AI is disrupting services. What does that mean for power demand?

About this episode

AI is still in the foothills of enterprise adoption, but the infrastructure buildout required to support it is already reshaping energy markets, capital allocation, and industrial policy. Data centers are driving demand for new generation and transmission, while billions of dollars flow into the hardware and software needed to train and run increasingly powerful models. 

That makes energy infrastructure a big constraint in the AI buildout race. Data centers require more power generation, transmission, grid capacity, and additional physical infrastructure — all major investments. And those energy systems will have impacts on the cost of running potentially more power-hungry frontier AI models. That’s a lot of uncertainty even before considering the economic transformation still to come.

If AI moves beyond software and begins to fundamentally change how services like law, accounting, insurance, and finance are delivered, it could reshape a far larger part of the economy and create a new wave of demand for both digital and physical infrastructure.

In this episode of Critical Capital, Alfred Johnson sits down with Jake Saper, a partner at Emergence Capital and an early investor in enterprise software and AI infrastructure. Saper argues that the biggest transformation from AI may not happen inside software companies at all. Instead, it could reshape the much larger services economy, creating a new category he calls AI-native services (AINS).

Saper explains why he believes companies built from scratch around AI may have an advantage over traditional service-business roll-ups, and why the biggest challenge for these companies is delivering their services with AI rather than simply adding more human labor. The conversation also explores the increasingly blurry line between software and physical infrastructure. 

Critical Capital is a co-production of Crux and Latitude Studios. Learn more about how Crux is financing the future of energy.

Episode transcript

Jake Saper:  The big unknown, which like no one really has a good answer to, is how costly it will be to run these models in the future. I think what's clear is that demand is going to be very, very high and is going to be an exponentially higher number than it is today.

Alfred Johnson: One of our recent guests on the show said, "You can't talk about AI without talking about energy." Well, it turns out the inverse of that is also true. The AI boom is driving the largest energy infrastructure build-out in a generation. It has the potential to drastically reorder society, reshaping everything from capital markets to industrial policy.

And beyond the headlines about breakthrough models and billion-dollar valuations lies an even bigger challenge: building the infrastructure to support an AI-powered economy. Utilities are racing to add generation and expand transmission. Investors are pouring hundreds of billions into energy and digital infrastructure, and policymakers are grappling with difficult questions around permitting, reliability, and America's position in the global AI race.

But the real economic transformation will come when AI begins to fundamentally change how businesses operate. Services are a dramatically larger market than software. If AI reshapes those industries, the scale of change will be unlike anything we've seen before. But how that happens is still a question mark.

Jake Saper: What is not clear is what is the infrastructure amount needed as the models get more efficient. And I think anyone who says they have a strong opinion is guessing.

Alfred Johnson: This is Critical Capital. I'm Alfred Johnson, the CEO of Crux, the capital platform for the clean economy.

My guest today is Jake Saper, a partner at Emergence Capital. Jake is a leading AI and software investor who's gone deep on energy. He believes we're still in the earliest stages of enterprise AI adoption, and he argues that the greatest opportunities won't come from retrofitting AI into existing operations. The best opportunities will come from building entirely new, AI-native companies.

Jake coined the term AI native services, or AINS, in 2024 when he first argued that AI would fundamentally reshape professional services. We'll talk about that, what it means for energy infrastructure, and how he sees the rapidly evolving AI landscape.

Jake Saper, welcome to Critical Capital

Jake Saper: I am so thrilled to be here, Alfred. Brings me so much joy.

Alfred Johnson: It is the best to have you here. I have been looking forward to this conversation since I asked you to do it. That is in part because you are one of my closest friends. It is also in part because I think you are a rare person to talk to at this moment because you have been a prolific SaaS investor. You are one of the leading voices on this emerging category of AI-native services, which we're gonna talk a lot about. And you started your career in energy and power working in solar in India, so let's set the scene there. I remember a story you tell of going over one of the solar fields in a hot air balloon that led to your DroneDeploy investment.

Jake Saper: That's right.

Alfred Johnson: How did that happen?

Jake Saper: The thesis behind the firm is that we wanna be early experts in emerging business models, hence the name Emergence. It's founded in 2023. The business model then that was emerging was SaaS.

The first investment was Salesforce. We did a lot of great early investing in horizontal SaaS. The next big wave was vertical SaaS, building, you know, SaaS specific to industries. The biggest hit that Emergence had, and actually the largest vertical software company today, still is Veeva in the pharmaceutical space.

So naturally, when I joined, in that era, I said, "I know a lot about natural resources, so I'm gonna develop a thesis around the natural resources cloud." So that is cloud software in agriculture, energy, water, et cetera. One of the very first investments that we made behind that thesis was a company called DroneDeploy, which was aiming to do drone software that would allow energy workers, construction workers, farmers to fly drones autonomously, gather a bunch of data, and make decisions.

This was in contrast to what was popular at the time, which was actually investing in drone companies. This was just a drone software play. Part of what gave us the conviction to make that investment was, they had talked about solar as a potential category. And to your point, when I was a developer, I had to get aerial imagery of my projects, both early on to scout the land, but also to prove to the banks that it was done. In Western India, in the Thar Desert in Rajasthan, it's really hard to get aerial imagery.

Alfred Johnson: That makes sense.

Jake Saper: So I ended up hiring a hot air balloonist who dragged his hot air balloon to the middle of the desert and launched it and took a bunch of pictures. And, you know, it was expensive even though it was India, because it was just such a hassle.

Alfred Johnson: Novel use case.

Jake Saper: It was a novel use case. So anyway, fast-forward to my job as an investor, and they're like, "Hey, could we use drones to do, like, solar stuff?" And I'm like, "Yes. I definitely know the answer is yes." And you fast-forward, you know,  a decade since we made the investment, and solar and energy is actually one of the largest categories for DroneDeploy. It's now the largest software provider to drones in the world. Very first board I joined. I'm very grateful to that experience, and it's been cool to see it happen.

Alfred Johnson: All right. So, now we're sitting in the present and everything is energy and compute, right? So we have come from this moment where you had to be squinty in order to see value creation and impact in clean energy to the largest physical infrastructure build-out in human history happening to serve hungry data centers. How are you seeing the coincidence of energy and software investing today? Where do you think that there are exciting opportunities and not?

Jake Saper: We made a decision three years ago, like 2023, shortly after GPT-3.5, that this open-source model ecosystem was gonna become a dominant way in which AI gets delivered. And so, what that basically means is that enterprises are ultimately gonna opt in large part to use open-source models because they will be cheaper, they will preserve more data privacy, and they will ultimately be more customizable, and enterprises will want that.

So we went big into a few companies, but one in particular called Together AI, which is a provider of inference and training services for open-source models, which is a heavy user of compute and data centers, and have gone deeper into the data center stack over time. That has played out well. We did the Series A there. They had a couple million in revenue. They just closed a series, I don't know what it is, and they're over a billion in revenue three years later.

I also like Baseten, which is another player in the space. I'm a big fan of that one.

 What's interesting is that Together investment was originally sort of a software investment in the sense that we were using software to optimize training and inference for open-source models. What it has become is an infrastructure investment, and infrastructure moving from software infrastructure to physical infrastructure. There's a blurring of the worlds at this point.

Alfred Johnson: Yeah, there is this really interesting other blurring, which is what will the ultimate adoption by the enterprise be? So a lot of people have quoted [Marc] Benioff saying that he spent some huge amount of money on AI last year, but it's a small fraction of what he pays software engineers. And there's this question of how far into enterprise adoption are we?

Are we on the doorstep or are we, you know, coming well up the curve? What's your sense on that as it then flows back into the question of how much compute we actually need?

Jake Saper: I think we are not even on the doorstep. What's interesting is Demis [Hassabis] uses this concept of like we're in the — I like this term — the foothills of the singularity, which I think is kind of like a beautiful and also very ominous phrase. I think I'm not smart enough to comment on whether or not the singularity is nigh.

What I can say is that I think we're in the foothills of enterprise deployment. Meaning, I think that we are so, so, so early, and I don't think I know because I see what traffic goes through Baseten and Together. I know kind of the types of companies that are using this and what types of companies are starting to experiment with it, et cetera. And we're still very early days, particularly for the larger, kind of more traditional enterprises, non-tech forward ones.

What I'd say is that like the vast, vast majority of dollars in AI have gone into developing these models. There's been obviously a shift over the past 12 to 18 months toward building the infrastructure to actually, you know, run those models.

But there's a third bucket of capital that is paltry relative to where it needs to go, which is deploying this technology across our country. And I think this is the thing that we in Silicon Valley miss, right? I think a lot of people here that never leave this 10-mile radius and take Waymos everywhere think that once AI hits AGI, first of all, they think it's gonna be a binary thing, which I don't believe. But they think there's gonna be a moment and then this is just gonna spread across like the speed of light and everything is gonna change. The real world is not that way. It's going to take a lot longer to deploy this stuff. Eighty percent of the American economy is a services-driven economy, and so until we find a way to get AI into our services businesses, this stuff is gonna take a while to deploy.

And so the amount of capital needed, and to your point, the amount of compute needed, is still infinitesimally low relative to the ultimate need.

Alfred Johnson: Do you think we'll look at hyperscalers spending $750 billion or whatever on CapEx this year in a few years and think it looks small?

Jake Saper: Yeah, I think it's very possible. The big unknown, which no one really has a good answer to, is how costly it will be to run these models in the future. I think what's clear is that demand for tokens is going to be very, very high and is going to be an exponentially higher number than it is today.

What is not clear is what is the infrastructure amount needed to deliver those tokens over time as the models get more efficient. And I think anyone who says they have a strong opinion is guessing.

Alfred Johnson: Yep. Okay, so let's talk about services. Much larger part of the economy than software. I was at your recent AI-Native Services Summit in San Francisco, which was awesome. You threw up a chart of just how much larger the services spend in the economy is than the software market. That has led you to a thesis on AI-native services.

What are AI-native services definitionally?

Jake Saper: I made that term up, so it's admittedly not a well-understood or well-known thing. The way, very simply, I think about how you define it is an AI-native service is any service that can be delivered using AI and is faster, better, and/or cheaper than the incumbent service. So that could be an AI-native accounting firm, an AI-native law firm, an AI-native insurance brokerage, an AI-native investment bank.

These are all examples of services across our economy that are today powered by labor, and my thesis is that you will be able to use AI to power a significant majority portion of that job to be done, and that for many of them, there will still be a highly paid, highly valued human who sits on top of that AI and checks it and provides their stamp of approval from their experience to say, "Yes, I bless this," and give it on to the customer and complete the outcome.

Alfred Johnson: So some investors have been focused on roll-ups — that is, buying and merging existing companies — whereas you've been focused on new builds, companies starting from scratch.  Why new builds versus roll-ups?

Jake Saper: I think they're gonna work better, I guess is the TL;DR. Ultimately, I think that the attractiveness of the roll-up is, okay, I've already got distribution. I can just buy a bunch of existing accounting firms, slam them together, and then throw AI on top, and then I've got distribution, I've got AI, I'll have a great business.

I think that, too, ignores the realities of humanity, which is that behavior change is really, really hard. Roll-ups in general are hard. Trying to make a bunch of cultures come together and work efficiently together has always been hard. The idea that you would then develop some magical AI platform and convince the people that you don't fire to change the way they do the job they've always done the job to do it with AI and work in this kind of conglomerate — I think it's just really, really hard.

I think it's much, much easier, although still exceedingly difficult, to start from scratch and say, "I'm gonna build an accounting firm. I'm going to build the GL from scratch using AI, and I'll build that GL for whatever use case I'm gonna be focused on. And then I'm gonna build agents that work specifically with that GL. I'm gonna try it with a few small customers, and I'm gonna do it myself as the lead. And then as it starts to scale, I'm gonna start hiring really impressive senior accountants to sit on top of this thing and train it, ensure it's good, ensure the outcomes are really good, and then kind of scale that way from an organic perspective."

The thing that people don't understand about AI-native services is that for the best versions of AI-native services, you generally aren't demand constrained. Because if you're selling an existing service faster, better, and/or cheaper than the incumbent, everyone's gonna wanna try you. So the question isn't "Can you sell a lot?" The question is "Can you deliver this primarily with AI in a high-quality way?"

And that is the biggest trap for AINS today.

Alfred Johnson: Yeah, that totally makes sense and resonates with what we see. If you're rolling up a bunch of accounting firms, you need to deal with the existing service providers, the people that are already in place, the general ledgers of those firms or record systems. So give an example of a company that you think is able to be significantly better, faster, cheaper using AI as a new build versus what incumbents are able to do, and what does that look like?

Jake Saper: To push on the accounting analogy a little bit, we invested in a company called Hanover Park, and what Hanover Park is, is a fund administrator. A fund administrator is an accountant focused on private equity firms. So private equity has their own kind of bespoke accounting, and obviously fund flow is critical to this business.

So there are very large existing services businesses called fund admin that serve private equity and venture capital firms to do all that work, to do capital calls, distributions, accounting, all that stuff. Historically, these have all been powered by people and very legacy, old software that's often sold by another firm and then used by the services firm that you're contracted with you as the venture fund or the private equity firm.

Hanover built the GL from scratch that's specific to their use case to private equity. They built these agents that talk to it, and they now serve private equity firms with a service that is much faster and much higher quality than the legacy providers. So to give you an example of what that means, if you wanted to do a distribution of a public stock, for example, the way that's done today, if you have a legacy, kind of labor-based services firm for a fund admin is, you know, first of all, the partners have a discussion and say, "Okay, I wanna sell, you know, X number of shares of Salesforce."

You then have someone on your accounting team email your fund admin. The fund admin then works through a bunch of Excel files to figure out, okay, which LP owns what, and how do I figure out the right sale process, and which bank do I contact, and what are the emails I need to send to the LPs once we actually do distribution? And then someone queues up all these emails. It's a process that can take days, if not weeks, to actually get done. But I wanted to sell Salesforce today, and if I have to wait a week, that's-

Alfred Johnson: This is not investment advice.

Jake Saper: This is not investment advice. Yeah, yeah.

Alfred Johnson: It's purely anecdotal.

Jake Saper: Yeah. I just pulled Salesforce, you know, randomly. But the broader point is, that's an example of where a faster service done by AI has real dollar value attached to it. And so, unsurprisingly, lots of private equity firms want to buy this. And so our challenge at Hanover Park is not sales. It would make no sense for us to buy another fund admin because we don't need that distribution.

Alfred Johnson: How do you think about that trade-off, right? We live this every day, 'cause when you're building an AI-native services company, you have to build the company, and the company has to be better at doing the thing and/or faster, cheaper, and you have to figure out how you're gonna redesign the business process.

And ultimately, there's this trade-off between revenue today, right? Like Hanover Park, go out and sell this until the cows come home. But then they would be putting less resources on innovation and actually making the services better. How do you think about that trade-off?

Jake Saper: It's messy. The biggest risk in AI-native services is what I call mirage product market fit. And what that means is you sell a bunch of your stuff and your customers could even be happy with the service you're providing, but to perform that service, you've had to hire a bunch of people. You're not actually an AI-native service, you're just a service that took the wrong form of capital. So you need to find a way to have an AI platform do it, even if it means growing more slowly. So some of the best AI-native services businesses, including Hanover Park, intentionally stop selling periodically so that the delivery capability and the AI platform can catch up to ensure that it is able to do 80 to 90% of the work and not people, which is a really hard muscle for a CEO to build.

Like it's crazy to think you would say, "No, I'm gonna stop selling," but the best ones are doing that.

Alfred Johnson: Yeah, I mean, there is this other interesting trade-off there of when you have something like that, right? Like they have a better product. Presumably others will come and try to do the same thing. So there is this competitive pressure that you have to go as quickly as possible, and there is benefit to having more reps, right? Like, the more throughput you can put through the system, the more data you can acquire, the better the system can be. Again, how do you think about when you're telling a company to go slower in order to build better? What are the things that you're looking to see them do?

Jake Saper: Well, I think, just like all things in venture and in life, it depends on the situation. There are some businesses that are in highly competitive industries where everyone is kind of nipping at your heels. Contract law, for example. There are now a number of contract law-focused AI-native services firms, so there's a bit of a market grab for that.

Something like fund admin is a little more obscure. Also, building the GL from scratch is really, really, really hard because you have to be a software company and you have to be a services company. The thing I often say about AINS businesses is that they're way, way harder to build than software companies because you have to build Stripe and McKinsey at the same time and have them integrate seamlessly. That is really hard to do.

Alfred Johnson: Yeah, let's talk about legal for a second. You just used the example of contract law. So there are a number of these new law firms that are established to be AI in the first instance, and I think it's a good example of a category. It's a category you know well. You are on the board of Ironclad. We see every day in our transactions there's an enormous amount of legal work that needs to be performed around the transactions, and we're seeing increasing adoption of applications by the law firms that isn't yet really flowing through into client value or lower costs.

Tell me how you see that category as an AINS category.

Jake Saper: I think that the last point you made is the promise for these contracts-focused AINS law firms, right? The idea is, look, we're gonna cut the crap. It's not like Harvey sells to Kirkland & Ellis, and then Kirkland & Ellis figures out how to use that more efficiently, but you never actually see the benefits in terms of fees or speed. We're gonna do it all. We're gonna take on the burden of figuring out the AI, and we're gonna use that to do the contracting faster and cheaper than the legacy provider, and we're probably gonna evolve the business model. Instead of charging per hour, maybe we just charge per contract, which is disruptive, obviously.

Alfred Johnson: I've been really interested in this theme recently because there was this phase in early AI, like post-GPT 3.5, where a bunch of the applications were doing a lot of post-training and putting a ton of resources in. And in categories like legal, that was good for a minute, and then Claude came out with a better legal model, and then it looked like all that post-training wasn't worth it.

I'm tracking in financial services. Bridgewater has this paper out this past week where they show that with an open-source model that they did post-training on, they were able to achieve dramatically better results at much lower costs. So it seems like there is a little bit of this rotation back to wanting to own the control of the model.

What do you think that that looks like as the models continue to get better and people get more sophisticated in their use of the tools?

Jake Saper: So I think there's three things that are driving that. The first is like there's a little bit of a hangover around the token-maxing spring that we all experienced.

Alfred Johnson: Those were heady times.

Jake Saper: Heady times, and it was a solid six weeks that that lasted. But I think there's still a little bit of a “Whoa, whoa, whoa, that was crazy talk. How do we actually get value out of this stuff?” Which pushed people to double-click on open-source models at the same time that the models just got even better. And I think that enterprises, specifically tech-forward ones, have gotten smarter about how to track data to build evals. I think in the beginning of this gen AI rush, people were like, "Yeah, I have a bunch of proprietary data," but no one actually knew which data was valuable and how to actually use it to properly post-train something.

Fast forward three years, the core open-source model, your post-training has gotten way, way better, and you have probably done a better job of collecting the eval data you need to actually post-train these things in an efficient way.

Alfred Johnson: Cost is a really interesting thing for AINS companies versus the applications because you, as the application, you're constantly having to figure out how to moderate your compute spend. But in AINS, you're typically operating in high gross margin transactions. And so it's more about how do you make the deal process better for the counterparties than it is how do you minimize the token cost associated to it.

How do you coach your AINS companies around that question?

Jake Saper: I don't think it's always true that the AINS businesses are operating in high gross margin environments because often the service that they're displacing is a 10, 20, 30% gross margin business. So they've gotta figure out how to do it mostly with AI. They're still gonna have some people cost on top of it. So we still have to be mindful of gross margin. We're not in a free lunch type of world.

And obviously, the tokens are replacing the labor, so if the tokens are equivalent cost to the labor, then you're not getting any alpha. So you gotta find a way to have tokens be efficient. I actually think that a tailwind for AINS businesses, a core competency they have to figure out, is, "How do I be as efficient as I can with model routing and my own token consumption?" And so I think part of the reason why enterprises are gonna want to outsource more and more stuff to an AINS business is they're not gonna wanna be the ones who have to stay on top of all the latest open-source machinations and how do I train this and whatever. Part of the value you're gonna pay to a third-party firm is for them to deal with that headache.

Alfred Johnson: Yeah. It's something that we're thinking a lot about because the transactions that we operate in are very high gross margin. At any moment, I would trade quality and speed for cost. At the same time, when I see something like the Bridgewater thing come out and it's dramatically better, there's this question of can you get to much better outcomes with more model control? Or can you just ride the wave of the foundation models getting better and better at different things, and you build better routing for different things that you're doing within the service, and that then leads you to a different conclusion of the way that you're setting up the underlying stack?

Jake Saper: To me it's an and. I don't think it's necessarily an or. I think you're relying on the frontier labs to push forward the frontier by definition. That's what their name suggests they will do. And your job is to figure out what are the tasks in this transaction that need that type of juice, and which ones can do my fine-tuned thing, and which ones can just use this super cheap open-source, off-the-shelf thing that's basically free to run.

Part of your value as the AINS provider here is to make that decision so your client doesn't have to.

Alfred Johnson: What other categories do you love? What do you think are the particularly high-value AINS categories?

Jake Saper: So we've done a lot in insurance. I think we continue to do stuff in insurance because it's a space that obviously has very high value, very high human labor, high data. So those are all places. The other thing that's nice about it is it's generally recurring transactions in insurance, and so that's another thing that makes it interesting, I think, as a place to play.

One other flag that's interesting that's a little counterintuitive is I tend to bias toward liking categories that should have a human in the loop indefinitely, which seems counterintuitive because you're like, isn't the whole point of this to get rid of the human and have infinite gross margin? The risk obviously is if you're in a service that truly can be provided exclusively by AI and there really is no need for a human for judgment, for cover your ass, for throw up to choke, for whatever, then ultimately maybe the frontier labs will just do it and you'll be out of business. Part of why I like the AINS model is that by keeping the human in the loop, by keeping the liability you are taking that on, that's part of the value that you're delivering as the vendor, and I think that gives you a little bit more of an enduring position.

So there are some businesses where regulatorily you need a human in the loop. One business I have not invested in but I'm curious about is customs brokerage. Customs brokerage is super paper heavy. It's always changing because Trump's constantly changing the tariffs. And it's done by these very highly credentialed people that have to go through this crazy intense, licensing process. And so it makes sense to hire a few of those people, have them train a big model, and then sit on top of the model and deliver the service.

Alfred Johnson: So let's make the transition into thinking again about what the impacts will be to the world that we live in. If all these things that we are foreseeing to be true come true — the enterprise adopts AI in huge numbers, we see a lot of people that had a certain kind of job no longer be necessary in that job, entry-level work becomes a lot harder — what do you think becomes true about society and the way that we need to operate?

Jake Saper: It's a question for a long hike.

Alfred Johnson: Yeah, let's do that, too.

Jake Saper:  I feel nervous about it. I am concerned that we are veering between extremes as a society and politically, and I'm concerned that our ability to manage all this AI stuff will get caught in those extremes. Because the reality is, AI is not evil, nor should we have a world that AI is unregulated.

I do think that this has become so unpopular that we have to figure out a way to address it.

What's interesting is the reasons why AI is unpopular are kind of all over the map. Part of it is electricity pricing, which I think is imminently addressable, by the way. I think there are things we can do. We can make the grid more flexible. We can make it cheaper to build. There's things that we can do that actually should help in most cases have data centers lower energy prices versus raise energy prices.

I also think that we as an industry have done a piss-poor job of explaining the benefits of AI to the end consumer, besides the vague promise of curing cancer someday. That's not enough. Until it cures cancer, we need to be clear about the things it's actually doing that are beneficial.

Alfred Johnson: I think it's gonna be really hard to get those politics back into a good place. Let's imagine that they do that. Substantially, the hyperscalers bring their power to the grids and it's to the benefit of consumers. I still think there's so many structural factors that are pushing up electricity prices — we just don't have enough power for all of the stuff, for air conditioners in Texas and EVs, and AI data centers as the least of them, right? That then creates this reality where electricity prices are just gonna go up. AI is a very obvious boogeyman on that as opposed to your air conditioner, which you probably like.

Jake Saper: I think that's probably true. I think you're right that this pinch point would've probably happened regardless. This is probably a Pollyanna too rosy version of the future, but there's a version of the future where we use this moment and say, " We're gonna use the build out of AI to try to support the communities they're building in and try to be as mindful as we can about energy price increases and try to find cheaper ways to build and provide power," all those things. And it can be a convening, coming together moment.

What's not clear to me is who? Ultimately when you're arguing for a logical, non-screamy policy, it's kind of a centrist-y thing, and centrist-y things are kind of out right now, so I'm worried about that.

Alfred Johnson: I think there's something on the positive side. I think regulation is hard. I think we have just gone through this era of big industrial policy — the infrastructure law, the Inflation Reduction Act — and then we have all of these exogenous forces. We have increasing geopolitical competition with China. We have rising electricity prices. We have the growing awareness that we need domestic supply chains of components and materials. So I actually think there's a pretty good case for more industrial policy to support the energy manufacturing complex. I think that that will probably happen.

I think that the hard thing is how you actually put in place the safeguards against true risk. Like, if we think that things like Mythos are just the beginning of what could represent a massive cyber threat globally, then we haven't seen shit yet.

Jake Saper: But to your earlier point, maybe we just need that to happen. Maybe there needs to be some terrible thing that happens with a Mythos-like product, and that is the catalytic moment where there's some public-private consortium that says, "No, no more, no mas. We're gonna fix this."

Alfred Johnson: Yeah. What's your take on the robot apocalypse?

Jake Saper: I'm less fearful of the robot apocalypse. You know this, I invested in a company called Bedrock Robotics, which is doing self-driving excavators and dump trucks and all sorts of construction equipment. Unsurprisingly, they have a lot of work with data centers and buildout of a big physical infrastructure. When that works at scale, we should be able to build data centers faster and cheaper, which ultimately hopefully will translate into cheaper power costs and everything.

So me, I am more of a bull that thinks that this will ultimately be beneficial. There are absolutely questions around the labor impacts of these types of things, but what's interesting is you have to be more granular than that and not be ham-fisted. Like in the case of excavator drivers, for example, there's a huge labor shortage, and frankly, the immigration crackdown has driven that shortage to be even higher and more extreme. So there's a lot of robots that have to be deployed before it actually impacts current people who are working.

That being said, there's gonna be transitions. My own view on this has changed a little bit over time, which is the most successful possible scenario for AI, I think, in our society is a medium deployment pace. If it goes too slow and we're outpaced by China and others where it is unfettered, that's not good. If it goes too fast and the societal changes required by massive unemployment that happens very quickly, that's also really scary and bad. So you almost want this world of the natural friction of our economy will prevent it from being rolled out too quickly, but but quickly enough hopefully, and people have time to retrain and re-skill and what have you.

You can hope for that scenario. You gotta plan for the tails. My fear is that we don't have enough people that are planning for those tails.

Alfred Johnson: Does the current pace feel medium to you? What does it feel like to you?

Jake Saper: I think the current pace of deployment feels medium, if not slow. What's interesting is if you leave your little tech bubble and you ask people, first of all, no one knows what Anthropic is. If you ask your friends, ask your mom and maybe your mom will know, but most people aren't gonna know.

And then, you know, I tell people that I take a robot every day to work and they don't believe me. When I am outside of a major city in America, there's just disbelief that that's true. And certainly you leave the country and it's abundantly clear that none of that stuff is deployed. So that kind of details my earlier point, which is like we're so obsessed with building this technology, we have not given as much thought or resources to the deployment of the technology throughout our economy and world.

And I'm glad in some ways that the deployment, I think, will not be instantaneous, but I do think we need smart founders to be working on the deployment part of the job.

Alfred Johnson: Okay, so back to the beginning. You went to solar in India because you felt like that was the best place where you could apply your talent to something that mattered, and it was related to things that you had seen growing up with your parents as founders. Where do you draw your energy and your joy in your work today?

Jake Saper: That's such a good existential question, Alfred.

Alfred Johnson: That's what I'm here for, baby.

Jake Saper: I mean, the first thing that you and I have talked about extensively, which is true, is like becoming a dad changes you, right? I think a lot about my girls and who I want to be for them and the type of world I wanna create for them or help create for them. It's cheesy, but that is absolutely a motivator that wasn't there before.

I have worked really hard, and you know this, to distance my sense of self from accomplishment.

Alfred Johnson: How's it going?

Jake Saper: It's a work in progress. I will say, being a dad and frankly getting older and doing more mentorship and less doing is actually really helpful there. You're still, there's a derivative of like, okay, I'm feeling proud of the impact I've had on someone else, but it's less like, "Look at the thing I did, Mom, give me praise."

Alfred Johnson: it feels less self-serving. You're actually helping somebody do something.

Jake Saper: Even though it is self-serving, because ultimately it makes me feel good about myself to feel useful to another person.

And I think actually being a VC is largely that, too. I work with these founders, my job is to sit with you and try to give you whatever advice I can to slightly increase the chances of being an iconic company. And if I get positive feedback from you, like, the best thing is, you know, it's great when you tell me, "Oh man, that was transformative. Here's how I changed my business as a result of that conversation."

That gives me joy in a way that has increased over time. I guess the way I think about it is like I'm getting more derivative joy now from trying to feel useful to other people

Alfred Johnson: Jake Saber, So great to have you here. Thanks for coming on Critical Capital.

Jake Saper: Thanks, Fredo .

Alfred Johnson: Jake Saper is a partner at Emergence. Critical Capital is co-produced by Crux and Latitude Media. Our production team includes John Sheehan, Jenna Herzog, Anne Bailey, Stephen Lacey, and Sean Marquand. Matthew Filler mixes the show. Additional production by Emily Hughes and the excellent team at Crux, the capital platform for the clean economy.

You can find Critical Capital on Spotify, Apple, or wherever you get your podcasts. I'm Alfred Johnson. Thanks for listening.