Expeditors International of Washington, Inc. (EXPD) Earnings Call Transcript & Summary

Good morning, everyone, and welcome to our webinar today. My name is Olivia Tan, and I am a senior geopolitical analyst at Onyx. So we offer a different webinar topic each month. And this time, we will provide our AI outlook. So the AI boom has led to a physical buildout of chips, service, data centers and power infrastructure globally, and the speed and scale of this hardware build-out is actively reshaping supply chains and logistics strategies. So join our analysts today as we assess the impacts of the AI boom to air and ocean freight markets, shipping lanes and discuss longevity of this build-out. Now before we begin, we've seen a lot of interest in the application of AI to logistics operations, notably in shipment visibility and service improvements. So however, in this webinar we'll focus instead on the impact of the AI investment boom on freight markets and supply chains. Before we begin with the content, there are just a few administrative details to cover. We will have about 45 minutes of content to share, and we will save the last 15 minutes for the Q&A session. Please submit your questions in the Q&A box, and we will do our best to address your questions during our Q&A session. A copy of the presentation, notably the slides, will also be available later. To receive a copy of the presentation, please fill out the brief survey that will be e-mailed to you shortly after this webinar. And please visit our website as well and subscribe to receive information on Onyx' future webinar. We would also like to invite you to explore our latest insights on LinkedIn and the Vantage Point blog, which features a makes offshore updates and in-depth articles. So please use the QR codes at the top to follow us either on LinkedIn or to subscribe to our Vantage Point blog? For those of you who are not familiar with Onyx, just to do a quick introduction, Onyx is a consulting division of Expeditors, and we help clients build more efficient and resilient supply chains. We are uniquely positioned to help our clients in identifying geopolitical, regulatory, economic and operational disruptors, which can then be translated into a more forward-looking and reply change. All of these are done through advisory engagements and is. Projects are also tailored to client -- individual client needs, either as one-off projects or ongoing retainers. These are Onyx' services lines. In a nutshell, our service offerings cover various areas within the supply chain, like planning and strategy, trading compliance, sourcing and manufacturing. Please contact us if you have a project or a need where our advisory expertise can assist. So on to our speakers today, I'm excited to introduce the speakers who will be presenting today, myself, Adam Karson and Suryo Nugroho. I'm a senior geopolitical analysts and Onyx, and I work primarily on the Indo-Pacific. I hold honors degrees and history of international political economy, and my work has been featured in the Pacific Forum, the Asia Times and other media outlets. Suryo Nugroho is a seasoned policy expert with 14 years of experience across geopolitics, policy analysis and supply chain management. He currently serves as Onyx' senior geopolitical analysts leading the firm's Southeast Asia coverage. And on to Adam Karson, Adam has more than 20 years of experience as an economic adviser to global leaders across a range of industries. He has extensive experience in the U.S., Europe and Middle East. Adam most recently worked at Chevron as a senior economist and is responsible for Onyx' macroeconomic analysis and forecasting. So with that, I think we will cover 1 more slide on our content today in the webinar. So what we will cover in the next 45 minutes or so with 15 minutes for Q&A is the current state where the AI build-out stands in terms of capital, supply and delivery. We'll cover hyperscaler CapEx plans through 2030. We'll look at data center power demand as it relates to infrastructure and then the supply chain and critical minerals for the supply chain. Adam will cover our possible future trajectories, bring you through three scenarios of AI development moving forward, a fast-build scenario, a slow-built scenario and a pullback. And lastly, he'll touch on what it means for freight and what to watch. So with that, I will hand it over to Suryo to kick us off. Thanks, Suryo.

Thanks a lot, Olivia, for the kind introduction. So good morning, everyone. So we will -- as Olivia has mentioned before, so we will start by basically providing you with our analysis as well as insights on the current state of AI development, right? So where are we going with AI? What's the state of play of AI currently, right? So there are four elements that we want to cover here. First one is about capital, right? Capital is readily available around big 5 data AI CapEx projected to exceed USD 1 trillion by 2029. And then the revenue scaling as this AI companies have already started to also offer like enterprise AI functions, right? So it's the revenue for them is scaling up as well. And in terms of economic impact, the economic impact is significant especially for the U.S. economy. Now AI contributes around 1.5% to 2% of the U.S. gross domestic products as well as 50% of the growth net imports, right? And supply chains are heavily concentrated, and it benefits a handful of countries. We are seeing -- we're talking about Taiwan, South Korea, China and Mexico, carry the value chain. And then -- but there is one thing or one risk that need -- you all need to monitor, right? So we will discuss this in the second part, and Adam will talk about this later on in more in depth. So delivery is lagging from two different sides. The first one is on the power supply or energy supply construction. So right now, it's only about 5 gigawatts of the roughly like 16 gigawatts of U.S. capacity announced for 2026 as well as we are talking about components, right, and equipment as well. So there are a couple of bottlenecks emerging in the AI supply chain, namely high-bandwidth memory, chips, bottlenecks as well. And then we are also seeing that packaging is also experienced -- starting to experience a bottleneck as well. So moving on to -- I'll talk a bit more detail about the CapEx plans, right, the first box. So hyperscaler CapEx plans will grow by about 28% per annum from 2025 to 2030 after like a tremendous growth of CapEx from 2020 to 2025, around 114%. This -- most of the CapEx will go to inferencing. So why why inferencing is so important here because right now, AI is at the pace of implementation. So now more and more people are using AI. And then inferencing is the cost of inferencing is really high, right? We're talking about because of the scale. So inferencing happens millions or billions of time depending on the use. And because of that, they need more data center. Hence, you can see the figure on the top that data center -- CapEx on data center is also increasing tremendously, right? Because they need data center, more and more data center to do inferencing, right? And then to do like a better inferencing we're talking about speed as well. So instantaneous responses, which require advanced and basically power-hungry GPU. So there more investments are needed to basically acquire a more advance a GPU for doing a better inferencing, right? And then complexity as the model becomes more advanced and can solve complex task, it requires more computational resources, which led to more intense data centers, what we said before, need to be built. And with that, that comes with the rising cost of energy as well as the wear and tear of the hardware, in this case, the GPU. So the -- you see that the CapEx plan is really high, right? So it's tremendous. So for that for -- basically, for this -- the AI companies to justify the CapEx plans, they need to get high margins as well as sustained rapid growth, right? So we anonymize the company here, but this company A, B, C and so on and so forth represents the top AI firms, globally. So when you talk about -- I think the most important graph here is the one on the right side, right? So we're talking about the revenue growth required to break even and NPV. So there's one firm more specifically company B that relies on frugality to generate mastic ROI, right? So company B, basically, they need a smaller revenue -- lower revenue growth to basically make a breakeven -- for the breakeven point -- to reach breakeven point, right, meaning that they spend more efficiently compared to the others. So it's a different strategy, right? And the other strategy is basically company A, C and D, they require -- they rely more on the strong revenue streams, but they spend more on the infrastructure. They spend more in the data centers, and that's why they required a stronger revenue growth in order for them to justify the CapEx. So moving on to the AI supply chain mapping, right? So as I've already said before, the AI supply chain is heavily concentrated in a handful of countries, more specifically here in 5 countries. So number one -- the first one is Taiwan. So we're talking about chips, the leading edge logic and advanced packaging. So 90% is -- 90% of the sub-5-nanometer of logic output is produced in Taiwan by 1 company, TSMC. And in Netherlands because Netherlands -- ASML is headquartered in Netherlands, and they are the only supplier of EUV lithography at the moment. So yes, it's -- for the EUV lithography is pretty much concentrated in this one company right? In Japan, about 50% of global silicon wafers is produced in Japan. We're talking about photoresist and sub film as well. In South Korea, we're talking about high bandwidth memory. So there are only three manufacturers -- three companies that manufacture high-bandwidth memory, right? One based in the U.S., Microtechnology and the other two SKHynix and Samsung are based in Korea or a Korean company. So clearly, the South Korea is -- I mean, really the high-bandwidth memory is pretty much concentrated in South Korea. And I think we cannot leave China out of it because China basically dominates of basically Domino's critical minerals refinery. So about 99% of primary gallium refining is done in China. Also a couple of different rates as well, talking about germanium, tungsten. And Olivia will talk about this about the potential export control because the Chinese government is actively basically developing measures or -- yes, measures to basically control the export of this refined critical minerals for gaining geopolitical advantage over the U.S.

Absolutely. I think Suryo has taken us to, I think, a wonderful overview of the AI industry. And the next three couple of slides, I think, as we finish out the context setting before we move on to the scenarios that Adam will take us through. Here, we're really looking at infrastructure and specifically power demand. So a headline for us today is that data centers power demand are projected to surpass heavy industry by around 2030. And that demand remains the highest in the U.S., China and Europe, while Southeast Asia more than doubles as well by this time frame. So when we think about it from an infrastructure perspective, for Southeast Asia, in particular, that data center power demand will be driven by hubs in Singapore and Southern Malaysia, which makes assessing, I think, the country capabilities of each of these regions to provide things like reliable power, affordable power, it becomes much more critical. And despite the fact that data center energy demand absolute growth is much smaller, they tend to cluster geographically, which makes grid integrated than other sectors like industry electric transport or appliances. And I think as we move on to critical minerals, what we've done for you here as well is to summarize a list of cortical minerals most exposed to export controls. So as we know, critical minerals are a pretty key upstream component of AI supply chains. And when we think about the geopolitics of the longevity of the AI build-out, critical minerals are a big part of this. As Suryo mentioned, China is very dominant in refining, and it prefers to use these upstream inputs and trade confrontation. So we've listed out here, I think, a couple of minerals that are most exposed. Something that I really would like to highlight for us here today is that -- last year in 2025 in October, China kind of put in a series of export controls on critical minerals. Those are expected -- that pause is expected to expire in the next two months, which makes the upcoming Trump-Xi Summit very critical. -- as we will see both governments try to reach, I think, an extension of that pause and to prevent some of these expert controls from coming back into place. So a key milestone, I would say, in the next two months. That being said, critical minerals are -- there are workarounds that exist for mining and price coordination, but diverse China in refining is a long-term process. So we've identified a few workarounds at the supplier and government level. We offer and, I think, tell our clients that you can work with suppliers to stores re-refined inputs from partner nations, conduct audits on geographic origin of your raw materials, coordinate on the minimum percentage of "Western refined inputs," if that is a strategic requirement. Of course, license monitoring is a big part of your strategy here, really confirming the status and the speed of export license approvals. I think a lot of nations are also being quite active in mineral alliances and stockpiles. Specifically in the U.S., you have project Vault strategic reserves and members will essentially subsidize the difference from the preset minimum price if China tries to crash the price of germanium or another commodity crowd-out new miners. So these alliances and stockpiles definitely exist as alternatives, but refining is still a charcoal that will take time to work [indiscernible]. A lot of that are currently underway, like the Canadian Ohio pipeline, Vietnam's processing plants, a project in France, all of these will only kind of kick in, in the next couple of years and will take time to ramp up. So refining will still be a to charcoal in the next few years. So I think just to round off this section, I think we really want to maximize the amount of time available for the scenarios, which is really the key part here. For this context, we've kind of gone through an overview of the AI industry, its financial viability and then as well as some of the infrastructure and geopolitical constraints of the growth of this industry. So with that, I'll pass it over to Adam, and he will take us through the scenarios.

All right. Thanks, Olivia. Thanks, Suryo. So I thought was a really good kind of background level setting of kind of where we are in the AI growth cycle, and how the ecosystem is global and complex, which leads to some pretty interesting and equally complex scenarios that we need to think through and how they impact the logistics market, even the broad economy, but specifically how they impact air, ocean and trucking. So what I'd like to start with is kind of when we think about scenarios, I think it's really useful to think about who the main actors are, what are the decisions that need to be made, and how do those kind of factors drive which path we're on? And if we take kind of a step back then and just think about the environment, I think there are three groups that really determine the path forward here. It's their customers, investors and suppliers. And the first two customers and investors are really pushing the acceleration here where you have a urea mapped out this rapid increase in the infrastructure build-out and the investor money flowing in to support those ambitions you have customers on the downstream side of that, consuming a lot of that AI bandwidth, consuming tokens, not just everything that we're doing individually hacking away at LLM all day, but also some enterprise solutions popping up, and that's starting to scale. If you look at Anthropics revenue over the past year, it's like gone up 10x or something in that kind of order of magnitude. So those two kind of actors are really full throttle pursuing the opportunities here. Then you have the third actor suppliers, and this is everything from power to chips and equipment everything that goes into the building into the data center buildings and supports that ecosystem. And this is where things are struggling a little bit, as Suryo mentioned, kind of, in the first couple of slides, where we're not quite keeping up. Now it's not doom that it's not that this is a major obstacle today or necessarily kind of holding out the build-out today, but this is something that we really need to pay attention to because it's -- we're probably the most immediate risks fall in terms of what pace the ecosystem can be built out. So I just kind of want to frame that because keep those factors in mind as we talk about the scenarios and kind of where the risks and opportunities may lie. So we came up with three scenarios that really turn on revenue and delivery. And I'll explain what I mean by that. But those are the two things that really determine how those three actors kind of evolve. The three centers we came up with our best case, which is a fast build. So even an acceleration from where we are today. And this means that customers are realizing accelerated productivity gains, driving a lot more revenue growth. So the 10x type growth we've seen over the past year, that continues. And that just becomes a flywheel. As that revenue grows, the hyperscalers build more, the investors put more money in there, achieving higher return on investment. And a key here really is that in the best case scenario, it would rely on suppliers innovating past their bottlenecks, which I think is a pretty fair assumption actually with this much money and capital flowing into something and if the prize is really as big as some people think it is, then the problems that need to be solved on the supply side probably aren't that complicated. They just require some dedication and some capital. So really, this best case is that kind of everything kind of comes together you get that, that flywheel as I said. One caveat here where it's paying attention to in the best case is the full impact really depends on the labor market outcome. And that's a whole other kind of 1-hour discussion we could have on how AI is going to impact the labor market. I'll kind of maybe touch on that a little bit as we go on. But I just want to highlight that as a key caveat for that best case scenario. In terms of mechanisms for this best case, I think some key things worth highlighting are that. This depends on things like behind the meter generation. So if we're going to rely on public utilities, building out massive power generation and grid, I mean that's going to -- that would take a decade or more, right? So this requires things like innovation behind the meter generation. It requires CapEx rates to remain very high, but shift over time and get a little more creative on kind of how much can be prefabricated instead of kind of build on-site construction. And also things like the siting of data centers follows the power, not necessarily the demand. So in the U.S., we're seeing certain states kind of put some restrictions up around what data centers can be built and whether or not contingencies on whether they have an impact on the grid. So I think you'll see more and more data centers kind of move to geographies that have existing excess capacity on the power side. So moving across the screen here so that's the best case. Base case is a slower build and kind of a plateau or even a bit of a slowdown from where we are. Thinking like 2026, 2027 is probably sort of the peak rate at which we can build out the infrastructure. So in this case, there are a couple of conditions, right? So customers are slower to achieve these scalable productivity gains. So there's some -- there's certainly some return on investment, but it's -- the capital comes at a higher cost and a lower ROI than the best case. And suppliers are more in a management mode. They're managing constraints that don't really get ahead of them. And so therefore, the buildout kind of plateaus or slows in the next year or the key mechanisms here that construction schedules are kind of slipping. We're seeing these like longer and longer lead times for some of the key infrastructure components and that just kind of becomes the norm. CapEx growth rate certainly decelerate from the super rapid growth we've seen over the past couple of years. And then back to the constraints, I think here, we would see the constraints evolve, right? Like right now, you're seeing tightness for memory and the prices are shooting up, then probably now also for electrical equipment. Next, it might move to the grid. So you'd see kind of this whack-a-mole approach to trying to deal with these constraints. Then you have the worst case scenario, which actually comes in kind of two forms, you could have either a demand-side shock or a supply side shock. And this is just where the economics of the system kind of break down. And so for example, on the demand side, you could see a situation where the economics for the consumer just don't make a lot of sense. And they're not -- because they're not achieving productivity gains. So they pull back on some of their AI spending or vice versa, the economics for the big developers, the Frontier kind of models don't work either. I think it's very plausible where there's a situation that in order to achieve scalable productivity gains, sort of at an enterprise level, you don't necessarily need the frontier models and pay that premium for that -- for those models. And actually, the models that are more like the fast followers that have a very clear business case and are solving very kind of very discrete problems. Those are the ones that kind of build up scale and -- but there's not a really strong revenue model behind those because they're more commoditized. So in that kind of environment, you would see a much lower return on investment, investors kind of pulling funding or maybe even facing some losses on some other bets. So that's kind of the demand side. On the flip side, you could have a supply side problem. Olivia touched on some of the constraints the policy-driven constraints that we might see in kind of the AI ecosystem going forward, specifically around critical minerals. Now this is one where maybe the models are working well. Productivity is going -- is scaling up, but there just isn't enough supply capacity to keep driving out the infrastructure investment and the cost of those materials becomes prohibitively expensive. So the mechanisms here are that to pay attention to are the revenue -- either revenue falters and/or there's some supply side issue that just finally gives way. And then the financing becomes tested. And then when kind of the tide rolls out, you have some very expensive assets with long lives that don't match the debt that was needed to finance it. And then also to watch out for the critical mineral export controls, whether those were churn or not could be a big turning point. So those are the three scenarios. Now what do we make of them? How do we think about the impacts here? And I'm trying to break this down into two major buckets, the economic impacts and then the logistics and supply chain impacts. And let me preface this with saying these are the direct sort of first order impacts, okay? So specifically related to the scale and pace of capital expenditures. Later on, I have a slide on sort of how this sort of multiplies or kind of multiplies the indirect impacts across the economy. So for example, just a kind of foreshadow that. In the worst-case scenario, if you were to have a collapse in capital expenditures, you would also have other parts of the economy kind of falling as well. You have like these multiplier effects. Stay with the best case scenario, you get that flywheel effect, you're going to have all kinds of impacts on the labor market, revenue growth, et cetera. So you have a lot of indirect impacts as well. But again, let's just focus here on the first order impacts, what happens to the economy and to the logistics markets under each scenario. I won't read everything on here, but just to highlight a few things. First, if we compare the economic impacts and the best case scenario, curious where you get the productivity gains really accelerating, and you get a very material boost to U.S. GDP growth for like the next decade, upwards of -- I think a conservative estimate would be, say, 0.5 percentage point or 0.6 percentage points per year above baseline growth for the next decade. So that's -- that may not sound like a lot, but if you compound that over 10 years, that's a pretty big jump in the size of the U.S. economy. Again, the cat share is what happens with the labor market depending on what those indirect impacts are, you could have something above or below that 0.6. The base case is our baseline view, right? So here, we're looking at GDP growth in the low 2s over the next decade. I think one thing to pay attention to in the base case is that inflation increases before output. So go back to the mechanisms of this scenario where you have suppliers dealing kind of triaging constraints as they come along, which means you're going to have kind of continued waves of inflationary pressure, like kind of what we've seen over the past year or two. And then in the worst case scenario, whether you have a demand shock or supply shock, you kind of get different outcomes. The demand shock, I think, is probably -- I would argue the more likely, and here's where you get, I think, much more negative impacts to the U.S. and global economy, where you're talking about potentially putting at risk the whole financial model that's kind of backing this endeavor. And if you get significant write-downs, equity repricing, here's where the direct impacts really multiply across the economy and you get -- you probably get a recession at the end of the day. And then for the worst-case scenario, this is really more of kind of an inflationary scenario in addition to marginally slower growth. So then if you look at that last row, what happens to logistics and supply chains. On -- in the best case scenario, this is really where freight kind of takes off, especially heavy oversized ocean cargo becomes a real growth engine because we have to build out the really heavy physical infrastructure on the power generation and equipment side. I think in the base case, the air market is really kind of thing to watch. That stays tight for the next year or two, but then may normalize, right? If we're essentially kind of plateauing on how fast we can build out the infrastructure, then the air cargo market balance kind of mirrors that over the next couple of years. One thing to pay attention there is project cargo. There's long lead times there. So that kind of cargo probably has a longer kind of a longer peak cycle over the next -- maybe to '28, '29. And then in the worst case scenarios, here's where you get volumes falling. Again, tracking the investment cycle, volumes and rates fall together commensurately across both the the kind of chip and technology side and then also the kind of power and infrastructure side. And so both air and ocean are hit in that scenario. Air is hit disproportionately. Let me move on to onto this slide. So kind of going into this in a little more detail, and this is kind of mirrors the last slide a little bit, but going into slicing it a couple of different ways to think about what's improving or getting worse in each scenario. Again, I won't read everything here. I just want to highlight a couple of things. And most importantly, I think power is the only input that in -- basically, it gets worse than that in every scenario. And that -- and really in the short run, power, we're short on supply. The best case scenario assumes that, that kind of catches up over the medium term, becomes more kind of a comfortable supply-demand imbalance over the long term. But in the base case, this is -- that's something that really is a stress point and creates those constraints that I've spoken about. I think another thing to kind of pay attention to is how sourcing changes. And by sourcing, I mean, diversification and whether or not we can kind of debottleneck or derisk some of the sourcing that is happening right now. And in the best case scenario, we assume that kind of naturally diversification kind of naturally happens as the ecosystem evolves and gets a little more innovative and creative on where we're sourcing from. But in the base case scenario and certainly in the worst-case scenarios, that diversification doesn't really happen much at all. And so we're kind of living with some of the inherent risks in the AI supply chain that we have today. In particular, I would focus on Taiwan as being the main source of chips. And then what I really wanted to get to here is kind of more direct impacts on the logistics market. And I'll just spend a minute on this slide. And as I mentioned, I think air has hit the hardest, right? So when we think about where the volumes are today, I think the numbers I've seen suggest that about, I think it was 7% or 8% of global air cargo is related to AI, and that's on a volume basis. On a value basis, it's something like 40%, 50%, some crazy big number. So when we think about how -- and then on the ocean side, it's probably less than 1%. North America domestic trucking, probably 1%, but kind of concentrated in particular segments of trucking. So there are some areas of concentration there. But when we think about how these scenarios impact logistics, air definitely hit the hardest. So the -- just because of the scale, right? So if you have a pullback in CapEx spending, you would have a commensurate decline in air freights and that 7% of volume become some significantly lower number. So these are just kind of directional what to think about, what would happen on the demand side growth for airfreight, ocean container freight, project cargo and heavy haul and then trucking, specifically North American cross-border and last-mile trucking. And obviously, we're in a kind of when you look across the board, we're in a relatively kind of tight market, certainly in air, ocean, trucking, not all of that is related to AI. There's -- we've held other webinars on the geopolitical factors, driving that market tightness. But in the best case scenario -- while it might be the best case scenario for sort of anyone playing in the AI ecosystem, not a best case scenario for anyone outside trying to ship things because it just means kind of tight market for the foreseeable future upward pressure on rates. Base case, you kind of get a milder version of that and potentially in the long term kind of a neutral impact. And by long term, I mean, say, 5 years out. And then in the worst case scenario, that's where volumes kind of really take a big hit and rates come down commensurately. Now one thing I want to kind of highlight, oops, there we go. How we calibrate those were obviously qualitative kind of measure, but we can calibrate this a little more precisely. And I think the dot-com era provides a useful comparison to calibrate the downside scenario. So if we look back to the telecom sector 2000 to 2003, spending -- CapEx spending fell about 80% from its peak and then took about a decade to soak up the fiber optic glut that we've build out. So if we just kind of take that as a very rough example, I think a conservative estimate then would be in a worst-case scenario, AI CapEx falls by, say, by half, by 50%. AI-related goods, as I said, we're about 7% of global air cargo volume in 2025. So a 50% CapEx reduction means that it goes from 7% market share down to 3.5% market share. So all else equal, air cargo volumes would fall 3.5%, okay? So in a one year, that's a significant hit, right? That creates some looseness in the market. And certainly, the transpacific lanes are much more exposed to U.S. high-tech air imports. So you would see a lot more loose on particular lanes. But just globally, 3.5% is something that -- a 3.5% decline is something that could be absorbed over a year or two. Like it's not necessarily sort of an existential issue for air cargo. And then when you look at ocean and trucking, when the volumes that we're seeing today are fairly negligible, 1% probably less than 1% of global volumes are related to AI on ocean. So this is not enough to move the market, really. Now I'll get to the indirect impacts on the next slide, which which probably would be big enough to move the market. But again, just the direct impacts, not a major shock to ocean. On cross-border trucking, you could see more of an impact there, and I think, in particular, on flatbed and heavy haul. So if you're kind of -- if you're using those services, that might be something that gets hit a little bit harder, and you could see some slackness in that segment of the market. Now I've talked about these -- I've talked about direct impacts and now the indirect impacts, I think, is where the real money is here. So this is where you get a compounding of both upside and downside. And so when we look at the best case, if you are getting this direct impact that you have some modest acceleration in CapEx in project cargo, air cargo, airfreight, et cetera. But then you get this flywheel effect that spills out across the rest of the economy, and you have the U.S. economy just consistently growing at 3% or higher. That obviously has spillover effects into other sectors of the economy, in particular, consumer spending, and then you get even more rapid growth. So I think you get spillover effects that compounds, not necessarily evenly, but you get much more kind of positive upside for ocean in that scenario as well. Base case is kind of what it is. Not -- there's not a whole lot there in terms of indirect impacts because we're kind of -- it kind of looks like the world does today. So you get some kind of positives and negatives positives on maybe the consumer side, some negatives because the AI CapEx plateaus maybe even shrink a little bit. So we're basically we kind of end up where we are today. But then on the worst-case scenario, I'm highlighting the demand side version of that worst case is that if you get a -- again, so you get that 50% reduction in CapEx. Right now, the current level and growth rate of AI CapEx is contributing somewhere between like a third and a half of U.S. GDP growth. Those are -- that sort of third-party range of forecast. My forecast is closer to 50%. So if the U.S. economy right today is growing at 2%, a full percentage point of that is coming from AI. So if you basically get rid of that, and the sector is not only flatlining, but actually declining, the direct impacts alone might put us in to a recession, certainly would probably flatline growth. But if that happens, you have potential -- you have equity -- the bear market -- bear equity market, you have bonds getting defaulted on, you have kind of ripple effects throughout, the investment community, the consumer sector, this is almost definitely a recession, probably worse than what we saw during the dot-com bubble. And if that's the case, then you have these compounding effects and impact not only to air cargo, but then to ocean cargo as well because retail sales are probably going to decline for a solid 2, 3, 4 quarters. So I just want to highlight here that like these indirect impacts really compounds, both on sort of the book ends of these scenarios. Now let me end up quickly with a couple of slides, then we can get to Q&A. So just a couple of things to watch over the next year or so. First, always pay attention to memory prices and transpacific air cargo and volume rates. I think those are sort of the canary in the coal lines of how fast things are moving and is the buildout kind of keeping pace or even accelerating going forward. But the next thing to really watch after that is what happens with Chinese export controls in November? Is that pause, extended or not? Then we want to look at the power generation order books at the year-end. This is a great forward-looking indicator of where -- whether things are staying hot or cooling down. And then when we get into February, we want to look at the fourth quarter plans for the Q4 2026 plans and then kind of 2027 construction starts. What does that pace of growth look like? And then finally, what I'll end you with here is kind of two sets of conclusions depending on where you sit. Basically, are you in the AI value chain or not? If you are, things to kind of pay attention to are the wait time for power equipment is not going to get shorter. We think, for the time being, chips and packaging stay in Taiwan, I think medium to long term, that we could derisk that as we're kind of building out the North American capacity. But not quite yet. Mexico stays as a strong U.S. assembly hub and that sort of cross-border lanes or something to really pay attention to. And then there's a lot of talk about racks getting kind of more dense, heavier, hotter. And so the -- it requires more power and cooling freight per server. So there's just more demand for power and cooling. If you're not an AI value chain, the bottom line here is you're competing with AI for the same capacity, right? So air and flatbed and cross-border trucking in particular are going to remain hot in most scenarios. AI CapEx is really kind of setting the rate. That's kind of that's the marginal good being moved right now, and that's what sets the rate and air spot rates are up 38%. Not all of that is AI. Obviously, there's the geopolitical stuff going on in the Middle East, but that demand just kind of keeps that upward pressure. And then the power and component costs, although you don't have -- you're not may not have direct exposure to that, that ends up coming -- end up feeling that no matter what. So electricity prices are up, memory prices are up. So say you're in the auto sector, looking for chips. I mean these are -- the cost pressures are only pointing up to some of these key inputs. And then just pay attention if there's a pullback. So if we're in a downside scenario, that's really kind of the main driver of rates coming down. But the flip side of that is that remember those indirect impacts. If we fall into that worst-case scenario, you're -- depending on where you sit in the economy, you may get kind of rolled up in those indirect impacts as well. Okay. So I went a few minutes over. We have about 10 minutes for Q&A. So I will stop sharing, and we can go to Q&A.

Absolutely. We've got a long list of questions here. And first of all, thanks Adam, that was a wonderful set of scenarios. And I'd like to kick off a question here in the Q&A box. I'm going to take a little bit of liberty here in rephrasing it. But whatever you assume that the AI infrastructure build-out is a government-backed adventure, right? And it's not strictly based on market dynamics, but rather on strategic capability. How would this kind of affect the stability or the time frame of the AI infrastructure build-out? I think either for Adam or for Suryo?

I love that question. I'll take a stab at it. This is something I've been thinking about a little bit as we prepared for this weather. I think I think there's a strong case that you could make that parts of the AI ecosystem are -- could be treated like a public utility. One hypothesis would be, this is not a recommendation. It's just a hypothesis. Is that frontier models could be treated as a public good and funded -- and just have a completely different funding mechanism with a lot of taxpayer dollars. And there's a ton of precedent for something like that. Think of like how NASA operates, right. You go to the moon, all that's publicly financed, but all the benefits that spill out of that, all the technologies that are created as you kind of you embark on projects like that. So I think you could make a very good argument that the Frontier should be at least partially public financed. And that would also kind of open the door for some different tax models. So how you prepare for -- how we prepare for labor market disruptions kind of like the corollary to that. So imagine that world where that kind of financing mechanism exists. I think that would add a ton of stability to the AI infrastructure build-outs and alleviate some of that pressure on the margin and allow more of a focus on really the enterprise scalability, fast follower models that are much less expensive to operate and consume. And I think you have a very -- so you kind of have a bifurcated view of the market that way. And I think it's much -- in my mind, it's easier for investors to have a clear return on investment if they're not having to participate on the frontier all the time. So I think that's really -- I love that question, and I think it's a kind of a whole field study that should -- is kind of emerging right now. Sorry, I don't know if I made any sense at all, I answered, but.

Yes, I think you made a really good point here, Adam. And a couple of countries have already started that, in fact, like the U.K., for example, is -- I mean, you mark about GBP 1.1 billion for developing national AI supercomputer, right? So this is to expand national computing capacity by 2030. So that's a clear indication there that state has already started embarking on this journey, right? And then we are also seeing a couple of different countries. Canada, France, India and the Gulf countries, like the UAE, Saudi Arabia, they have already started embarking on AI infrastructure as well, right? And a couple of them, they have really started also public private funding is basically to reduce the risk of private investment in Frontier sector as you say it, Adam. So yes, I think it's getting there. I think the traction is moving forward towards that direction. I can totally say that.