’s a potentially disruptive technology as you said in your intro. So there’s a lot of companies that are nervous about being disrupted or see an opportunity to be the disruptor. There’s a lot of interest naturally that generates. We’ve been debating for 10 years whether or not and when AWS should be in the quantum industry.
Until a few years ago, we had concluded that it really wasn’t worth the time of most of our customers to focus on this technology. There just wasn’t quite enough center ground. The technology just wasn’t mature enough. As I’m sure you’ve heard many times, we’re customers-first company. We don’t do stuff just for the sake of technology. We do it because customers find it useful. So for a long time, it just wasn’t quite there.
Three or four years ago, we said, “You know what? It feels like it’s starting to come together.” As you said, different modalities, different ways to build a quantum computer, interestingly different in terms of how they fit use cases, clearer sense of what those use cases are. We felt it was worth the time and trouble for some of our customers, not all, to start investing in resources to really grapple this technology and to plan for the future because they all have the obvious question. When’s it going to happen? What sort of people do I need to hire? Which bits of my business are going to get impacted first? Is this a threat or is this an opportunity? So we launched Braket and I suppose the overriding goals are just to make it simpler. We think a lot at AWS about what we call the innovation flywheel, just generally try and knock down the barriers, bring the right people together and try and just accelerate the process.
As you said, it’s a super fragmented industry. The guys that build the hardware aren’t particularly plugged into the applications and vice versa. The educators don’t have easy access to the machines, unless you know a physicist or are prepared to spend potentially millions of dollars a year to get access to one of these devices. If you’re a corporation, you couldn’t get your hands on these systems. All different commercial model, different developer toolkits, different user experiences, super fragmented. So in a sense, it sounds a little bit cheesy, but we think of Braket almost like the town square for quantum computing. We want to bring the hardware guys, the software guys, the consultants, the educators, the researchers, corporate users together in one place. If you can take away some of the barriers, then perhaps you can accelerate innovation and just get there quicker and that’s what we’re all about.
Patrick Moorhead: Yes. So not only are there different modalities in the core technology, which by the way-
Richard Moulds: Which is a scary word for some people.
Patrick Moorhead: No, I know, but what I do love is I think we all realize in the industry that not one single technology is going to rule them all. In fact, I like to make parallelisms to things that have happened already and let’s take GPU compute. Not all solutions are best for different model sizes. One is gigantic models. We have training, we have inference, and we have everything in between. There’s also different familiarity with the technology that on the AI and ML side today, whether you want to go to the metal and you serve up hardware or you want a completely abstracted SaaS app with Textract or something like that, you have it. It looks like you’re trying to build with Braket opportunities for different levels of experience for people to be able to experiment. Am I reading that correctly?
Richard Moulds: Yeah, absolutely. That word experiment, that is the key word. There’s a lot of mythology in this industry as I’m sure you’ve detected in a-
Patrick Moorhead: We’ve hit quantum advantage. We’re done.
Richard Moulds: Right. Yeah.
Patrick Moorhead: Three years ago, supremacy. Sorry. Yeah.
Richard Moulds: Yeah, I know. This is still at the research stage. Companies are not using quantum computers in production systems to save money or do stuff faster. We’re not quite there yet and we might not be there for a while, but that’s not to say there’s not value in getting started. This is radically different stuff. It is applicable in a certain set of use cases, radical different set of skills. This is not something you pick up overnight. Our job, of course, is to break down some of the barriers so you don’t need a physics PhD to actually do this stuff, but still, it’s a curve and you need to get started. But yeah, you’re right, some customers really want to be abstracted away from the hardware. They don’t necessarily care how you build your qubits. Fair enough. But some other folks actually literally want to get in and play with the microwave pulses that literally control the atoms on the chip.
We just have some features in that function in that area, so we could talk about that later, but yeah, it’s a tool stack and it gets thinner the higher you get. A lot of the partners we’re working with are trying to close the gap between the hardware there and the actual use cases. Some people actually have problems that we think quantum can solve.
Daniel Newman: I see parallels with AI and HPC. I know it’s different state of maturity, but there was a period of time. I remember meeting someone that was like, “I got a PhD in AI in the ’70s,” and I’m like, “Whoa.” It was 50, 40 years before that really started that theory. So quantum obviously, we’re in this stage now where people are starting to realize there’s certain genomic challenges that it’s going to be able to solve. So you’re seeing large pharmaceutical and biotechnology companies investing. They know that eventually, they’re going to want to be there. Or financial services firms that want to be able to use it for anti-money laundering and fraud and understand that that heterogeneous relationship between quantum and classical is going to be a game changer.
But what you said is really great, is more experimentation now. The work they’re doing, they could still do it faster on a HPC with what they know today, but they know in the long run, they’re going to be able to do it faster tomorrow. These people you’re working with, these personas, these are almost like the innovation labs inside of big corporation, government, cities. Talk a little bit about that. Who are the people that are locked in the lab right now plugging into Braket, starting to figure out how to make this work in that parallelism of moving away from classical to what I think is mostly going to be a marriage of quantum and classical in the future?
Richard Moulds: Yeah. I think that’s one of the myth, again, one of the preconceptions, it’s not a myth, that somehow, a quantum computer is a supercomputer. It’s just the next biggest machine that somehow replaces existing classical hardware. That’s not the case at all. Quantum computers are really co-processors. They do certain sort of mathematical functions. In the end, we think better than classical machines, but they don’t do everything. You’re not going to be running a spreadsheet or video rendering on a quantum computer. It’s always about a classical world.
Daniel Newman: Where’s the keyboard? You know what I mean?
Richard Moulds: In the same way we think about a math coprocessor or a crypto coprocessor. It’s being developed.
Patrick Moorhead: How about a GPU as an example of that where it’s API-driven? Listen, the original AI algorithms or ML algorithms were developed in the ’60s. The problem is it takes big data to do machine learning and it was too expensive and we didn’t have the accelerators to be able to train it and do inferences. Then the folks at the University of Toronto came up with this ability to find images. By the way, I worked for one, a GPU vendor and I saw banging our heads against the wall trying to use this thing something other than playing games or doing 2D drawing. Is that a good analogy that we should look at where quantum is the accelerator and classic computing is running a set of the application?
Richard Moulds: I think it is. Yeah, I think it is. I think you could imagine an HPC workload today that’s struggling in certain areas. You can imagine leveraging a quantum computer in the same way we leverage other instance types. So use them in parallel, a manager workflow across a variety of resources, some of which are quantum, most of which are classical. So the notion that at some point, there’ll be GPU, CPU, QPU instance types, it does make sense, yeah. Calling it a quantum computer is almost a misnomer. You should think of it much more as a quantum processor for doing a certain set of things. What we are doing now obviously is enabling customers to explore what those things are and to ultimately figure out how do you coordinate a workload that spans these different processor types. How do you actually formulate a problem so you know this bit of the problem can go onto a quantum machine, this bit of the problem should stay classical?
And how do you actually orchestrate the workflow between those things? Because look at AWS, it is effectively an infinite pool of classical resource. Quantum computing is not there. These are individual machines that aren’t even alive for most of the day. You got very different operational characteristics, so organizing what we think of as hybrid algorithms that have quantum and classical components. Actually managing those workflows and orchestrating these things together is not so simple, and that’s what Braket does. We don’t just provide access to quantum computers, but we actually manage that interplay between classical and quantum compute.
Daniel Newman: So-
Patrick Moorhead: Oh, go ahead.
Daniel Newman: No. You got me thinking, so I kind of hit you on the personas.
Richard Moulds: Okay, which I don’t think I ever really answered.
Daniel Newman: I don’t think you ever really got you, but I’m going to join two thoughts together because I think it’s going to be persona plus enterprise. Because these personas are being hired by these companies, universities, these government entities, and they’re asked to explore. Amazon and AWS is that company that says, we want to meet our customer… They’re customer-obsessed, meet the customer where they are. So you’ve gotten into this business because there’s a journey going on.
Richard Moulds: Absolutely.
Daniel Newman: So I’d love for you to align that journey of the enterprise, the persona and how it’s all tying together that you’re helping them get there. Because they got to be coming to you saying, “All right, we know we need to do this. Help.” How do you take them through that journey?
Richard Moulds: There’s a couple of different ways. Some enterprises, I think of them as the quantum pioneers. They’re doing exactly that. They’re recruiting professors from universities and creating teams, significant teams in their organization.
Patrick Moorhead: Typically within the research group of the CIO or the-
Daniel Newman: CTO.
Richard Moulds: Yes.
Daniel Newman: Big banks.
Richard Moulds: Yes. Yes, exactly. They are scouting their organization to find the use cases and try just to get ahead of that curve. Some companies, they hear about quantum, they’re intrigued by quantum, they come in. They want a quick assessment. Is this a quick win for me? I’ve got problems to solve. I’m busy.
Daniel Newman: That’s easy. The answer is no.
Richard Moulds: Which is most companies. They have day jobs. Oftentime, the conversation, we’ll pivot to, “Yeah, quantum’s interesting. You should keep an eye on it and this is how you should do so, but in the meantime, we can make your classical stuff just go better.” Oftentimes, surprise, surprise. Not everybody’s maxing out what they could achieve classically. So a lot of that conversation turns back to classical technology.
Daniel Newman: It’s an $80 billion business for you guys now.
Richard Moulds: And growing quickly. Some people, hyperfocused. They are leading us in many ways. Part of that obsession is to listen to what they tell us in terms of evolving the service. We do have a professional services team. We launched it at the same time we launched the Braket service. We call it the Quantum Solutions Lab. It’s within the AWS Pro Serve organization, but it’s a specialist in quantum computing. It’s entirely staffed by quantum PhDs and really, that’s geared to corporations that really want to dig in, collaborate around algorithm development. In addition to that, we have a lot of partners or software partners. It’s interesting. In this industry, everyone focuses on the hardware because it is super cool. The way these work is amazing.
Patrick Moorhead: Right?
Richard Moulds: Awesome pictures. They definitely have the best pictures, but people don’t often think about the solution guys, the quantum software providers. Different folks, different skill set. Don’t really know anything about how to build a machine, but boy can they close the gap between the machine and the use case. There are dozens of software partners that have come out of the woodwork since we launched Braket. Our proposition to them is A, it’s an operational platform that can actually build a service on themselves, but also gives them the choice. It gives different instance types and they tend to be very industry-focused. They’ll be focused on chemistry problems or they’re focusing on application problems or fraud detections, the sort of things you were talking about.
They oftentimes are the first interface to corporations because they’re intrinsically solution-oriented and they can, in some cases, bridge the gap between quantum future and classical now and they can start taking those customers through that journey. In a way, Braket, we’re still at the infrastructure there. That will evolve over time, but now we are really keen on building that community of software experts to find those chinks of light in terms of where this technology really fits into different industries.
Patrick Moorhead: By the way, that’s how GPU compute started.
Richard Moulds: Exactly.
Patrick Moorhead: And now, look where it is today. It’s evolved and it’s delivering incredible value. I wanted to shift the conversation to academic research and open source. As far as I can remember, some of the biggest breakthroughs have been driven by research. I always like to clarify, don’t confuse R&D. I know we say R&D, but research is very different from development. Research has higher risk, has a higher failure rate, but boy, when it hits, it hits. If you look at whether it’s wireless technologies, compute, storage, anything that was huge out space came from first, research, and then we answered a bunch of questions and then we got development. A lot of that research is done through academic institutions for quantum. In fact, every company seems to be aligned with an academic.
Richard Moulds: They’re oftentimes direct spinout.
Patrick Moorhead: Exactly, and there’s funding. Then you have open source, which at least in the past 20 years, has been this growth multiplier where the promise is you make some improvements and get share of everybody else. You can take advantage of all the improvements that they put in and AWS has basically built on open source software. I’m getting to a question here. How does Braket help these two audiences, academic research and also the folks dealing in open source?
Richard Moulds: I mean the research piece is essential and it’s good that you drew the line between R&D.
Patrick Moorhead: Drives me a little crazy, I’ll admit.
Richard Moulds: Yeah. Sure, there’s a lot of engineering work going into building quantum computers these days, but we should never forget that to get to the point of a fault-tolerant quantum computer, there are significant scientific advances still to be made. There’s Nobel prizes to be had in this area. This is not done and dusted. This is not just making it faster and cheaper. One of the primary use cases for quantum computing right now is figuring out how to build better quantum computers fundamentally. A lot of researchers out there are diving into, really, I think two areas. We have the dream of delivering a fault-tolerant quantum computer, which we know can deliver exponential compute power. But we don’t have one of those machines yet and we might not for a while. So in the meantime, we are living with these error prone noisy machines. People have coined this phrase NISQ, noisy intermediate-scale quantum, which is a way of saying we’re going to use classical computers to make quantum computers work better in the near term. In the long term, we think quantum computers make classical computers work better.
Patrick Moorhead: Entanglement.
Richard Moulds: Right, but for the next, who knows, five years, maybe longer, classical computers make quantum computers work better, which is what NISQ is all about. We have researchers that are absolutely focused on the long-term game, error correction. Quantum error correction is the elephant in the room. We have to fix that problem. There’s no chunking out a million qubits of today’s quality.
Patrick Moorhead: Sure.
Richard Moulds: You just simply couldn’t use them. So we got to fix the world of error correction and there’s a ton of research going into that. Part of what Braket does, some of the functionality we talked about the stack. The reason why you want to dive down the stack is so you can research error correction, so you really can understand how these qubits actually interact at the physical level. The other area of research is the near-term stuff, is NISQ. It’s, how can we think about almost the notion of a self-training quantum algorithm? We used to trade the concept of training, obviously machine learning, but you can take similar approaches, iterative, convergent processes to enable quantum algorithms to be tweaked over time to better deal with the error profile that they see on machines that they’re actually working with.
There’s a whole discipline of figuring out what these hybrid or variation or algorithms actually look like. A lot of researchers are in that space because that’s where the near-term win is. That’s potentially the route to quantum advantage and is an endless debate in the industry as to whether that’s ever going to be fruitful. But you have to try because if it’s not going to work, you just have to wait until a fault- tolerant quantum computer comes along, which might be a while. So in the meantime, we have to see what value we can squeeze out of current devices, and that’s all about, again, [inaudible 00:23:14] together quantum and classical resources. As quantum computers get bigger, that amount of classical resource to support the quantum computer gets bigger quickly because there is this exponential gap. If there wasn’t, we wouldn’t be bothering building these machines.
People shouldn’t think of a quantum computer in isolation. You don’t go behind the curtains and tinker with the device. It’s all about really how you bring these resources together and today, that’s the algorithmic level. How do you make this? How do you get around these noisy devices? But over time, that hybridization exists at the HPC level, where you have a general problem that you are then distributing across a wide variety of devices.
Daniel Newman: Yeah, I love what you say about the quantum computing as a whole, not thinking so much about computing processing. We’ve been in rooms with ion trapping machines. We’ve been in rooms of superconductor. It’s not flipping open a laptop.
Patrick Moorhead: Is this on?
Daniel Newman: Yeah. I say that because I think there’s still a lot of that misconception. One of the biggest opportunities still is to just break that mythology between the average person and what quantum can be, and of course, all this stuff happening that you mentioned in research and then eventually it comes. Like you said, the early classical computing machines were very different than what we see today in that same type of evolution. It’ll probably happen faster because of how fast technology is moving. But more than anything, Richard, I want to thank you for joining us. This topic is not over. We have many more conversations and I’m hoping they’ll be with you that we can bring here to talk about the progress. Because this is one of those things that every year, there’s progress being made, whether it’s more fidelity, more quantum supremacy. But I personally like more usability and realistic examples of software, of cryptography, things that are going to be done with quantum.
Richard Moulds: That’s why it’s in the cloud. That’s fundamentally why we felt it’s really, really important to deliver Braket as a cloud service within Braket because it’s all about leveraging the quantum computer in the context of the platform, in the context of all of that compute resources. Our large corporate customers, they live and breathe the AWS platform. They like the tool that exists. When we launched Braket, it was really important that we adhered to all of the security mandates, operational mandates, the integration with the various tools and services that are across the entire ecosystem of AWS, and customers love that. They want to be able to control access to these resources just like they want to control access to databases in EC2.
Daniel Newman: Well, Richard, thank you so much for joining us. We got to end there.
Richard Moulds: You’re welcome.
Daniel Newman: Really appreciate having you at AWS re:Invent, The Six Five on the road at this event. Great guest. Great conversation.
Richard Moulds: That was fun. It’s great.
Daniel Newman: There’s so many more of these to be had. If you’re out there and you’re watching, we have several more of these conversations. Go ahead and click that subscribe button. Join us for all the episodes here and all our other episodes. We love having you as part of the community, but for this one, time to say goodbye.
Patrick Moorhead: Thanks, everybody.
Daniel Newman: See you later.