Transcript | Investing in Quantum with SOSV

Quantum computing has been surrounded by hype from the beginning. While these machines may exceed our most ambitious imaginings in the future, investing in them today brings some challenges to VCs. Find out how they vet these emerging technologies and how quantum winter comes into play in this episode of The Post-Quantum World. I’m your host, Konstantinos Karagiannis. I lead Quantum Computing Services at Protiviti, where we’re helping companies prepare for the benefits and threats of this exploding field. I hope you’ll join each episode as we explore the technology and business impacts of this post-quantum era.

Our guest today is a quantum software investor and general partner at SOSV, Bill Yao.

Welcome to the show.

Thanks, Konstantinos.

Good. It’s nice to see you again. We got to meet before in a chat, and I have to warn my listeners, Bill also does stand-up comedy on the side, so we’ll see what transpires.

Hard to be funny on a podcast, though, really.

This is a different kind of episode, so I just want to ask how you found your way to quantum investing.

Back in 2012, my daughter came down with type 1 diabetes, and I started to look at whether I could make insulin at home, because we live on a five-acre organic farm in the middle of nowhere, and we’re self-sufficient on everything — power and food and all that stuff. And I noticed back then that biotech was getting cheaper and cheaper, because every time I looked at whether I could make insulin at home, there was a new advancement that made it super cheap. And so I launched our biotech practice in 2014, and we’re now the highest-volume biotech investors in the world.

And along that journey, I discovered that from a biotech perspective, we have run out of compute. I know people with labs that literally produce more data than they can crunch, and because of that, there’s a huge bottleneck in processing everything from protein analysis through to sequencing. And it is amusing that a lot of biologists get into biology because they hate computers, and now they’re really dependent on them and they’re running out of compute. Biology has progressed far faster than Moore’s law. Our understanding, and our ability to collect data in biology, has outstripped Moore’s law. And so I was also looking at some of these amazing deep-tech startups, going, “How are they going to get their compute?” And that’s how I came across quantum. I was, like, “Quantum computers are a thing.”

So it’s a very personal journey. Then, the biology data thing — just look at genes. There are four units there instead of just, like, zero and one, like we had in computing before quantum computing. You could see how there’s just so much potential for more data, even in that one example.

That’s probably not a great example, because those four units, two of them only combine together. DNA is actually binary. But look at a protein — proteins, they’re complex. You can create a strip of DNA that will code a protein, but the protein only works when it’s folded in exactly the right way. It’s like extreme origami, and understanding how those myriad billions of combinations can suddenly fold into a shape that exactly works — that’s really hard.

And then, of course, gene expression, that takes that simple environment and explodes it exponentially in different ways. But now, on the topic of quantum and investing, I want to start with a controversial question right away — something that I’m sure a lot of people are curious about in this current climate: Do you think we’re in a quantum winter?

Undoubtedly — the hype has passed. People are, like, “I want to get my quantum startup funded.” There is a dearth of capital out there. We are in a double winter, because the VC market overall has taken a punch in the face. A couple of weeks ago, on an idle Thursday, we started to hear some shaking and trembling in the capital markets about Silicon Valley Bank. By Friday, we looked at our portfolio and realised that there was an enormous amount of money that had been raised that was sitting in Silicon Valley Bank accounts. And if those bank accounts weren’t fully guaranteed — one startup had $50 million. They just raised their Series A, 50 million bucks in the bank, patting themselves on the back: “Woohoo — we’ve done the raise! And on the Sunday morning after, they were looking at being able to recover $250,000.

That’s a pretty small percentage you’re able to be guaranteed there.

It’s not even payroll. And so you’ve got a lot of people nervous anyway. And then VCs have very distinct time horizons, and so you’re looking at, for quantum compute, you’re looking at a long time horizon, and then you’re looking at already-dominant players. There are quite a few platforms out there that are very well established, and they have decent runways and they have decent roadmaps, and it’s hard to think of a new way of innovating that hasn’t already been figured out by one or other group that’s already well-funded. So what are VCs chasing in terms of deals? So you’ve got a double winter.

And do you think AI is going to make this worse in some ways? I was thinking about this recently: It’s so hot right now to be thinking about how you can make money off things like ChatGPT, whatever. Do you think that will draw some investors away from looking at quantum while they’re trying to make up for things like losses this year?

It definitely has. It’s also weird, because ChatGPT in and of itself — how do I put this? If, in 1991, I had said to you, “Would you write me a document on this following topic?” and you said, “Yeah, sure.” and then I said, “By the way, can you use Word to write that?” you’d look back at me and go, like, “What are you, weird? Of course I’m going to use Word. I’m not using WordPerfect, because everybody uses Word. Why would you not use Word? Why would I not use Word? What are you thinking?” Writing a document without GPT-4 now is a bit like that. It’s a bit like saying, “You should write that with GPT-4,” and anybody with a brain is going to do that, because it’s going to provide somewhat better results somewhat faster.

Word did not spawn a whole lot of venture business models. People are saying, “GPT-4 is going to spawn a whole lot of new startups.” I’m not so sure about that. GPT-4 is an incredibly well-trained model and has a nice user interface, but that doesn’t mean that it’s going to spawn a thousand businesses.

There are privacy concerns. When you’re using GPT-4, your data is going out to the internet, to OpenAI. That’s why there are all these open source ones now that are appearing, too, that let you train your own data internally, query your own CSV files, whatever. That’s an interesting space.

I’m not so worried about that. If you buy the API from them, which is what most people who are using it are doing, you do get your own compute space there. So, there are some T’s and C’s in there that give you at least some level of ownership on what you’re generating if you’re doing it by the API. But even people using the API, I don’t think that there are a million new business models. Just because something is popular doesn’t necessarily make it have the kind of utility that you can build a new business around. Does that make sense?

Yeah. This could be a temporary distraction, too, in terms of investing. What do you think gets us out of a quantum winter?

In part, AI is part of the answer, because AI itself has run out of compute. Just because GPT-4 is useful doesn’t mean it’s as scalable as we need it to be, because frankly, we need more compute. And quantum is one of the few ways of doing that. We are headed toward, in traditional compute, the barrier to Moore’s law of being able to dissipate energy through a system fast enough that you create a hard stop to progress unless you can figure out a different way of doing the compute. And while quantum isn’t perfect for that, it certainly can improve it enormously.

We’re also looking at ways to train models with things like quantum-inspired approaches, tensor networks and things. It might help that industry in parallel, and maybe that’ll regenerate some new interest, too.

It’s more than “might help.” It is going to be essential to cover the compute need that AI has. Nvidia makes a separate processor suite for AI. The number of transistors on those boards makes the top-end GPU in the world look like a pocket calculator.

They’re building solutions for hybrid quantum simulation, things like that, too, to actually run these types of problems on classical hardware. And before we delve deeper into investing stuff, I’ll give a little disclaimer here: I have to specify that while Bill is an expert in the general area of investing, none of the information provided in this episode is intended as investment advice, as an offer solicitation of an offer to buy or sell, or as an endorsement of any company securities fund or other securities or nonsecurities offering. There we go. I’m a very professional podcaster. Now we can move back. Now, you say whatever you want, and it’s all on you.

What advice would you give to someone who’s interested in investing in quantum computing stocks but doesn’t have experience in the field? Common mistakes to avoid. Granted, it is a winter, but it might be a good time to invest.

The best time to invest is always when the market’s down. I am utterly incompetent when it comes to investing in the stock market. I’ve cofounded and exited two very nice large public companies, and so I got to know intimately how to communicate to the market that we were any good. But when it comes to actually picking stocks, no idea. When it comes to VC investing, I’m much better at that, because VC investing is about building conviction, and I’m reasonably good at looking at where you can match your investment hypothesis to the kind of companies that look promising and you can build conviction around.

In terms of investing in quantum startups, while we, through our HAX programme, are one of the most prolific hardware investors in the world, I have yet to see a quantum hardware play that I would be going after, because there’s a lot of capital already chasing that, and the return profile is pretty slow. The sweet spot, the investment hypothesis that we’ve developed, is that quantum hybrid/classical compute that can be quantum-enabled is the place to look. And there are always a lot of good software startups around to invest in.

The ones that have already started to investigate quantum now are the ones that are going to be winners, and here’s why: You’ve got a five- to 10-year time horizon to build a software company of real worth. Yes, we see overnight successes, but every overnight success I’ve ever seen has 10 years of history behind it, including OpenAI. If you get in now, by the time those businesses have built a successful business around classical compute, if they are thinking about and are incorporating either quantum hybrid techniques or just preparing for an overall move to quantum, quantum will be a really big change in their cost base.

You might have a company that in five years’ time is looking to go profitable, and then suddenly, it creeps along for a while, and then suddenly, quantum compute catches up and they can change their margins enormously. And then you’ll have a spectacular return, because a lot of investment that I’ve seen in the quantum space, people are thinking, “We’ll build it, and then we’ll wait for quantum computing to come. And then, somewhere along the line, we’ll be expert at what we’re doing.”

And then, by that time, they’re going to have a long road ahead of them just to make the business successful, because a business is all about the customer, and if you can’t deliver to the customers already, that takes time to build. Even if you’ve got the best quantum solution in the world and you come out of the blue, it takes quite a long time to get adopted. My investment hypothesis is, go for companies that are getting toward adoption or are adopted on a classical basis but that have deep understanding of what quantum algorithms they’re going to need to change their cost base.

So you’re looking at their timeline forward and how they’re going to make that shift. Does the age of the company, when you catch sight of them, matter? Is that a big part of your equation? How long do they have to be in business for you to be interested in them?

Our overall investment hypothesis is precede. We like companies that are very early, because we run programmes that de-risk early-stage businesses. One of the things I spend a lot of my time doing is skilling up scientists and engineers in how to be hustlers, because I’ve discovered over time that the scientific covenant is convened around doubt, and the entrepreneurial covenant is convened around conviction. And so you have to build a bridge between the two that a scientist-entrepreneur can walk backward and forward in because without doubt, there’s no scientific method, and therefore it’s not science. But without conviction, it’s not a business. We’ve had enormous success skilling up scientists and deep-tech engineers to become entrepreneurs. I have not ever had success trying to get a salesperson to be a scientist.

What level of rigor you put toward the scientific idea itself, the kernel that the company has? Usually, for what we hear on this show, too, when I have guests on, is, it’s usually like, someone did their Ph.D. in something, and maybe they met someone else doing a similar Ph.D. in something, and then you’ve got a company. How do you evaluate that initial kernel of an idea?

One of the things that stood me in good stead over the years is, does it break the laws of physics? If yes, no investment.

Is it a perpetual-motion quantum computer?

Exactly. And I have had people come to me with exactly what you just described. Thank you very much. Bye. Thermodynamics is a thing even in the quantum realm.

Definitely in the quantum realm.

There is definitely a need to filter on, does it work? The next thing to filter on is, really, are you taking science risk, or are you taking engineering risk? If you’re taking science risk, you should be going after grants and not VC, because VC is locking you into a time horizon that if you have science risk, you’re probably not going to be able to deliver on, and that’s not going to be a happy outcome for you or the investors.

That makes a lot of sense. That is the realm of the grant: We’re trying to prove this thing — we’re trying to advance this thing — and you’re saying, “No, when you come to me, it better be that we’re going to have this in the market by this date.”

Exactly. Show me your working prototype. Oh, good. Now, show me how we can scale it. That’s true for software, hardware, biotech — it’s true for all of them. If you’re about to have massive science risk because you both have a good idea with your Ph.D.’s, no, thank you. Bye-bye. Now, here’s another area where I think AI is going to make a huge difference: Write your grant with GPT-4.

For the boilerplate stuff.

Even beyond. There is just so much utility. It takes so much of the pain out of grant writing to use the AI and work with it.

Not to get too far off-topic, but because you like the technology, how do you feel about machine language detection? OpenAI even has one, and there are a bunch of other ones, and they give you varying scores — 90% sure that it was written by AI, etc. How do you feel that that colors someone’s perception of a document once it goes out into the world?

If you can’t figure out how to confound that, you’re not looking very hard.

I’ve seen all sorts of crazy ways to do it. There are sites that do it. Some of them make the language really weird. I’ve seen prompts that are designed to change different types of approaches of writing, paragraph by paragraph. It just makes me wonder, though, about things like watermarking. There have been discussions about watermarking documents so that when they go out there, you can still detect that they were written.

Let’s look at chess. When Deep Blue beat Gary Kasparov, a lot of people said chess was over: The machines can always beat us. There’s no hope. And then there’s this whole new raft of centaurs out there, which are basically an AI and a human working together, or AI and a human team working together. And chess has become contested at a whole new level, and chess is now more popular than it ever was. Now, if you are playing a computer and you know you’re playing a computer, that’s fine. If you’re playing a computer-assisted human and you know you’re playing a computer-assisted human, that’s fine. But if you’re playing a human and they are using the computer exclusively, then that’s somehow cheating, and those people get kicked off Chess.com. The sentiment out there is all about transparency, really. It’s, like, can I do a better job using the AI? Yes, I can. Then I can be completely transparent about that. Why wouldn’t I be? Why wouldn’t I give the best answer possible in the least time using the tools available?

Yeah. Where do you draw the line? To go back to the example of Word, Word always had a grammar check. More recently, it became an editor, this feature that’s even smarter, like Grammarly. And Grammarly is another example. You’ve had these assistants for several years now that can make your writing better. But now, right inside of Word, Microsoft’s plan is to have, click, “Write me a letter to my boss asking for this,” and then it just appears in Word. At that point, it’s like it’s a tool inside Word. Is that OK to use? Do you have to be transparent? It’s such a gray line, a blurry line.

It is a blurry line now. It’s going to quickly be swept away. Very quickly, people are going to go, “It’s stupid how we’ve been doing this previously.” Occasionally, you write a handwritten letter, and it’s somehow more special. But you don’t do that every day. It’s the same thing. Sometimes, you’ll turn your AI off, assistant off, and handwrite something — great. But the reality of most modern communications for business is that we’re after efficiency anywhere we can get it. It’s going to be obvious that you haven’t proofread your AI-written document. So, it’s not like you can just write it and send it and forget it. You have to put some work in — it’s just less work.

And is it AI on both ends? This reminds me of a meme I saw recently: I have these three bullet point ideas for a business — feed it to GPT-4 and say, “Turn this into a proposal.” The proposal goes to someone, and that someone gives it to GPT-4 and says, “Reduce this to three bullet points for me.” And it’s like the same thing on both ends. The boilerplate in the middle becomes an intermediary or something.

I saw that meme, and I can see the appeal of it from a human perspective. The question is, is the information transferred better from your brain to the other person’s brain? As a communicator, your primary job is to take responsibility for how what you wrote lands. And if you’re not taking responsibility for that, it may end up going through the GPT analysis, but if we end up with that situation where it’s bullet point to bullet point, that’s hugely inefficient, and that will show up. Wherever there are huge bottlenecks in transferring information, it generally shows up. That particular scenario will become obvious, because people will start screwing up.

It’s like the game of telephone. I don’t know if those three bullets will transfer perfectly when you have all that in the middle.

Exactly. And then those mistakes will become obvious.

They’ll be glaring. So what do you think are the most promising quantum computing companies out there right now that you’ve seen recently? Whether because they gave you a great proposal using GPT-4, or for other reasons, what do you think of some of the most promising ones?

That’s too tough a question, because so far, I’ve seen a lot of companies, and so far, we haven’t made our first investment in the space. It’s difficult for me to get behind something when I haven’t put my money there.

It says something. You gave these criteria before, and you’re saying that the companies that are coming are not meeting them.

Exactly. I’ve seen a lot of people who are super early and they are taking science risk, and then I’ve seen a lot of people that are very much quantum purists, and they’re, like, “We will only ever do it with quantum. It’s no use doing it with classical.” And I’m, like, “So, you’re not going to do anything to build the market for what you’re doing. You’re not going to engage with any customers. You’re just going to work on the quantum algorithms that you need. Bye.”

Because of those types of risk and where they are, do you think government funding and policy will affect any of that? Will it change some of the science-risk side — getting them with grants or something more ready?

There are a lot of grants available that are helpful. It’s going to be interesting to see what different governments do there. I know that at least one company I spoke to has since moved to China and has been given a lot of money, and that’s very interesting for a lot of reasons, geopolitical among them. I know that there are European grants in the space, I know that there are U.S. grants in the space, I know there are U.K. grants in the space. And grant money is genuinely more able to take science risk, because otherwise, why would you do it?

And as you start finally giving the green lights to some of these up-and-comers, do you think that diversification is going to be important? Would you feel that it’s better to have different types of companies in quantum — not just 17 that have a tool for writing algorithms, but maybe across the space? Would that be more interesting to your company?

As an investor, we like to invest in verticals, but our verticals are pretty broad. We’re the number one by deal flow investors in biotech, and we’re the number one by deal flow investors in hardware with a specialty in robotics and medical devices, and in the biotech cyber specialty in alt proteins. And we also have some therapeutic tracks, etc. So, I’m looking for companies that fit into my ecosystem of other investors, because I want to be able to look at them and say, “Talk to my other investors in the series A and series B and say, ‘This is great, and it fits your hypothesis.’”

As an early-stage investor, half my job is getting the next round put together. Otherwise, I’m not de-risking. So, it makes sense for me to fit into the investment hypothesis of the future. Diversification is the focus for someone like us. And I don’t think it’s worth overdiversifying. It’s worth, sure, having some diversity within the vertical. Absolutely. We try not to invest in two identical companies doing exactly the same thing. At the same time, it is much more interesting to get them into our ecosystem and then be able to talk to our ecosystem intelligently about them for the next round and the next round.

If you can name one additional thing that a company would want to do to make sure their quantum company got funded, what would it be? Obviously, so far, you said, have a working prototype, things like that, on some level. What could be one other bit of advice that you would throw out there that no one’s really doing that’s messing them up?

Have three customers — seems obvious.

You’re seeing that as a big failing right there.

 I’ve got lots of people coming to me saying, “We’ve got this piece of IP” or “We’ve got that piece of IP locked down,” as if that’s the thing that’s important. But you have all the IP locked down in the world. If you have zero utility to anyone —

Would you recommend, then, aggressive strategies, like even severely reducing the price of the product for those initial three customers just to get that proof of concept and viability?

The best thing in the world is three letters. It’s a TLA — three-letter acronym: NRE. Nonrefundable engineering cost. If you can get a customer to pay you for some of the engineering and then become a customer that way, that’s excellent, because that means that they’ve bought what you’re selling. And they may not be buying in volume yet, but they’ve certainly paid out in order for you to tune what you’re doing to their needs. That’s a good buying signal. And very early-stage companies can get significant NRE, and you don’t have to give up any IP to do that.

That’s great. And before we close down, I wanted to know, what are your thoughts in general on the future of quantum computing over the next two or three years?

I’m super excited. I see we are getting closer and closer to being to having real utility. We’re going to see utility way before we have the qubits that we think we need, because I’ve seen some very creative approaches to hybrid work, and that’s going to show some promise real soon — and I’m looking for, where is the rubber hitting the road in an early-stage startup? I’m looking for that to put some cash behind. We’re still a ways away from ubiquitous massive compute, but I do see that there are software investments possible around even things like metrology, where the quantum piece is very small, but it has an impact overall. And the last thing is, I do notice that we are slowly increasing the number of quantum algorithms that are available, because the number is very small.

It really is the quantum zoo. It’s tiny. It’s like a petting zoo.

Yeah, it’s a very small petting zoo. When I started looking at this, it was single digits, and now it’s like — I wouldn’t even hazard a guess, but it’s not big. Whereas you compare the number of algorithms we have in traditional compute, and that’s a lot more. And that’s a very interesting property for utility.

And some of the original ones weren’t even practical. They were just proof of concept to prove that you can do anything. The whole point of the Deutsch-Jozsa algorithm — it’s just to show that quantum computers can run.

Exactly — “Oh, wow. I’ll cut a check immediately.”

This is some real practical advice, and our listeners could see that you can’t just say, “Quantum — give me money.” You have to show some utility.

Exactly. Saying, “Quantum — give me money” is a bit akin at the moment to saying, “Fusion — give me money.”

That was one of the first requirements — it was perpetual motion and tabletop fusion. You have to have a working prototype to get a patent. Those are the two — for good reason.

Thank you so much for joining.

Thank you very much for having me, Konstantinos. I look forward to hearing from your listeners.

Yeah — they might challenge you.

I’m up for it. I really am.

Now, it’s time for Coherence, the quantum executive summary, where I take a moment to highlight some of the business impacts we discussed today, in case things got too nerdy at times. Let’s recap.

Bill feels we’re in a double quantum winter, but he is bullish on the technology. He acknowledges that quantum will have fantastic use cases and boost available compute resources for other areas such as AI. Bill points out that the best time to invest is when the market is down. But his firm also has set criteria it looks for when investing in quantum and other emerging technologies. This investment hypothesis is focused on quantum hybrid offerings from startups.

SOSV knows there’s a five- to 10-year time horizon for products in the space. They look to precede companies that have solid science at their core. They’re not interested in companies taking science risk, and feel those should be looking for grants — companies taking engineering risk instead are more attractive. They have a working prototype and have to scale it, for instance.

Bill also points out that they’re interested in companies that have customers. These customers can even be helping cover some of the early engineering costs. This kind of trust can help make a case for utility and viability.

That does it for this episode. Thanks to Bill Liao for joining to discuss quantum investing, and thank you for listening. If you enjoyed the show, please subscribe to Protiviti’s The Post-Quantum World, and leave a review to help others find us. Be sure to follow me on all socials @KonstantHacker. You’ll find links there to what we’re doing in Quantum Computing Services at Protiviti. You can also DM me questions or suggestions for what you’d like to hear on the show. For more information on our quantum services, check out Protiviti.com, or follow ProtivitiTech on Twitter and LinkedIn. Until next time, be kind, and stay quantum-curious.