In this episode of Innovators Inside Podcast, Ian Bergman and Emmanuel Vallod, Partner and Head of Venture Research at Hivemind Capital, explore what happens before a scientific breakthrough becomes a startup.
Much of venture capital is designed to evaluate companies. But some of the most promising work in AI infrastructure, deep tech, robotics, computing, and other frontier technologies begins long before there is a company, a product, or even a traditional investment opportunity.
That gap is the focus of Vallod’s work with Dark Matter Lab, an initiative designed to support researchers as they move promising technology closer to commercialization.
The conversation ranges from space-based data centers and next-generation computing to research grants, technical risk, robotics, and the role private capital could play in accelerating breakthrough innovation.
Here are the five key takeaways from their conversation:
Traditional venture capital usually enters when there is already something recognizable to invest in. There may be a founding team, an early product, customer demand, or at least a clear path toward building a business.
Frontier technology does not always develop that way.
Some potentially valuable technologies begin as research inside universities and laboratories. The intellectual property may exist. The technical potential may be significant. But there is no company yet.
Vallod argues that this creates an important funding gap.
Research that could potentially reach meaningful technical milestones in one or two years can take much longer when researchers lack access to sufficient capital, compute, technical support, or commercialization expertise.
Money alone does not solve every deep tech problem.
Emmanuel Vallod explains that investors working at the earliest stages of frontier technology need to understand the underlying technical work well enough to evaluate the risks involved.
That can mean understanding the code, mathematics, system architecture, hardware constraints, or scientific reasoning behind a research project.
This is especially important when there is no conventional MVP or established market category to evaluate.
In traditional software investing, an investor may be able to focus heavily on distribution, customer acquisition, market size, and execution. In frontier technology, the core competitive advantage may still be the technology itself.
The lesson for investors is simple: capital is useful, but technical understanding becomes increasingly important when investing before the product layer exists.
Research is rarely linear.
A team may begin with one hypothesis, discover something unexpected, change direction, and eventually reach a more valuable result.
Emmanuel Vallod argues that traditional grant structures can struggle with this reality.
Grant funding can take a long time to secure, money may be distributed over several years, and future funding can depend on completing predetermined milestones.
Those milestones can create structure, but they can also push researchers toward meeting administrative requirements rather than following the most promising technical direction.
Ian describes the extreme version of this as a “grant zombie” situation, where teams begin optimizing for the next tranche of funding instead of the breakthrough itself.
This does not mean grants are unnecessary. Vallod makes clear that they remain important. The issue is that grants alone may not provide the speed, flexibility, or resources required to move certain frontier technologies toward commercialization.
How do you evaluate something that has never existed before?
For Emmanuel Vallod, traditional frameworks are not always enough.
If a technology sits outside the historical range of products, companies, or systems investors have previously seen, comparing it to existing categories may produce the wrong answer.
Instead, investors may need to return to first principles.
Is the underlying technical problem important?
Does the researcher demonstrate strong critical thinking?
Has enough technical risk been removed to describe what a future product could become?
Could the technology meaningfully change user behavior or how an industry operates?
The product does not necessarily need to be fully built. But the research needs to advance far enough that a credible commercial direction can be articulated.
That transition is important because it marks the point where a research project may become understandable to more traditional venture investors.
One of the biggest ideas in the episode is that pre-seed investing itself could change.
If Dark Matter Lab and similar models work, Vallod believes larger pre-seed investments could become normal for highly technical projects.
That would represent a shift in where private capital enters the innovation process.
Instead of waiting for researchers to independently turn their work into startups, investors could begin supporting promising technologies during the technical de-risking process.
The potential impact goes beyond funding individual companies.
Earlier capital could accelerate the commercialization of AI infrastructure, new computing architectures, robotics, energy technologies, industrial systems, and other research-intensive innovations.
Vallod believes private investment in this area could eventually grow from hundreds of millions of dollars to billions.
The broader question is whether venture capital can build a repeatable model for funding the space between academic research and startup formation.
The episode also provides a glimpse into the technologies that could emerge from this new model.
Vallod explores the engineering challenges behind putting data centers in space, including cooling, communication, launch costs, maintenance, robotics, and new computing architectures.
He also discusses research exploring a very different future for robotics.
Humanoid robots may make sense in industrial environments built around human workers. Inside homes, however, intelligent machines may take much more familiar forms.
Instead of robots that look like people, researchers are exploring ways to embed intelligence into ordinary objects such as tables, screens, kitchen objects, and other parts of the physical environment.
It is a useful reminder that breakthrough innovation often looks very different before the market decides what the final product should be.
The central idea running through the conversation is not simply how to fund more startups.
It is how to make sure important technology does not stall before a startup can exist.
For founders, investors, researchers, and innovation leaders, that means paying closer attention to the stage between discovery and commercialization.
The next major AI infrastructure company, robotics platform, computing breakthrough, or deep tech business may already exist in some form today.
It just might still be sitting inside a research lab.
Have a question for a future guest? Email us at innovators@alchemistaccelerator.com to get in touch!
Timestamps
🧭 1:36 Meet Emmanuel Vallod
🚀 2:20 Why Space Is Hard for AI Data Centers
🛰️ 4:15 What Space-Based Compute Could Actually Look Like
⚙️ 10:13 The Technology Needed to Make It Work
💡 15:16 When Breakthrough Research Has No Company Yet
🧪 17:12 Why Valuable Research Gets Stuck
💰 20:09 The Idea Behind Dark Matter Lab
🔬 26:07 How to Evaluate Research Before There Is a Product
🏛️ 29:39 Why Grants Are Not Enough
🧠 32:17 Why Frontier Tech Needs Technical Investors
📈 36:14 Can Private Capital Fill the Research Funding Gap?
🤖 39:44 Why Humanoid Robots May Be Wrong for the Home
🚀 45:07 What Changes If Dark Matter Lab Works?
🔎 45:51 Where to Follow Emmanuel and Dark Matter Lab
00;01;36;27 - 00;01;38;29
Ian Bergman
Emmanuel. Good afternoon.
00;01;39;01 - 00;01;39;25
Emmanuel Vallod
Good afternoon.
00;01;39;26 - 00;01;45;02
Ian Bergman
It's awesome to have you on Innovators Inside. Thanks for joining us.
00;01;45;03 - 00;01;47;00
Emmanuel Vallod
Well, thank you for having me. Excited to be here.
00;01;47;01 - 00;02;15;19
Ian Bergman
I hope you're excited because I'm going to drag you straight up into outer space. And here's why. There's a little bit of a paradox in my mind right now, because, as we all know, like space is one of the coldest places that we can think of, right? It's a vacuum. There's no heat, whatever. But your team just sent over a paper showing your research that saying it's one of the worst places in the world to cool a computer.
00;02;15;21 - 00;02;20;16
Ian Bergman
Do me a favor. Explain that in 60s and why it matters.
00;02;20;18 - 00;02;59;15
Emmanuel Vallod
You cannot transport cold. You can only transport heat and to transport heat. You need fairly sizable equipment. When you start doing that in space, and not only you need sizable equipment, which is expensive and heavy, but you are very constrained on where to put that equipment. So while it's not necessarily impossible and crazily stupid to go and put at Space Data Center well into orbit, there is quite a few major technical hurdle along them.
00;02;59;16 - 00;03;31;10
Ian Bergman
And these technical hurdles. I imagine, need some fundamental research to get burned down, and some of that is what we're here to talk about today. So there's a fundamental question that informs this episode, and that is what happens before a breakthrough becomes a company. So for the audience, today's guest, Emmanuel Velo, is partner and head of venture research at Hivemind Capital, a global investment group working at the intersection of traditional markets and on chain and AI economy.
00;03;31;12 - 00;04;01;22
Ian Bergman
Emmanuel, we're going to want to talk about what you're launching, Dark Matter Lab, which I understand to be a first of its kind program supporting AI blockchain researchers before company formation with partners like UC Berkeley, Google Cloud, Gunderson Dettmer, Goodwin, Proctor and you provide, as I understand it, at least $1 million in resources to these companies. And along with your own wealth of perspective on research funding, commercialization.
00;04;01;24 - 00;04;14;07
Ian Bergman
So we're here to talk about why the next generation of breakthrough companies may need to be built before traditional VC is ready to invest. And yeah, welcome aboard. Are you ready to go?
00;04;14;11 - 00;04;15;13
Emmanuel Vallod
Oh, ready to go.
00;04;15;15 - 00;04;40;27
Ian Bergman
Walk me through. Why? This is something that you're paying attention to. Because the news today is filled with data centers in space. SpaceX recently IPO. There's extra launch capacity, and everybody envisions a box, a container getting thrown up on a rocket, chucked out the side of the rocket, and now doing compute. And now we have AI for everyone.
00;04;40;28 - 00;04;45;04
Ian Bergman
So what is compute in space actually?
00;04;45;05 - 00;04;52;22
Emmanuel Vallod
I mean, you're taking a bit of a step back on that. I started pondering on that.
00;04;52;24 - 00;05;21;01
Emmanuel Vallod
Really like a year, year and a half ago, and it eventually made its way into that, that that paper that came together already, like during the month of June. I was trying to tackle it purely from a technical and engineering standpoint rather than a business standpoint, because it's very easy to forget technical obstacles when we start talking billions, hundreds of billions, trillions of dollars.
00;05;21;03 - 00;05;47;09
Emmanuel Vallod
And it was actually super interesting to dig in because it brought back memories from my first grad school and then things where actually some of the technical work my own startup was doing, but basically where I landed was space. Data centers are a possibility, with a long road of technical challenges to be overcome. But if you were to succeed, here's what it would look like.
00;05;47;16 - 00;06;17;07
Emmanuel Vallod
You would basically have a constellation of data center satellites around Earth. They would most likely have to be positioned at the border between the half of Earth that's in daylight, and the half of Earth that's in basically the darkness. Right? Because then one side gets the energy from the sun with solar panels all the time. So you have continuous service.
00;06;17;11 - 00;06;40;14
Emmanuel Vallod
The other side evacuates the heat produced by the GPUs into the interstellar void. Right. And so it needs to be always in the dark. So it's as efficient as can be. And then within each of these satellites you basically have your usual rack of GPUs. You could pack them as much as you need inside the the satellite. Things don't change.
00;06;40;19 - 00;07;01;18
Emmanuel Vallod
Now of course across satellite things would have to change because you need to rethink the way they communicate with one another. Right. And then the way they communicate, of course, from low orbit to Earth, which comes with different constraints. And the other part was also very interesting. That was purely in a way an engineering problem was, well, great.
00;07;01;19 - 00;07;43;08
Emmanuel Vallod
Like let's say you have all these satellites at some point you need to do maintenance, right? So who is going to show up in space to replace a transformer, to replace a transistor, to replace a GPU, to replace a, you know, mother, a motherboard that opened an entire can of worms on robotics and industrial robotics. And of course, you have a tie back to space beyond Elon Musk talking about data centers in space, which is these need to get shipped and to get shape, they're going to require very big, very efficient spaceships, which is, of course, what he's working on.
00;07;43;08 - 00;07;52;22
Emmanuel Vallod
But that's at the end of the day, what it looks like versus that effective vision of a massive warehouse that happens to float in space. It's definitely not going to be that.
00;07;52;23 - 00;08;12;14
Ian Bergman
Yeah, it really is a very different picture. And that's actually why I wanted to start here, because I love the visual. This is not a warehouse or a container floating in space. These are solar panels the size of a basketball court, you know, aggregated, as you said, right around the horizon line. Is that the Terminator line? The.
00;08;12;16 - 00;08;13;27
Emmanuel Vallod
Yeah, yeah.
00;08;13;28 - 00;08;38;19
Ian Bergman
In orbit. So in a very interesting kind of orbital pattern with presumably, you know, fleets of as yet to be invented and proven, you know, maintenance robots. And we talk you talk about space, but it is interesting because your paper identifies the cost of launched orbit that you actually need to achieve in order to make some of this work.
00;08;38;20 - 00;08;47;18
Ian Bergman
And I think I saw 3 to 13 x cheaper than what we see today. And interestingly enough, Starship is targeting pretty much that number, right?
00;08;47;19 - 00;08;49;10
Emmanuel Vallod
Yeah. The top end of that number.
00;08;49;10 - 00;08;55;13
Ian Bergman
Is this pretty much the unspoken thing behind SpaceX's entire IPO these days.
00;08;55;15 - 00;09;11;21
Emmanuel Vallod
I mean, I think that, look, they are smart. They are brilliant product engineering people. I'm sure they've done the math the same way I did the math, which is why we end up basically at about the same numbers, the same intervals. Right?
00;09;11;24 - 00;09;38;04
Emmanuel Vallod
Pretty exciting to work on. Definitely going to be a lot of work. I don't think that this is work that's purely the matter of, okay, give me money and I'm just going to go buy the parts. I'm going to put the machine together. I think there is research work, theoretical work, foundational system architecture work that also must happen right before we actually build the stuff.
00;09;38;05 - 00;10;13;27
Ian Bergman
Let's use that as a bridge, because I think you've talked about a number of these, you know, not just theoretical technologies, things that are in the pipeline of development, but evolution and transistors, software evolution, supercomputing, etc.. But like, you know, how well understood is the research path to from where we are today to a world where it's viable to be putting trillions of dollars of hardware into orbit in a way that actually makes market sense.
00;10;13;29 - 00;10;37;01
Emmanuel Vallod
There are parts that I think we understand very well from a research standpoint. There are parts I don't think we know what we don't need. Right. So what we would what we understand well, in my view is, okay, could we make the data center.
00;10;37;03 - 00;11;13;13
Emmanuel Vallod
Much more condensed. Right. So that whatever for the size of a truck, instead of throwing out whatever number of teraflops or petaflops, you could do 5.6. That's right. For the same size there we have a very good understanding at the research level how to do that, right. In particular, there is wonderful research out of MIT Semiconductor Lab on replacing atom gating transistors with proton gated 20 stores.
00;11;13;15 - 00;11;46;22
Emmanuel Vallod
And they are the point where they have prototypes for cards. It's not like it's a paper with some formulas. They have actual cards on which they've run a GPT two GPT three model. They've compared it to same size form factor and GPUs, and they show that they are able to effectively consume 80% less energy and produce six x more floating point operations per second.
00;11;46;25 - 00;12;11;13
Emmanuel Vallod
So there I would say research wise, we know what's a realistic pastel world, right? There is a second example I like to use, which leverages some of the work I did in grad school and actually leverage this work my own startup was looking at, which is called generalized heterogeneous computing. The idea that you could mix and match different chips from different vendors.
00;12;11;16 - 00;12;39;16
Emmanuel Vallod
So some AMD with some Nvidia, or even within the Nvidia you could make some B 200, B 100, 100 running jointly. Right. And the advantage of doing that, of course, is it makes getting to a point where you can run in production way faster because you can just progressively send stuff, progressively build stuff, and they all communicate very well with one another.
00;12;39;23 - 00;13;16;20
Emmanuel Vallod
It makes maintenance much easier because now if one of the racks breaks and you need to replace, you don't need to replace with exactly the same thing as what's in all the other racks, you replace with whatever you have, right? That today, at the research level, we don't know completely how to solve, but we know the path that needs to be taken to solve, and we have enough proof points at the local scale to be confident that we would be able to reach generalized scale.
00;13;16;21 - 00;13;49;14
Ian Bergman
The term generalized here is actually very, very interesting because, you know, as I understand it, your typical data center, the highest efficiency data centers now are wildly predictable. You know exactly the latency. You know exactly the memory bandwidth, you know exactly the interconnects. The software is perhaps, you know, fragile to any changes in those. But when you say generalizable, you're talking about things like model inference or potentially model training happening across a very unpredictable substrate.
00;13;49;14 - 00;13;56;25
Ian Bergman
And so and so what you're saying is that that is a frontier research area in software, computer science and probably physics.
00;13;57;01 - 00;14;18;02
Emmanuel Vallod
And it's mostly in computer science, both at the algorithmic layer, middleware layer and hardware layer, because it's one of these fields of research and engineering where you need to operate in a hardware software co-design. Right. Versus first, I have my hardware separately.
00;14;18;02 - 00;14;23;21
Ian Bergman
So you're talking about moving the abstraction layer up a little bit for all of the substrate the stuff's running on.
00;14;23;21 - 00;14;24;19
Emmanuel Vallod
Correct.
00;14;24;21 - 00;14;48;01
Ian Bergman
You know, I think it's really easy for the audience to grasp that a new type of transistor is in the lab. I think it's really easy for people to grasp that. Okay, maybe we don't actually know how to do, you know, super payload launches at scale efficiently, yet we all see, you know, the launches on YouTube. I think it's more hidden in the world.
00;14;48;04 - 00;14;51;08
Ian Bergman
Some of the software side work that has to be done.
00;14;51;11 - 00;15;02;21
Emmanuel Vallod
Yes. It's much less intuitive, right. Because the intuition dictates that, well, if it's two GPUs, they are both GPU. So therefore you should be able to do the stuff on both at the same time.
00;15;02;27 - 00;15;14;12
Ian Bergman
And that you can easily balance any difference in characteristics like in software. That's intuition, but that's probably coming from someone like me who failed linear algebra and therefore didn't become an engineer.
00;15;14;13 - 00;15;16;07
Emmanuel Vallod
You know, there is a path to recovery.
00;15;16;08 - 00;15;39;05
Ian Bergman
Yeah, this is really interesting. And I wanted to start here because your papers conclusion is that there's trillions of dollars in outcomes all all over the place. But let's talk about in space data center and space compute. That depends on technologies that exist in the lab today, several of which you just articulated. But that means there's nothing for someone like me to invest in yet.
00;15;39;07 - 00;16;08;10
Ian Bergman
There's no company there. There's IP. This is the stage that I understand. You just built Dark Matter Lab to fund. And so I want to dig in on Dark Matter Lab. But I want to start with a really fundamental question. What made somebody who knew knows things about crypto, a crypto CIO, decide that the highest leverage place to invest is on a research lab.
00;16;08;12 - 00;16;36;24
Emmanuel Vallod
So that's where I come from. And I think this is how it came to be. When I was in grad school, I developed things that were basically the early days of AI infra, right, new algorithms, new system architectures interacting with high performance computing chips. I developed cool stuff in the world of financial engineering, like, you know, their record models.
00;16;36;26 - 00;17;12;01
Emmanuel Vallod
Some of that made its way into my startup back in 2018. We were doing I interrupt before that was sexy. I'll skip all the boring to me. Very sexy details on the technical front, but long story short, for that dark matter context, I had plenty to many conversations with VCs that just did not understand. First, that there was actually demand for AI infra, second, that didn't understand that the mode was technological, not distribution.
00;17;12;01 - 00;17;46;10
Emmanuel Vallod
Distribution would be a result of superior technology. It was extremely frustrating and in parallel to that and becoming a VC. But I continue to be a professor. And so I continue to see repeatedly these situations where there is, to me, legitimately, commercially valuable research that's getting stuck, because on the other side of the table, people are unable to realize the importance of it.
00;17;46;13 - 00;18;15;27
Emmanuel Vallod
And so what could take a year or two to achieve significant technology? There is King will take 4 or 5 years because of capital constraints. Not enough cash, not enough compute, not enough support on how do you basically do the tech transfer from the lab out? Open source is one route that's probably the less difficult one, but open source is not the only path forward for frontier tech.
00;18;15;28 - 00;18;20;03
Emmanuel Vallod
Right. And so it's this.
00;18;20;05 - 00;18;49;21
Emmanuel Vallod
Combination of restrictions hindering research progress and my own experience that led to basically saying, hey, how would I go about this if I were to start with a blank slate? Can I actually get my alma mater? And it happens to be a unicorn factory? Can I get them to trust and support and buy in onto this initiative? Right.
00;18;49;24 - 00;19;17;27
Emmanuel Vallod
And and they did. And I think they did not just because, you know sure. We are startup campus. Let's try that. I think first and foremost they did because coming from Rich Lyons and his team, there is that symmetric perception that their PiS, their lab directors, are hindered in bringing things to market because of a gap in resources at that stage inside the labs.
00;19;17;28 - 00;19;48;00
Ian Bergman
Well, and it's really interesting. I mean, I'll be honest, we at Alchemist see this exact same problem set from a different direction. Right? And as an increasing amount of our portfolio starts to look like science backed deep tech, you know, novel research based companies, we actually struggle ourselves deeply with this question of where are we on that technical de-risking and commercial de-risking curve?
00;19;48;07 - 00;20;09;08
Ian Bergman
Right. Because there's areas that organizations like ours are well equipped to help with. And there's ones that were not. And, you know, it's interesting, even as you're saying things like you can end up in 3 to 4 years of technical de-risking when it should take one. I'm thinking, oh, I've seen that. I've seen that in our applicant pool, in our portfolio.
00;20;09;08 - 00;20;26;02
Ian Bergman
But let me ask you, so what's the what unique insight did you have about how to fix this? And what's the non-obvious bet that is backing what you're doing with with Dark Matter Lab?
00;20;26;05 - 00;20;49;21
Emmanuel Vallod
I mean, I think the unique insight is I'm a VC who happens to be a professor, who happens to have been a deep tech founder, and we still is a researcher in that space. And so I end up kind of seeing all the sides in very extensive details that I'm therefore able to connect the dots. Right. I think that just boils down to that.
00;20;49;23 - 00;21;06;21
Emmanuel Vallod
And some of the uncommon things that as a result, we are backing or looking to back.
00;21;06;24 - 00;21;37;29
Emmanuel Vallod
In terms of sector, there are going to be your usual suspects, right? We have things in AI efficiency. We have things in physical AI, we have things in energy grid and industrial applications, and we have things that have to do with modernization of payment architectures. We have things that have to do with world models and mitigation of financial risk for real estate or infrastructure assets.
00;21;37;29 - 00;22;10;16
Emmanuel Vallod
So I would say as very high level topics, they are broadly topics where this is look, but they tackle major technical hurdles. And I'll give you two examples. So when you look to the at real estate, whether it's commercial, residential even like data center warehouses. Right. And you think about how do we price the risk of a natural catastrophe damaging these buildings.
00;22;10;19 - 00;22;43;26
Emmanuel Vallod
The way you do it today is you basically have historical data on whatever white fires, hurricane, earthquakes, right, per geographical region. And you basically say, okay, here's the likelihood that it happens. And if it happens, here is the damage surface. It's a very static way to look at things. It's also a very crude way to look at things, because within one arena, a building built in 1950 and a building built in 2030 of very different beasts.
00;22;43;29 - 00;23;19;16
Emmanuel Vallod
Right. A high rise tower versus a warehouse is a very different beast. Right. So there is work carried at UC Berkeley by actually mitosis advisor on doing that properly and properly means having a dynamic three dimensional representation of each building and their environment understanding, you know, not just the how many buildings you have, but the actual shape and condition of the building.
00;23;19;19 - 00;23;53;27
Emmanuel Vallod
Right? Or the windows, wood or metal. Is the roof significantly aged or new? Is it steep? Whatever steepness does it sit on? Flatland. Is it flatland with rocks or with sand or with dirt? Is it typically tall trees or shrubs or bushes? Is it weed? Is it perennials? She has that dynamic understanding as a result. Right. You have the ability to understand much more accurately why a particular natural catastrophe would happen.
00;23;53;27 - 00;24;10;24
Emmanuel Vallod
And if it were to happen, the extent and nature of damages that would be caused. You're therefore able to price things in a much more responsive, accurate manner. And then, of course, that should satisfy the entire financial services industry.
00;24;10;24 - 00;24;12;18
Ian Bergman
And it's I mean, it's it's interesting.
00;24;12;19 - 00;24;39;22
Ian Bergman
Right, because anybody can understand that being able to do what you just described requires a level of data and a level of compute infrastructure that would have been impossible to imagine until very recently. But I think perhaps the integration of all of this is underappreciated, like how much work it takes to take the compute infrastructure, the data and figure out what to do with it.
00;24;39;22 - 00;24;41;02
Ian Bergman
And so.
00;24;41;04 - 00;24;42;04
Ian Bergman
Using this example.
00;24;42;05 - 00;25;32;01
Ian Bergman
Or others, I have to imagine that you have exit criteria, right? You have a milestone that you expect that type of research to achieve. And I'm curious what that is. Is it you can show me a viable MVP? Is it that you have proven that all of the financial institutions, the insurers, whoever it is, have demand for this solution or will buy, like how far does the research go and when do you say, okay, this project has advanced far enough to maybe go to more traditional channels for the next round of funding.
00;25;32;03 - 00;26;03;10
Emmanuel Vallod
Very fair, very fair. I think the step even before that is worth covering, which is what entry criteria in the first place make me say that putting a resource to start with makes sense, right? Because there you come before there are. But the insurance companies have already expressed demand like you're way before that to me, the entry. But it then drives the exit, which is why I want to start talking about the entry.
00;26;03;13 - 00;26;07;16
Emmanuel Vallod
To me, the entry is.
00;26;07;19 - 00;26;43;27
Emmanuel Vallod
Do you, as a researcher, demonstrate to me your ability to have critical thinking and out of the box thinking on a problem of industrial and societal importance of massive proportion? Right? I don't want knowledge on the problem. I want critical thinking and out of the box thinking that at a high level. In a nutshell, that's the entry point right there.
00;26;43;29 - 00;27;18;04
Emmanuel Vallod
The exit is really a resultant of that, which is okay, we knew where you were. You had a sense of the direction you needed to take, whether that was the right direction or you need a pivoting along the way. Did you arrive at a stage where technology is the rest, and technology? Is the risk enough that a product can be articulated?
00;27;18;04 - 00;27;25;01
Emmanuel Vallod
It doesn't have to be built because product could still be ten different shapes, but you can articulate at least what the product would be.
00;27;25;03 - 00;27;34;08
Ian Bergman
Is TRL level the right framing for this? Or like how how how do you think about making that assessment in a data backed way?
00;27;34;11 - 00;28;17;04
Emmanuel Vallod
In some cases, it is more art than science, right? Because in some cases you are creating something that has no precedent, and so you cannot apply. In my view, you cannot apply an established evaluation or categorization framework to something which is completely out of your historical distribution of observables. Right? So you have to go to first principles. So I'm a very big believer at that stage in the first principle reasoning and first principle reasoning in some cases were mapped back to standardize whatever product conceptual evaluation frameworks.
00;28;17;04 - 00;28;30;15
Emmanuel Vallod
But in some cases, we'll go back to what you and I were discussing in preamble, which is can it out or spontaneously user behavior NY.
00;28;30;17 - 00;28;53;16
Emmanuel Vallod
That to me really the models operandi. If you are at that point where it's like, yes, it can. And here is why. And to a degree that will deem satisfactory from the standpoint of reasonable educated individuals, then I think that's the stage where you should go to the standard VC wrapped to fuel the next phase.
00;28;53;18 - 00;29;18;06
Ian Bergman
Yeah, this makes sense. I mean, I love what you're building, but I want to ask the kind of hard question, or at least maybe the semi-hard question. I think, you know, a devil's advocate is going to look at what you're building, and they're going to say there's already a massive pipeline of available grant funding, right? There's already university based incubation labs and paths for professors and their postdocs to go out and attempt commercialization.
00;29;18;06 - 00;29;39;07
Ian Bergman
There's already tech transfer offices. There's already an infrastructure of resources that theoretically and I'm I'm I'm weighing heavily on the theoretically here are designed to take stuff out of the lab and get it into market. So like why is what why this not grants.
00;29;39;10 - 00;30;14;26
Emmanuel Vallod
No. It's super question. And what I call each of these so grants first in many cases have drastically compressed beyond tenable in the last year and a half. Right. So that on its own adds to the limitations grants had in the past. Grants to me are in. They should come back anyway. Different topic, but grants where they are insufficient is essentially the cadence over which they happen.
00;30;14;28 - 00;30;47;03
Emmanuel Vallod
Right? To get a grant. Typical process is a year and a half to start receiving the grant money, right? Grants are never disbursed fully upfront. They are dispersed over time, and time is not counted in weeks or months. It's counted in years. And how the next disbursement is triggered is based on delivering certain milestones that aren't necessarily what you and I would consider as useful milestones.
00;30;47;04 - 00;30;53;27
Emmanuel Vallod
They may be much more bureaucratic milestones, therefore misaligned with the reality of how research works, which is not linear.
00;30;53;28 - 00;31;06;09
Ian Bergman
Yeah. And we've all seen that simply the requirement for the next tranche of grant money completely redirects focus in a, in a team. Right. You end up in these kind of grant zombie situations. So I do get that now.
00;31;06;10 - 00;31;41;15
Emmanuel Vallod
Incubators, incubators, accelerators play a role that to me is symmetric of dark matter inside universities, where they would cater well to things that are essentially the application layer and they don't cater well to what's the pick and shovels in the back end. Part of that is their roles as incubators is more democratic vis a vis the poor of potential projects than my role.
00;31;41;23 - 00;32;06;18
Emmanuel Vallod
Right. So they have to spread resources. Second, they typically deal with projects that actually are not IP sensitive, so you don't have to interface tech transfer office. And then third, they are typically not themselves researchers, geeks and founders. Coming from that space.
00;32;06;19 - 00;32;17;14
Ian Bergman
I want to double click on that. Can you just kind of expound on that? Because I think that's a very important insight. But it also speaks to how the infrastructure helps people self-select into the right track.
00;32;17;16 - 00;33;04;06
Emmanuel Vallod
To me, if you want to be helpful to someone at the precede stage building backhand infrastructure in Frontier Technologies, you need to have the technical product, operational and commercial expertise to actually have some level of useful insight. Right? And the commercial and operational expertise can be acquired through operator experience, the product and the technical expertise, unless it has been your field of study, unless you've been a researcher and an engineer in that field, you are not going to have that right.
00;33;04;07 - 00;33;35;03
Emmanuel Vallod
And when you're tackling, precede in infra and you're trying to really resolve the technology risk question. Right. There isn't much there going to be able to opine on if you don't even understand the code, the mass, the logic, etc. in the papers, in the reasoning of research teams. So that's to me the distinction there in DNA of the investors.
00;33;35;03 - 00;34;16;09
Emmanuel Vallod
But that does like the incubator. Right? Right. And then the third world of tech transfer offices, tech transfer essentially focuses on the interface between IP created in the labs and the corporate world. So yes, they end up supporting IP commercialization, though I would say they end up typically supporting a subset of IP commercialization, which is IP that's either developed inside the lab but paid by your corporate sponsor, or is IP developed inside the lab.
00;34;16;09 - 00;35;03;23
Emmanuel Vallod
That results in a patent, and then the patent gets licensed to the corporate. Right when you do that. And yes, it commercialize things, but that's typically counter to the startup rent, right. Because of these IP and Cumbrians or these explicit IPO assignment rights back to external sponsor that tend to be large corporations. And I would also say there that it's maybe a bit less true the last two years than before, but still, large corporations aren't exactly where innovation flourishes in the best possible way, because they have their own priorities that aren't always aligned with that.
00;35;03;23 - 00;35;07;22
Emmanuel Vallod
So let's take a concrete example.
00;35;07;24 - 00;35;38;22
Emmanuel Vallod
If you are a meteor researcher, you have your new proton gated transistor. And on the other side of the table is Nvidia, who still operates on atomic gated transistor. You could look at that and you could say, well, that's going to be the future of these transistors. So we should go all in and make sure it's ours. Or you could say, hey, by the time right, this would start making revenue in N of size that we would care about.
00;35;38;25 - 00;36;11;21
Emmanuel Vallod
We know it's probably end of years, that we can just wait and let that thing get to a scale where we would care, and then we can see whether we go do it ourselves and plug the car with cash, right? Or whether we just leave them out. So these are symmetry of incentives for bigger corporations is, I think, another hindrance of correctly monetizing, commercializing the full scope of commercial breakthroughs coming out of the labs.
00;36;11;23 - 00;36;14;04
Emmanuel Vallod
Henceforth. The gap that I hope to plug.
00;36;14;05 - 00;36;55;02
Ian Bergman
I love this. I mean, Emmanuel, before before the call, we were talking a little bit about coming back in a year or two and seeing, you know, if a hypothesis on the episode bore out, you know, it strikes me that you just articulated the hypothesis that we need to revisit in a couple of years. Right. The hypothesis that private capital, a concentrated environment with the right input stream, right of people with the right sort of aspiration and personality can actually augment into an extent, replace NIH, NSF, other public funding trenches that are being disrupted and commercialize more quickly.
00;36;55;04 - 00;37;15;17
Ian Bergman
Right. That's innovation. And like it's really interesting because we talk about measuring innovation by behavioral change. It'll be very interesting to come back in a couple of years and say, hey, because of dark matter, has private capital moved on mass into environments like this? How long do you think it's going to take? When do we have to come back for the round two of this conversation?
00;37;15;23 - 00;37;40;17
Emmanuel Vallod
So all mass we need to qualify all mass, hundreds of millions, five years. I think it's within five years. It could be faster than that. Of course. Hope it's faster than that. But I would say it's within five years. Billions. Probably within seven. Like, to me, if we get to hundreds of millions, it will be pretty quick to get to billions.
00;37;40;19 - 00;38;06;07
Ian Bergman
But this is this billions. This is not to be very clear. This is not, you know, the next generation of Bell Labs and private industrial capital, right? This is this is capital allocators finding, you know, a new vehicle that makes sense to get early access to the best kind of science and research grounded concepts. And I think that's super interesting.
00;38;06;08 - 00;38;19;20
Ian Bergman
Well, and look what dark matter is doing is fascinating. I'm going to be watching closely from the sidelines and peeking under the covers whenever you're let me. And wishing you a ton of luck.
00;38;19;21 - 00;38;22;18
Emmanuel Vallod
Thank you. I'm going to need a lot.
00;38;22;21 - 00;38;45;26
Ian Bergman
Yeah, well, you and the team, but you know, you you got it. You guys have the track record, you've got the resources. And this is very important. This is an extraordinarily important, I'd say, test in the market for the last few minutes before before we have to wrap up, I want to nerd out with you a little bit and we can build on something that's happening in the lab.
00;38;45;26 - 00;39;02;04
Ian Bergman
We can go a little bit more broadly, but I want to talk about Frontier Technologies. I want to talk about those big, crazy ideas that are being incubated. And I want to ask you to pick, is there one that you think is real and one that you think is theater off the top of your head?
00;39;02;07 - 00;39;20;19
Emmanuel Vallod
Well, if it was theater, I wouldn't bother with it, but there are some where I am clearer than others on how much technology risk has been taken away.
00;39;20;21 - 00;39;44;04
Emmanuel Vallod
And I'll give you a so, for example, what my tests advisor has been doing, I'm very clear on how much of the technological risk has been taken away. Part of that is I'm actually a supervisor of that research. I know that research inside out for the last four years. Right. And part of that is, well, industry wide validation.
00;39;44;04 - 00;40;32;02
Emmanuel Vallod
And it happens to be an industry where I come from when I worked on Wall Street. Now I'll give you another one. That's how I met your researcher there, Cyrus Clark. And he is working on robotics. And he's not notably working on robotics interfaces for artificial intelligence with humans. He's basically saying that the humanoid shape will make sense in industrial environments, but will not make sense in household environments because it's going to creep people out, which I'm very willing to believe because I actually share that view.
00;40;32;05 - 00;41;02;19
Emmanuel Vallod
But he's saying, look what people will be comfortable with in the context of their home or mundane shapes that end up interacting with us. And so he's rethinking interfaces to be everyday objects, tables, chairs, screens. Right. And he's saying, how do we bring in physical expressions of this software intelligence that end up creating a relatability?
00;41;02;20 - 00;41;11;17
Ian Bergman
What a fascinating, which is one of the underlying motivations in the humanoid form factor and how it presents in like science fiction. Right?
00;41;11;20 - 00;41;30;12
Emmanuel Vallod
The premise we have in these movies are like, well, if it looks like me, then I'm more likely to identify it to it, right? But he is going to a much more primal instinct, which is if it looks like me, but I know it's not like me, I'm actually going to see it as a threat.
00;41;30;13 - 00;41;58;08
Ian Bergman
What a fascinating, fascinating research space, right? Because it encompasses robotics and encompasses psychology. And, you know, and frankly, like practical logistics. But it is really interesting, right? Because, you know, humanoid robotics, the demos so often not in the obvious industrial applications where it just makes sense because the whole workflow is designed for humans and humanoids. They're often in the home, the bar, the hotel, the, the, the, whatever.
00;41;58;10 - 00;42;19;26
Ian Bergman
You just flipped that completely upside down and you basically said, let me see if I can paraphrase this back. You basically said that there's a frontier of research happening on giving human to human relatability, like qualities to everyday objects, robotics, things that we interact with.
00;42;19;28 - 00;42;53;27
Emmanuel Vallod
They showed me a demo that was super interesting. They have little glass bottles used for cooking, right? And they say, okay, these glass bottles for some people could be reminiscent of a family member because that was glass bottles the family member used to for cooking. Right. And they said, okay, could we use the glass bottle as the trigger element to have you basically remember or be brought back memories of that family member?
00;42;53;27 - 00;43;17;07
Emmanuel Vallod
And so the example is out, of course, of the lab director. His mom loves to cook, his mom loved listening to classical music. And so they have a system where you have these bottles. They are, of course connected to a support, which is a smart support that's multi-sensory. When you open the bottles, some of them trigger classical music that starts playing.
00;43;17;09 - 00;43;42;23
Emmanuel Vallod
So you open one, it plays the piano, you open the second one, it adds the violins, you open the third one, it adds the copper. Suddenly you have a symphony of classical music, right? Concurrently, they may start producing a romance used in cooking from when you were a kid and your mom was cooking. And so they were saying, okay, that interface is a very natural interface for the user.
00;43;42;23 - 00;44;01;00
Emmanuel Vallod
It's everyday home components with a smart bass and some some integrated sensors. But it looks very, very natural, very intuitive. It produces an interaction to you, which is not human language and yet is a clearly relatable interaction.
00;44;01;00 - 00;44;06;17
Ian Bergman
It's relatable and it's emotionally resonant. That's those are emotional cues.
00;44;06;19 - 00;44;43;10
Emmanuel Vallod
To me. This was a fascinating conversation, and I can see what certain technology risks are. I can see what certain use cases could be, but being between I can see and I know that these are commercially sizable. There is a canyon of ignorance and philosophical considerations that I am in right and need to finish traverse. So that gives you two complete opposite.
00;44;43;11 - 00;45;07;08
Ian Bergman
I absolutely I absolutely love this and I think it made some of the research very, very real. And now I'm going to go around looking at, you know, what are the relatable qualities of the tools and objects I interact with? Well, Emmanuel, unfortunately our time is coming to a close here. So I want to I want to end with just kind of one final question about the work that you're doing with dark matter.
00;45;07;08 - 00;45;16;25
Ian Bergman
And that is, you know, if Dark Matter Lab works the way that you hope and envision what becomes normal, that is not normal today.
00;45;16;26 - 00;45;27;29
Emmanuel Vallod
I think what will become normal is much larger, precedes in a lot more very technical projects.
00;45;28;00 - 00;45;48;27
Ian Bergman
Okay, I've got to go raise a bigger fund. No, but I, I love it, I love it. Well, I guess and the last question to wrap up is for folks in the audience who are smart listeners, they're like, I want in. I want to learn more. I want to follow the work of Emmanuelle or Dark Matter. Where do they go?
00;45;48;28 - 00;45;51;24
Ian Bergman
Are you on LinkedIn? Do you go to the website? Where do folks follow you?
00;45;51;25 - 00;46;27;24
Emmanuel Vallod
So we have a dedicated website, Dogmatic Hivemind Capital, where people can drop research papers they think are interesting. Right. And we'll progressively update as we onboard new labs, new universities, new industry resources. The second part is I have a technical blog called Dark Matter TLDR on Substack, where every week I do a deep dive in either A or researchers work or a labs work, or a particular topic tackled by different universities.
00;46;27;25 - 00;46;50;16
Emmanuel Vallod
And I like those research papers in terms of how do they do what they say they've done. And then obviously, as your stereotypical VC, I'm too loud on LinkedIn, not saying much, but moving, you know, smoke and screens so they can look at the wind moving. And I am not on Twitter because I'm apparently not cool enough.
00;46;50;16 - 00;47;13;19
Ian Bergman
To well, you know what? You've got the opposite extreme there with LinkedIn. But most importantly, you've got the Substack for your street cred annual. Thank you so much for coming on Innovators Inside. It was an absolute pleasure, and I look forward to reconnecting in a few years. To see how your hypothesis on early stage capital into research intensive ventures is working out.
00;47;13;20 - 00;47;14;14
Ian Bergman
Thanks again.
00;47;14;15 - 00;47;16;18
Emmanuel Vallod
And thank you so much. That was a pleasure.