Chad Rigetti returns with Singletree, a quantum-accelerated AI server company targeting data center deployment

We're always just having a normal conversation and John will say, "Uh,

yeah, this would be better if we were in the dino experience."

Is that true? The uh the I'm not that much of the dino experience. Anyway, let's bring in Chad Regetti from Regetti Computing and Sagal. Chad, how are you doing?

I'm I'm doing great. How are you guys doing?

We're doing fantastic. Thank you so much for taking the time to come chat with us. Um, I would love to start a little bit with your your background and your journey. Of course, we're going to talk about the company today, but if you could give us a little bit of an overview of your journey in Silicon Valley. I think uh that might be informative. There's a lot to talk about there, and of course, it relates to what you're doing today.

You bet. Yeah, great to be here guys. Uh I I started uh I I got interested in quantum computing when I was a senior in college and did a PhD in this field. Uh and spent about three years at IBM research in the early days. Uh you know helping build up the quantum computing team there and then started my own company that was Regetti Computing in 2014.

Uh I was introduced to Sam Alman and uh you know he said we had coffee and he said hey well have you you should do YC and I said what's YC? And uh and so he explained to me what Y Combinator was and that was the first batch after Sam had taken over YC in 2014. And he brought in a bunch of uh hard tech companies into Y Combinator for the first time.

And so I got to be a part of this incredible group of companies in including Helon, uh

Ollo, which is now public, GO Bowworks,

GKO Bowworks. Yeah.

Uh yeah, Boom was a couple batches after me, but there was this cohort summer. But yeah, so any uh it was a fantastic experience. Ended up running Regetti for about 10 years. We took it public in in early 2022 uh through a spa transaction. They were the third quantum company I think to go public. And so that was an incredible journey. Uh and you know, so I've been in quantum computing, I usually say my entire adult life and in Silicon Valley for a big part of that.

But it's just a really fascinating mix and there there's incredible people working in this area. There's incredible technology that's being developed and it's gonna uh it's going to change change the rel relationship between artificial intelligence and computing infrastructure and that's we're working on at Sigle Tree.

Yeah. uh the journey of going public, all the market girrations. Is being a public company less predictable than venture and being private because there's still the whims of the private market whether you're in the hot category that year and venture investors are scrambling to get uh you know their position built up in a particular category. But the public markets seem like even harder to read on because you have retail investors and uh the stocks up and down and and things can repric on a minute-to-minute basis. What was it like psychologically transitioning from private company to public company?

I think either can work and there's a right answer for different companies and you got to ask yourself the question what you're trying to achieve.

Is it liquidity for your early investors? Is it a primarily a capital raising activity?

Sure. uh is it to provide you know have have liquidity for your early employees for example with some companies where you've got a 10-year exercise window for your options and uh you know zooming out in in the regetti kind of uh taking public journey that was a point in Silicon Valley when quantum computing was growing in in commercial maturation and the technology was maturing but a lot of the capital in the markets at that point had migrated for deep tech companies particularly just wasn't available in the private markets so when you look at 2020 into 2022, most of that capital was actually sitting, you know, a lot of it was sitting in spack trusts on the public markets and they were and those spaxs were hungry to cut a deal.

And so a lot of companies ended up going public during this wave simply because the founders, the executive teams were making the decision that that gave them the best chance of capitalizing the business going forward.

Uh and I I think there's a right answer for different things. And now in the past past month or so, Quantinum has gone public via IPO, a tremendous company that's made great progress. uh and so the quantum you know the public markets for quantum computing have reached a point of maturity. There's analysts that deeply understand the technology that are writing about and covering different companies. Uh it's a you know it's a very very interesting marketplace and then uh in terms of what it's like and the decisions that different companies have to make. I think the key thing is to take a long-term perspective on what you're trying to accomplish and what kind of business are you trying to build? What kind of cap table do you want to build? and what strategy best suits you know is best going to help you help you achieve that.

Yeah.

What kind of uh feedback did you get in the early days around naming the company after yourself? I've been surprised at more

uh there's so many generic names in the startup world now that's like the blank company of San Francisco or things things like that or you know all the neolabs have like the same sounding names. be like advanced super intelligence and then

there was a big boom in like Lys like friendly bitly musically there were tons of companies that were for a while but

and I only know one other I can only think of one other company Chris Amdon's company has Amazon heavy industries um but I'm sure I'm sure people thought you were a little crazy back then

uh well quantum was a different thing back then look I think there there's two quantum companies that don't have a Q in their name and I I started both of them. Uh one is Regetti and the other is Sigleree which is what you know what I'm focused on.

Yeah.

And uh but I will tell you when you think about you know advice for founders when you think about naming something and advice is worth what you pay for it. Um, but think of a name that can become iconic and if you that that means it's got to sound very fresh and new and different. And if every other corn company has a Q in it, maybe you try avoiding that. Uh, and that's what led me to Sigleree. Sigree, I I I love this name. It's from a Patrick Rothfus novel.

Uh, and he was an American writer. He wrote this incredible novel called Name of the Wind that came out in mid 2000 2010 or so. Um anyway, so Sigree is uh we're building quantum accelerated uh uh quantum accelerated AI servers for the data center to bring quantum technologies uh directly into the data center to act as a co-processor for the GPU or XPU pods that have become the unit of compute and AI infrastructure today. And uh we're based in Ann Arbor in San Francisco. Our hardware development is here in Ann Arbor, Michigan. Uh where it is hot and humid today. and um and our AI research team is right there in downtown San Francisco.

So what actually needs to happen? What is the path to you know I would imagine like cheaper tokens like is that the pitch like one day the tokens will be cheaper and we need to do X Y and Z to get there. Um what's X Y and Z? uh you need to well first of all quantum hardware is going to address a lot of different computational challenges today right so quantum computers were able to solve problems that are uh impossible they're very challenging to solve with any form of classical computing no matter what scale it reaches uh so at Sigler we're focused on applying that capability specifically to some of the computational challenges in AI to reduce the power and reduce the cost associated with training and deploying these models at at very large scale uh what needs to happen to get there well you have to build a quantum computer that meets the specific requirements for AI workloads and uh the strategy that we're taking at SIGLY is we are very focused on deeply understanding what those challenges are what needs to happen inside the data center to reduce the uh to bring these algorithms that can have a different kind of scaling complexity class than classical algorithms for AI training and inference um and then understanding what kind of quantum hardware is needed to run those and what we found is there's a set of requirements that you need to meet that uh probably are never going to be met by single modality hardware. What do I what do I mean by that? In quantum computing and quantum hardware, there's different kinds of cubit technologies that you can use to uh to instantiate the cubits. So there's supering cubits. That's what I did my PhD in and what my first company was based on. That's what IBM is focused on and largely Google has been focused on. But there's also trapped ions. Quantum and INQ are doing trapped ions and a long list of other companies. There's photonics. There's now neutral atoms. There's spin cubits and semiconductors. There's all these different hardware substrates that people are using to pursue and to build quantum computers based on those. And what we're doing at Sigle is stepping up a layer and saying from a computer architecture perspective, uh, you know, modern computers aren't built out of one aren't built out of one physical kind of bit.

There's not just one transistor type that makes up these computers that we're using today or the computers that are used to train large scale models and deploy them. There's a plethora of different physical technologies that are used to build these computer systems. And so at Sigleree, we're looking across all the different quantum modalities and hardware types and architecting computer systems to meet the requirements of AI based on the maturing uh path that all these different hardware modalities are on. And that allows us to build systems that are specifically tailored to AI and that we believe are going to be able to meet the meet the requirements of bringing quantum into the AI data center at scale.

How important is simulation at this point? Are you at a place where you can uh run this like like basically run the the the code of the future uh in simulation to understand like run it on a classical computer not see the performance gains but at least understand that uh when the computer when the quantum system is available uh there will be a cost savings. Yeah, we we've been able to do that largely speaking and you can do simulations of of something computer system or a jet or anything and varying levels of physical fidelity and and detail.

Uh the simulation we've been able to do so far indicate that we expect a level of you know several orders of magnitude potential speed up for key training tasks. Right? So this is not a factor of two or a factor of five increase that we're targeting with quantum acceleration inside the data center. it's several orders of magnitude, you know, when when all the pieces come together. Um, but that simulation you talked about is a really really important and powerful part of designing a computer system. You can't simulate all the all the logic of a quantum computer because that would require a quantum computer itself kind of by definition. But you can do load profiling, you can do uh you can do traces, you can understand how that's going to, you know, uh uh uh be b distributed across classical and quantum hardware and also simulate all the networking transactions in between. And so that's a kind of simulation driven design approach we're taking.

Yeah. Um I I guess uh what specifically in training benefits from quantum computing because the the example that everyone goes to in terms of uh quantum computing uh you know novel uh no novel algorithms that actually have potential to do something that a classical computer can't do. It's like shores algorithm cryptography usually. Um, but when people think about training AI, they usually just think a bunch of matrix multiplication. Uh, is there some different path that you plan on taking or do you think you can uh operate at sort of a hardware agnostic layer? uh much like we're seeing you know leading AI firms get off of CUDA like is there a world where you get off of classical and but but by and large it's the same training paradigm.

It's really interesting. I think the answer is both. So the our our starting point is we're looking at ways that you can insert quantum algorithms and quantum computing capability into the existing paradigm the ex the existing workflow for training and deploying very large models frontier models at scale. And that means that you're looking for an insertion point from quantum algorithm where the data in the data out allow you to then take a step that would take maybe you know uh uh a day or two classically and compress that down to hours or minutes and do that throughout the workflow.

Um the challenge is that quantum computing provides an exponential uh you know the possibility for exponential speed up with the right algorithm but it also has this issue with data in and data out. So it's classical data in which is can't be exponential in size and classical data out. And so the less you do that that translation between the quantum part and the classical part it's going to end up working better. So asmtoically where we're where we're heading is more quantum native models models that are designed in the first place to leverage a quantum computing capability tightly integrated with your classical infrastructure. But where you're not where you're probably not going to see is fully quantum, you know, quantum based models that don't include a substantial amount of classical compute as well.

Yeah.

So this isn't going to replace all the, you know, the AMD or Nvidia infrastructure in the data center. It's going to augment it. And our business model and our our focus and our product strategy is to take the to build a quantum accelerated AI server that sits next to the pod and acts as an accelerator for the XPU or the GPU pod in the data center and drive towards a very high attach rate of ideally one to one in the data center infrastructure of the future. Um and that's what's going to allow you to then run you know accelerate the current paradigm but also use that as a substrate to design new kinds of models that will fundamentally be better and more efficient. more efficient from a time perspective, from a cost perspective, from an energy perspective, but also uh these models are just a in a way just a representation of the computer hardware that they're based on.

And what's easy and hard from a computing and communication perspective on the hardware translates into the model capability and with quantum you have a fundamentally new resource in the data center that's going to allow new model capabilities to be developed and brought and brought to market. How are you thinking about, you know, timelines with this new with with this with the new company? Is it do do you think there's uh I imagine with the business right now is like an entirely more of like tech technical risk than execution risk. Is that is that the right way to think about it? Like there's a lot of hardcore research that needs to be done understanding, you know, the feasibility of of of the approach. Um and uh and and what kind kind of like conversations are you having with you know potential partners uh uh if at all right now versus you know about about kind of like the near-term application or are you you know are conversations like 2030s and beyond kind of thing.

Yeah, we're targeting we're talking to customers now. We've got several active you know conversations. I think partnerships and early engagement with customers is a big part of our strategy. The reason that's important is because the the challenges of really bringing a new compute uh you know capability into the AI data center are substantial and you got to be working with customers out of the gate to really understand those requirements, what moves the needle for them as an organization. And so that's what we're that's what we're doing and that's what we're focused on. Um in terms of timing, it's a fantastic time to start a company like this. the underlying hardware has made such tremendous progress in the past 1015 years and the market is you know with the amount of investment that's being made in AI infrastructure there is clearly a recognition that we need a new approach to drive down the cost per token to drive down the energy associated with these very large scale data center projects to make it fundamentally more efficient and quantum promises a you know a more efficient way of translating watts into intelligence. That's what this enables and unlocks in the long term. And to me, this is in in many ways a better idea than putting stuff stuff in space

because uh ultimately yeah, space gives you a lower, you know, cheaper access to energy and it gives you a better way to to uh dissipate that heat,

but you got to put it into space and that takes a lot of fossil fuels, that takes a ton of energy in the first place and it doesn't actually change the computational complexity of the computer hardware that you're running. Why don't the power the power challenge quantum can unlock much more than that?

Yeah, it's a it's it's a good point. Why why don't you think Elon has has made a a real run at quantum? I think the answer is that quantum is at the at this interface of deep science and engineering and a lot of what needs to happen over the next three to five years to bring this technology to market at scale is engineering risk but it is quantum engineering risk and it's not vanilla you know it's not not that it's easy not that any of the purely classical stuff is easy

vanilla rocket science

it's not vanilla rocket science and it's not vanilla fab at scale right and so even if you look at the leaders in quantum computing hardware It's not necessarily the Intelss of the world.

Incredible company that has, you know, propelled humanity forward for half a century, but they're not the leaders in quantum because quantum is a new form of engineering and I wouldn't characterize it as science risk. I think for quantum a lot of it that that is behind us even though there's tre there's tremendous work to be done but there is a lot of quantum engineering risk and that's an area where I think you need to see uh you know companies that are quantum specific bring the technology forward and at that point I think that all the big AI labs are going to need to lean in with quantum

when when do you think there will be a flip around sentiment uh from uh around around quantum it feels right now like at least in our corner of the internet there's so FUD around quantum and uh obviously

based on financials like Right.

Yeah. Yeah. Yeah. So that's that's what I want to know though like is is there a m like you know rewind 10 years if somebody said AI there was a very very small percentage of people that were like incredibly excited about it

and you know deeply involved and could see the trend line and could see that we would get to this point. I mean, Sam was talking about like people becoming best friends with a chatbot. I think in like 2015 or something like or3 was like losing money. It wasn't like making revenue yet.

Yeah. And that was even before that.

Yeah. No, I know. Well, well before that. And so, but then eventually it flipped and and it's really hard to, you know, there there's there's a lot of people that are AI bears and they and they talk about like overinvestment, but they can't deny the value of the products, right? like they're fundamentally pretty useful, right? Um, and you could argue that they're, you know,

well, some bears can, but yes.

Yes.

Some some bears would still figure out a way to argue that it's that they're not useful, but but I imagine like with both of your companies, you're predicting that like, you know, within the next five year, there's there's like a a flip. But what do you think is the first kind of like driver of that where um maybe the average uh the average person in Silicon Valley actually starts to say like, "Hey, I wasn't taking quantum seriously enough."

Mhm.

It there's a few things that need to happen. I think the the FUD is real because the companies that are succeeding and doing well in this space, you you can't tell by looking at their financials. You can't put on your kind of growth investor uh hat and say, "Yeah, this is going to be a tremendous company." and look at the metrics. It it doesn't work like that. You got to be able to analyze and and look at these companies and value them

based on their ability to buy down technical risk over over time and the progress that they've made towards that.

So it just creates a lot of uncertainty because it's a challenging task and it's subject to a lot of you know dis discussion and debate. Um but nonetheless I think there are clear there is clearly tremendous you know momentum and progress in the space. Now what's going to change it? I don't know. My bet is when we have quantum computers in the data center running production workloads and that you don't have to say hey that's a quantum computer for someone to care.

You care because it's a more efficient way of generating the you know the answers you need or or or you know training the model or deploying the model for inference. And uh that's when quantum is really going to become a mainstream category is when you don't have to talk about the fact that it's quantum anymore. And I think in a in a large part uh this is what we're trying to achieve with Sigle, right? The goal is that uh to take quantum computing and to obfuscate it underneath the hood of a classical computing system or underneath all the rest of the infrastructure that's already there and to not ask the end user to be programming it and writing code for it. That's all going to be done with AI anyway. And so uh that is just a better it's a better way to train your model and you know you need this thing or else you're it's going to take you too long and your customers aren't going to be happy with the quality of you know the outputs they're getting. That to me is a big inflection point and I think that can happen in the next 5 to seven years. I think that can uh but there's this whole march that needs to happen to take the technology from one proof point to then you know all the cost engineering that needs to happen the reliability engineering and that's going to be the really fun journey for quantum computing over the next decade is to get to that point where we're selling you know hundreds or thousands of units a year and uh and but that's the journey we're on and that's the march that quantum technology has been on for a good you know one two decades now

and then uh this is probably very obvious to somebody that is focused on quantum but But uh not not to me just cuz I I I don't I don't follow it closely. But like why why a new company? It feels like quantum like as you've explained it feels very obvious to apply it to uh data center buildout and and you you said it could be like a a meaningful inflection point for the technology overall. Uh why why why was a new company necessary and and you know why did you take this approach? Well, at a high level, I think all the different quantum hardware modalities have made tremendous progress and the right way to build quantum computers for AI is multimodality. That is a fundamentally new approach and uh it ultimately is going to in my opinion be very obvious in retrospect. It's going to work better. Uh but it is a it's such a fresh idea. It's got to be baked into your strategy, the DNA of your company. And then uh all the different quantum hardware companies that were out there before Sigler basically started with a thesis which was we've got the best cubit and so we're going to scale this cubit type up and see how far we can get by scaling it up. And that's why you have so much doctrine and like kind of organizational belief around a particular cubit choice. But in reality, you know, customers are buying a computer. They're not buying a you know the the the physical device or your your cubit technology. And so uh at Segree what we're doing is working backwards from the market application from the AI workload as the as the use case and using that to drive the specification of a system that can then be built from folding in whatever technologies are needed to meet those requirements. It's just such a totally different approach to quantum hardware. It's got to be a new company and uh that's that's that's signal technologies. That's the approach that we're taking. I think that that is uh ultimately what's going to unlock this new you know this market application of AI. The other reason is you said it's obvious, but it's actually not obvious at all to most people in quantum that quantum is going to be useful for AI. And in fact, it's not even a consensus view right now.

And the reason for that is because quantum algorithms themselves are still in this very this phase of discovery and in development. And obviously AI is going to help with that eventually as well to an extent. Um but quantum you know when you interview a set of leaders from across the quantum hardware industry the the you know the the median answer you're going to get for what the applications of quantum is going to be is you're going to use it for quantum chemistry you're going to use it for optimization problems things like that and applications to frontier AI is a new area that is just being developed now because it requires a development and extension of what current algorithms can do and then new algorithms altogether specifically for that that's what we're tackling at Sigle is that kind of quantum AI native research lab, right? Or a frontier a AI lab that's quantum native and uh and then we're doing that alongside developing our own quantum hardware.

Mhm.

Uh before before we jump, I didn't get we you you mentioned kind of the the history behind the name, but what is the significance of Sigleree in the novel that you mentioned?

Oh, well, you guys got to read the novel novel for one. It's absolutely incredible. And uh but uh the other thing is single tree is basically a discipline in the book that is learned at university and uh you know and it basically amounts to you inscribe runes on a particular object and by doing that you can imbue that object with properties that it wouldn't otherwise have or you can govern like heat and light flow and things like that. It's also a discipline where it's got a quant quantitative angle to it and if you do it wrong you can blow things up. So, it's got this this mix of kind of coding and hardware, but then a mysterious kind of angle of controlling things from a distance by how you do this uh these inscriptions. It's really it's a really amazing concept,

a little bit of magic.

Amazing. Thank you so much for taking the time.

Thanks, guys. Have a great rest of your day.

Cheers.

Let me tell you about Console. Console builds AI agents that automate 70% of IT, HR, and finance support, giving employees instant resolution for access requests and password resets. Our next guest is already here.

Very cool name. I wish you know what I'm