Deep Tech Strategy: Identifying Bottlenecks and Systemic Leverage

Lena: You know, I was looking at the latest hype cycle for deep tech—the kind of stuff that’s supposed to hit between 2026 and 2030—and it’s just a sea of headlines about "miracle" gadgets. But when you actually look at the system, it feels like everyone is looking at the shiny fruit and ignoring the roots.

Miles: That’s the classic mistake. If you want to understand where the real leverage is over the next few years, you have to stop following the headlines and start following the bottlenecks. I have High Confidence in this: the big winners won't be the companies making the most noise; they'll be the ones sitting on the obscure inputs that everyone else suddenly realizes they can't live without.

Lena: It’s counter-intuitive, right? We’re trained to look for the breakthrough, the "Eureka" moment. But you’re saying the real story is in the plumbing—the infrastructure that makes the breakthrough possible.

Miles: Exactly. Think of it as systemic thinking. You can’t have a massive leap in AI without a corresponding leap in compute, energy, and manufacturing. If you’re looking at Sam Altman at OpenAI or Jensen Huang at NVIDIA, you shouldn't just see a CEO and a chip designer; you should see two ends of a massive, straining pipe.

Lena: So, instead of asking "What’s the next cool app?", we should be asking "What is the one thing that, if it fails, stops everything else in its tracks?"

Miles: That is the only question that matters. And as we move toward 2030, those bottlenecks are shifting from the digital realm back into the physical world—into things like power grids, orbital mechanics, and the actual atoms of hardware. Let’s dive into how this system actually hangs together.

Lena: If compute is the engine of this decade, energy is the fuel. But I feel like we underestimate just how much fuel we’re talking about.

Miles: It’s a massive feedback loop. We have High Confidence that AI scaling laws aren't just a software problem—they are a thermal and electrical problem. Look at Jensen Huang at NVIDIA. He isn't just selling chips anymore; he’s selling the architectural blueprint for how a civilization processes information. But that blueprint requires a level of power that our current grid wasn't built to handle.

Lena: Which explains why we’re seeing deep tech founders moving into energy. It’s not a pivot; it’s a prerequisite.

Miles: Right. Look at someone like Bob Mumgaard at Commonwealth Fusion Systems or Michl Binderbauer at TAE Technologies. They are working on fusion—High Confidence claim here: fusion is the ultimate long-term solution to the compute-energy bottleneck—but the "hidden leverage" there isn't just the plasma physics. It’s the high-temperature superconducting magnets that Mumgaard is developing. Without those magnets, the whole fusion dream stays a dream.

Lena: So the magnet is the bottleneck within the energy bottleneck?

Miles: Precisely. And while fusion is the "holy grail," the immediate leverage is in advanced nuclear—SMRs, or Small Modular Reactors. Stefano Buono at Newcleo and José Reyes at NuScale are trying to build the "plug-and-play" power plants that could sit right next to a data center.

Lena: It’s like we’re seeing a re-industrialization of the tech sector. You can’t just be a "software guy" like Sam Altman anymore; you have to worry about where the gigawatts are coming from.

Miles: You really do. If you don't own your energy supply, your AI roadmap is just a wish list. This is the first-principles reality of 2026: bits require atoms, and atoms require heat.

Lena: We’ve been hearing about self-driving cars forever, but the conversation seems to be shifting away from "when will my car drive me to work?" to something much more systemic.

Miles: Because the bottleneck moved. For a long time, the bottleneck was perception—getting the car to "see." Now, we have Moderate Confidence that perception is largely solved, but "Physical AI"—the ability for a system to interact with a messy, unpredictable world—is the new wall.

Lena: I noticed Raquel Urtasun at Waabi is talking about a "sim-first" approach for autonomous trucking. That feels like a very different strategy than just putting more sensors on a truck.

Miles: It’s a brilliant move because it addresses the data bottleneck. You can’t drive enough real-world miles to encounter every weird edge case, so you build a high-fidelity simulation where the AI can "live" a thousand lifetimes in a day. It’s moving the problem from the road to the data center.

Lena: And it’s not just trucks. We’re seeing this in defense too, with companies like Anduril. Palmer Luckey isn't just making cool drones; he’s talking about autonomy as a "software-defined" capability for the entire military.

Miles: Luckey is a great lens for this. He realized that the bottleneck in defense wasn't the hardware—the Pentagon has plenty of hardware—it was the lack of an integrated, autonomous operating system that could connect those gadgets. That’s the "hidden leverage." If you control the autonomy software, the specific drone or sensor becomes a commodity.

Lena: So the leverage is the "brain" that coordinates the system, whether that's a fleet of Waabi trucks or a swarm of Anduril drones.

Miles: Right. And you see that same logic with Adam Bry at Skydio. They started with consumer drones, but they realized the real value is in "computer vision" for infrastructure inspection and defense. They’re not selling a flying camera; they’re selling the ability to navigate a complex environment without a human pilot.

Lena: Speaking of navigating complex environments, space seems to be moving from a "look at that rocket" phase to a "how do we actually live and work up there" phase.

Miles: It has to. If you look at the 2026-2030 horizon, the bottleneck isn't getting to orbit anymore—SpaceX and Elon Musk have largely smashed that with reusable launch. Now, the bottleneck is what I call "orbital logistics." How do you refuel? How do you fix things? How do you stay up there?

Lena: That makes me think of companies like Orbit Fab. They’re working on "in-space refueling," right?

Miles: Exactly. Daniel Faber at Orbit Fab / Deep space ventures is building in-space infrastructure. Think about the leverage there. If every satellite currently in orbit is a "single-use" asset because it eventually runs out of fuel, the person who provides the gas station suddenly changes the entire economics of the industry.

Lena: It’s that "indispensable before it’s obvious" takeaway. No one cares about gas stations until they’re stranded on the highway.

Miles: And it goes deeper. Nobu Okada at Astroscale is looking at the debris problem—debris removal and orbital servicing. We have Moderate Confidence that if we don't solve the "space junk" bottleneck, the entire Low Earth Orbit economy could collapse under the weight of its own collisions.

Lena: So the "hidden leverage" in the space race might not be the loudest rocket, but the most efficient trash collector or the most reliable fuel depot.

Miles: That’s first-principles thinking. Peter Beck at Rocket Lab is doing this too. He’s moving from just "launch" to "in-space systems"—building the actual satellite buses and components. He’s looking at the system and seeing that the rocket is just the elevator; the real business is what happens on the floors once you get there.

Lena: We’ve talked about energy and autonomy, but you mentioned manufacturing earlier. If we’re going to build all this stuff—the rockets, the SMRs, the AI hardware—we need a completely different way to make things.

Miles: This is where it gets really interesting for the 2026-2030 period. The traditional manufacturing bottleneck is "tooling"—it takes forever and costs a fortune to set up a factory for a new part. Tim Ellis at Relativity Space is trying to bypass that entirely with massive 3D printing for rockets.

Lena: I’ve seen their printers. They’re like giant robotic arms just "printing" an entire rocket body.

Miles: It’s a paradigm shift. If you can print a rocket, you can change the design on the fly. You’ve moved from "static manufacturing" to "software-defined manufacturing." And that logic is spreading. Look at the materials side—JB Straubel at Redwood Materials.

Lena: He was one of the early Tesla guys, right? Now he’s doing battery recycling.

Miles: Right. He identified a massive bottleneck: we don't have enough raw minerals to power the EV and grid-storage revolution. So, the "hidden leverage" isn't a new mine; it’s a "circular economy" where you treat every old laptop and car battery as a high-grade mineral deposit.

Lena: And then you have Gene Berdichevsky at Sila Nanotechnologies working on the anodes themselves—using silicon to make batteries more efficient. It’s all about squeezing more performance out of the same atoms.

Miles: We have High Confidence that materials science is the ultimate bottleneck for hardware. You can have the best design in the world, but if you can't make it out of something that survives the heat or stores enough energy, you’re stuck. The people who solve the "atom problem" are the ones who will define the next five years.

Lena: There’s one bottleneck we haven't touched on yet, and it’s a big one: the human one. Or maybe the "safety" one. As AI gets more powerful, doesn't the risk itself become the bottleneck?

Miles: Absolutely. This is the "social license" bottleneck. If the public or regulators don't trust the tech, the deployment stops, no matter how good the compute is. This is the lens through which to view Dario Amodei at Anthropic or Ilya Sutskever at Safe Superintelligence Inc..

Lena: Sutskever leaving OpenAI to start a company specifically focused on "Safe Superintelligence" feels like a major signal.

Miles: It’s a bet that safety isn't just a "feature"—it’s the fundamental bottleneck to AGI. We have Moderate Confidence that without breakthroughs in "interpretability"—actually understanding why a model does what it does—we hit a ceiling on how much we can let AI control.

Lena: So safety becomes the "infrastructure" beneath the breakthrough. If you can’t prove it’s safe, you can't ship it.

Miles: Exactly. And you see this in the "tech governance" space too, with people like Tristan Harris at the Center for Humane Technology. They are flagging the bottleneck of human attention and societal stability. If the technology breaks the social fabric, the technology itself gets throttled.

Lena: It’s fascinating that the "hardest" part of deep tech might actually be the "soft" stuff—the ethics, the safety, the trust.

Miles: Because those are the inputs that are "indispensable before they're obvious." Everyone wants the super-intelligent assistant; no one wants to think about the safety protocols until the assistant starts hallucinating or doing something dangerous. The "hidden leverage" is the person who can guarantee that the machine stays in the box.

Lena: While we’re looking at the horizon, I have to ask about quantum. It’s been "five to ten years away" for a long time. Is it still stuck in a bottleneck?

Miles: Quantum is the ultimate bottleneck story. The bottleneck there is "error correction"—getting qubits to stay stable long enough to do something useful. We have Low Confidence that a truly fault-tolerant quantum computer will be commercially available before 2030, but the players are positioning themselves for the moment that bottleneck cracks.

Lena: Like Jeremy O’Brien at PsiQuantum? I know they’re taking a "photonic" approach.

Miles: They are. They’re betting that the bottleneck isn't the qubit itself, but the "manufacturability" of the system. By using light instead of super-cooled circuits, they’re trying to leverage existing semiconductor manufacturing to scale up.

Lena: And then you have Chad Rigetti at Rigetti Computing and the team at IonQ working on different architectures—superconducting vs. trapped ions. It’s a multi-front war on the same bottleneck.

Miles: It’s a great example of looking for the "obscure input." If quantum does work, the "hidden leverage" might be in "post-quantum security"—the encryption that can survive a quantum attack. Michele Mosca at evolutionQ is already working on that.

Lena: That’s classic. While everyone is trying to build the "quantum key," he’s building the "quantum-proof lock."

Miles: That is exactly the kind of strategic thinking you need. You don't just ask "Who builds the computer?" You ask "What happens to the rest of the world the day after that computer is built?" That’s where the real opportunities are hidden.

Lena: So, if I’m trying to apply this analytical, first-principles mindset to the next few years, what’s my playbook?

Miles: First, you have to ignore the headlines and find the "unsexy" necessity. When everyone is talking about a new AI model, you should be looking at the companies building the specialized cooling systems for the data centers that run it.

Lena: Right, "follow the bottlenecks." If a technology is scaling rapidly, what is the one thing it's going to run out of first?

Miles: Exactly. Is it energy? Is it rare earth minerals? Is it "verified" human data? Second, you need to look for "systemic leverage." Don't look at a gadget in isolation; look at it as part of a stack. A drone isn't a drone; it’s an endpoint for a perception engine, which is an endpoint for a compute cluster.

Lena: And finally, "look for what becomes indispensable before it's obvious." Like the orbital refueling or the battery recycling we talked about.

Miles: Right. Ask yourself: "If this whole sector succeeds, what is the one service or material that will be under the most pressure?" That’s your hidden leverage. Whether it’s JB Straubel’s recycling or Raquel Urtasun’s simulation data, the real power is usually hidden a few layers down in the infrastructure.

Lena: It’s about being the "toll booth" on the bridge that everyone has to cross.

Miles: Precisely. In deep tech, the toll booth is usually a specific patent, a unique manufacturing process, or a massive, proprietary data set. If you can identify that before the "bridge" is even finished, you’ve already won.

Lena: This has been such a shift in how I think about "innovation." It’s less about the "spark of genius" and more about the "pressure of the bottleneck."

Miles: That’s the most rigorous way to see it. Innovation is a response to a constraint. When you see a massive constraint—like the energy cost of AI or the difficulty of orbital logistics—don't see it as a deal-breaker. See it as a map of where the next billion-dollar company is going to be built.

Lena: It’s a more grounded way of being optimistic, I think. You’re not just hoping for a miracle; you’re looking at the engineering realities and seeing the path forward.

Miles: It’s the "Marc Andreessen" approach: direct, skeptical of the hype, but deeply convinced by the logic of progress. The world is a system of competing bottlenecks, and the people who clear them are the ones who move us into the future.

Lena: I love that. Well, thank you for walking through this with me. It’s given me a lot to think about the next time I see a "breakthrough" announcement.

Miles: My pleasure. Just remember to ask: "Where’s the plumbing?" It’s the most important question in tech.

Lena: And thanks to you for joining us on this exploration. Next time you read about a new deep tech "miracle," take a moment to look past the shiny exterior. Ask yourself what obscure input is making it possible, and where the next bottleneck might appear. We’ll see you next time.