America’s industrial backbone took generations to build. Today, investors and entrepreneurs are betting that the United States can modernize its industrial infrastructure much faster with today’s technology. From defense technology and domestic manufacturing to nuclear energy and artificial intelligence, a new wave of companies is trying to apply the speed and mindset of the software industry to sectors that have traditionally been much harder to build at the speed of tech. What will it take to make that shift? How can startups compete in industries long dominated by established government contractors? And as AI infrastructure expands, what regulatory changes could make it easier to build in the United States? 

David Ulevitch joins us for an episode of Explain to Shane to help answer these questions. David is a general partner at Andreessen Horowitz, where he leads the firm’s American Dynamism practice. He previously founded EveryDNS and OpenDNS, the latter of which Cisco acquired in 2015. After the acquisition, David served as senior vice president and general manager of Cisco’s security business before joining Andreessen Horowitz in 2018.

Below is a lightly edited and abridged transcript of our discussion. You can listen to this and other episodes of Explain to Shane on AEI.org and subscribe via your preferred listening platform. If you enjoyed this episode, leave us a review, and tell your friends and colleagues to tune in.

Shane Tews: Your career began in software and cybersecurity. What made you more willing to bet on hardware and hard technology?

David Ulevitch: Cisco was the first time I had been exposed to any kind of hardware business. We ran the largest firewall business in the world.

At Cisco, I found out that hardware is hard. You have to plan years in advance. You have to deal with a complex supply chain. Unlike software, where you can keep making edits, revisions, patches, and changes, if you only have 95 percent of the parts for a hardware product, you have nothing. You need all the parts of your bill of materials in order to ship a working product.

That mindset is difficult and challenging, but I really valued the exposure. When I left Cisco in 2018 and joined Andreessen Horowitz, hardware was becoming more approachable. Applying a software mindset to hardware and manufacturing actually made sense. That made me more open-minded to hardware, which traditionally had been out of favor with investors and the entrepreneurial community.

When you joined Andreessen Horowitz in 2018, defense startups were still a difficult venture bet. What made you think that was going to change?

The government was starting to realize that it had technology gaps, supply problems, and production problems. This was still early, but Ash Carter had already stood up the Defense Innovation Unit to build a stronger connection between Silicon Valley and Washington.

The roots of venture capital in Silicon Valley actually had a strong partnership with the Department of Defense, but that had fallen out of favor. Hardware took too long, selling to the government was too slow, and building a large company took too long. It did not meet a venture timeline. It was much easier to invest in consumer and enterprise software companies.

Another thing that had happened was Project Maven. The government wanted access to the most advanced technology and had entered into a partnership with Google, but employees protested the project and some resigned. My reaction was, why wouldn’t we want the government to have the best technology and capabilities? Why wouldn’t we want the men and women who serve our country to have the most advanced software and the best ability to defend the country?

There was this mindset that you should not work with the government or support it. At the same time, the technology landscape was changing. We were seeing the beginnings of machine learning and computer vision. Entrepreneurs were waking up to the idea that hardware was possible and that you could apply a software mindset to it.

Drones are a great example. Could we apply commodity hardware and much more advanced software for computer vision, navigation, production, and design to dramatically lower the cost of making drones? I believed the answer was yes. Would the demand side from government start to transform? I believed the answer there was going to be yes, too.

My partner Katherine and I built American Dynamism around this idea that we could bring advanced technology into government, put it in the hands of war fighters quickly, and do production quickly. We felt the previous objections—that hardware was too hard and the government was a crappy buyer—were going to start to erode.

A lot has happened since then to accelerate that shift. COVID exposed massive supply-chain dependencies and showed that we lacked national resilience in many things we needed. The war in Ukraine changed how we think about mass-produced, low-cost systems operating in GPS-denied and electronically contested environments. In the Middle East, we have seen low-cost drones intercepted with multimillion-dollar Patriot missiles that we cannot resupply at a steady pace.

Those events have pushed different parts of government in the same direction. There is a realization that we have to change procurement, appropriations, and acquisitions. We cannot only bet on large, exquisite systems. We need to be able to field attributable systems and get them into the hands of war fighters quickly. The government has been opening doors, creating new programs, and facilitating new pathways for acquisition.

What makes you so bullish on nuclear microreactors, and what has building in that sector taught you about regulation?

I think being a founder prepares you to get punched in the face on a daily basis when you really believe you are doing something important. Building a nuclear power company, a reactor company, and a factory to make reactors—something that has never been done before, like we are doing at Radiant Nuclear—means dealing with an incredibly frustrating regulatory environment and a public perception that needs to change.

Large swaths of the American public have been told for 50 years that nuclear is dangerous, dirty, and creates a waste problem. At the same time, the regulatory, environmental-policy, and permitting environment has become incredibly calcified with requirements and policies that need to be streamlined or better facilitated.

When we look at what America needs and what improves the national interest, energy is a key component. When you dig into energy, nuclear starts to rise to the top as one of the greatest opportunities.

Then you start digging into the objections. How many people have died from nuclear power incidents in this country? Even at Three Mile Island, our worst nuclear accident, zero people died. Then you compare that with deaths associated with fossil fuels, coal mining, and their carcinogenic effects.

The next question is affordability. What inhibits making nuclear affordable? Maybe we need smaller reactors instead of very large reactors that come in over budget and over time. Can we put a reactor behind the meter? Can we avoid some of the grid-interconnection issues and other things that slow down nuclear projects? We looked at that and thought we could.

For me, one of the realizations that came from our work in defense is that there is no national security or defense industrial base without reliable energy and a reliable grid. There is no ability to grow our GDP without reliable energy. We are entering a moment where we have an insatiable thirst for energy, whether it comes from electrification, AI data centers, or something else.

That is where microreactors become really interesting. A microreactor is not about powering a city of millions of people. It is about solving a particular problem. That could be powering a forward operating base overseas where you do not want fuel trucks coming in every eight hours, keeping a domestic base running through a power outage, or powering a data center or charging center.

I tell people to think of a microreactor the way you think of a diesel generator, except instead of refilling it every eight hours, you are refilling it every five years and providing significantly more power. It is more reliable, much cleaner, and logistically much easier to deploy.

Radiant’s Kaleidos reactor, for instance, is about the size of a shipping container. You can put several together to increase the power output, move them on a plane, or transport them on a semi truck. Imagine a natural disaster where FEMA brings in four trucks with reactors and another four with desalinators. You could provide power and water for thousands of people within 24 or 48 hours and keep providing power for years. That is something we cannot do today.

Do we need a change in public mindset as much as a change in regulation when it comes to AI infrastructure and data centers?

We’re living in peak misinformation. What’s happening to AI and data centers is the same thing that happened to nuclear 50 years ago. They completely derailed it, and we can’t let that happen. We need to correct the misinformation.

People think AI data centers are consuming all the water. They think they are destroying farmland or taking away people’s jobs. Those things are not happening. We have seen that when we have major technological transformation moments, they create much more opportunity. We need to make sure that opportunity is evenly distributed, that people who get left behind get taken care of, and that new opportunities are created for new people.

AI and data centers are supremely important. We cannot get in our own way and allow adversaries to get way ahead of us. It is in the national interest, it matters for national security, and it matters for economic growth and development.

We should have more data centers. We need to build as many as we possibly can. We will find ways to use them all. It is like the internet. The internet has been an amazing boon to our country and given us a huge technical and economic advantage, and the same will be true for AI.

Looking toward the next decade, where should policymakers clear the regulatory path to make it easier to build?

Permitting reform is long overdue across a number of dimensions. We have bureaucratic organizations that have added rule after rule after rule.

The accounting and audit overhead for companies that sell to the government can also be dramatically reformed. When private capital is funding R&D efforts, we do not need the same amount of auditing, accounting overhead, and oversight that is appropriate when the government itself is funding the R&D. When we are selling commercial solutions to the government, we take on the R&D risk, and we can take on the accounting and audit risk. That can save taxpayers money and time.

I think the same is true for AI policy. We probably need federal rules that supersede all the state and local rules that are going to get passed, so companies that want to create AI companies or operate data centers have clarity.

We just need to pave the way. It is not that all regulation is bad, but I think the government should view itself as a regulator that facilitates growth and development rather than as an inhibitor. That is a mindset shift we need.

If we see that, the next 10 years can be an incredible economic growth opportunity for this country. We can revitalize the manufacturing and industrial base, not just across defense but across the entire country. We can satisfy our demand for energy with new power plants and factories that make nuclear reactors. I think it can be a really exciting time ahead.