AI Meets the Physical Economy

Morgan Stanley Research analysts Michelle Weaver, Ravi Shanker and Dave Arcaro discuss two industrial inflection points: how long it will be before autonomous trucking becomes a reality and why power infrastructure is racing to keep up with AI-driven demand.

Read more insights from Morgan Stanley.

----- Transcript -----

Michelle Weaver: Welcome to Thoughts on the Market. I'm Michelle Weaver, Morgan Stanley's U.S. Thematic and Equity Strategist.

Ravi Shanker: I'm Ravi Shanker, Morgan Stanley's U.S. trade transportation analyst

Dave Arcaro: And I'm Dave Arcaro, Morgan Stanley's Utilities, Power & Clean Energy analyst.

Michelle Weaver: Today, what we learned at Morgan Stanley's Industrials Conference about the changing economics of autonomous trucking and the increasingly tight power market supporting the AI build-out.

It's Friday, September 25th at 10am in New York.

Now, I know we're all on the road taking meetings post-conference, so the audio might sound a little bit different, but we wanted to bring you the latest from our annual Industrials Conference that recently concluded in Laguna Beach, where two themes really stuck out. The growing physical infrastructure demands behind AI, particularly power, and the shift in autonomous trucking from proving the viability of the technology to commercializing it at scale.

Ravi, after roughly a decade of development, you've said autonomous trucking is entering a critical 12 to 18-month period ahead of serial commercial production.

What's changed, and why is the debate shifting from whether the technology works to whether it can be commercialized at scale?

Ravi Shanker: I think for 10 years the industry has been focused on making the technology work. but with players like Aurora now putting up almost half a million miles of fully driverless revenue-generating operations, on public highways in the U.S., day and night, rain and shine, for different customers. With people like Kodiak, also running, several trucks, in revenue-generating service, for customers like Atlas, I don't think there is much debate on the technology itself.

And so, I think the debate is now moving from does this work to can this work for me? Where the next steps are going to be dotting i's and crossing t's on the path to actually pressing these trucks into commercial service rather than having to prove that it works in the first place.

Michelle Weaver: Your research suggests that autonomous trucking can deliver roughly a 20 percent lower cost per mile, while higher utilization could be an even bigger source of value. What are the key assumptions behind that math? And what still needs to happen operationally for fleets to capture those benefits?

Ravi Shanker: Yeah, so we recently updated our TCO math, on autonomous trucks and published a North American insight, where we revised and revisited our views on autonomous trucking with a lot of proprietary data, in there as well. And part of that new TCO math, again, I think revisited some of the changes in the split of operating costs of trucking over the last several years.

First of all, I'll kind of throw a huge disclaimer out there that your mileage may vary, right? Because, depending on who you are as a trucker, if you're public or private, small or large, dry van or reefer, heavy or asset light, long haul or short haul, your split of costs are going to be slightly different.

But we started out, by looking at the ATRI's national average. And labor accounts for 35 to 40 percent of the P&L of the average trucker. So, when you take the driver out and substitute that with an autonomous driver, if you will. Even after paying the autonomous technology company roughly 85 cents a mile, for the autonomous operation, you will still save a significant amount of money. Versus the 40 percent of the roughly $3 per mile that it costs for labor today.

In addition to that, fuel is another third of your cost structure. And there, an autonomous truck should be anywhere from 13 to 22 percent more fuel efficient. We have taken the low end of the scale to be conservative. And then you layer on insurance savings, maintenance savings on top of that. Even if you add some incremental costs, either for human drayage at both ends or for the truck itself being more expensive – we believe you will save about 20 percent per mile versus a human driver today.

And I'll point out that the unit economic savings are only about a-third of the total savings with the utilization benefit driving another two-third savings on top of that.

Michelle Weaver: But there, there still seems to be a notable disconnect between how much freight carriers and shippers think can be automated and how much of the network may actually be suitable to be automated. What's the industry potentially underestimating?

Ravi Shanker: Yeah. We have seen this in our conversations. Again, part of our report was conducting detailed surveys and in-depth interviews with a lot of our coverage companies. And I will say that there still needs to be a lot of education, of how these trucks work, where they work, what the unit economics are going to be out there.

There's still a lot of misinformation. For instance, there's this big perception that you still need human drivers at both ends of an autonomous truck move because these trucks can only operate on a highway. And here's where our AlphaWise analysis, comes in. I think it's the first of its kind analysis where we use geolocation data to pinpoint 10,000 plus of the largest commercial facilities belonging to the hundred largest commercial shippers in the U.S.

And we found out that the average [00:05:00] commercial facility is less than two miles away from the nearest ramp point. And these trucks can comfortably do seven to 10 miles, if not longer, off a highway on main roads to get to their end destinations. So, I think you just need a lot of education in the industry.

And that is part of the dotting of i's and crossing of t's that we think the industry needs to do in the next 12 months before we see the start of serial commercial production next year.

Weaver: Thanks, Ravi. I want to bring Dave into the conversation here, and that question of turning demand into real world capacity brings us naturally to power, where the challenge is also increasingly about physical infrastructure and execution.

Dave, coming out of Laguna, you describe management commentary across power equipment as notably positive. What surprised you most about what you heard on demand bookings and project activity?

Arcaro: Yeah, absolutely. What surprised me most was probably how consistent the commentary was across companies, across large frame turbine providers and the smaller, on-site power equipment players, the new entrants and the more mature companies in the market. Very consistent feedback. All very positive.

And I would say also what surprised me too was the lack of disruption across the board. You know, we all see the headlines about data center moratoriums, political pushback, community challenges that really, it seemed, to increase the risk of data center execution and delays out in the market.

But at least with the power equipment companies, they're just not seeing it. You know, in terms of the feedback that we heard from management teams across the board at Laguna, they review project timelines actively with their customers, and that's all still intact. We haven't seen any changes in bookings or slot reservations for equipment deliveries.

Still seems to be a very stable and very strong backdrop across the board.

Weaver: One of the broader conference themes was the availability of power is becoming a bottleneck for AI infrastructure. How are equipment shortages, longer wait times, and customers planning further ahead affecting pricing? And how far ahead can the industry see?

Arcaro: Yeah, we are seeing equipment companies booking out orders farther and farther. The large frame gas turbines, to give you a couple examples, from companies like GE Vernova, they're now in conversations to contract turbines for 2031 and 2032. Smaller equipment companies like INNIO, who make, smaller scale engines for data centers, they're in conversations with customers and taking reservations into 2029 and 2030.

So, what we heard from the conference as well was that utilities, which is a big customer for this equipment, they're looking out farther and farther now into the 2030s. That's new and that's a surprisingly long time in terms of how far they're looking out. And we're also hearing data centers looking out toward the end of the decade, you know, late 2020s in terms of trying to secure their power equipment in advance.

We would still consider it very much a seller's market. Pricing has been rising, and companies at the conference gave further indications that it's likely to keep rising, what looks like into the 2030s from here. We just haven't seen any signs of softening yet, really regardless of the company or the equipment type that they're selling into the market.

So still farther and farther out that we're seeing visibility into the order flow, and with that is also coming firm and even rising prices into the 2030s.

Weaver: Investors often frame the power debate as electricity from the grid versus smaller power sources built on-site at data centers. Based on what you heard at Laguna, how should investors think about the balance between those two approaches?

Arcaro: Yeah, it's an interesting dynamic. When you talk to utilities and some of the large frame turbine companies, they all say that all this data center demand is going to the grid. Eventually, it's all going to go to the grid. When you talk to the smaller equipment manufacturers and the power as a service providers, they say nobody wants the grid.

They see long-term opportunities to sell, on-site power equipment and contract it with their end customers for 15 to 20 years, and we're seeing evidence of that. So, I think, it'll stay It's an ongoing debate among, investors as well. On our end, we think the on-site power market is going to be an extremely large market as we get toward 2030, given limitations in how much power is likely to be accessible from the grid over time for the data center industry.

But I would say, my takeaway and my observation from the conference that I would highlight is that it's a really favorable market and favorable backdrop for both sides.

Michelle Weaver: From autonomous freight to the power needed to support AI, one message from Laguna was clear. The next phase of technology adoption increasingly depends on what the physical economy can actually build and scale.

Ravi and Dave, thanks for taking the time to talk.

Shanker: Thanks, Michelle.

Arcaro: Thanks for having me.

Weaver: And to our listeners, thanks for listening. If you enjoy Thoughts on the Market, please leave us a review wherever you listen to the show and share the podcast with a friend or colleague today.