Logistics for the AI Era: How to Move the World's Most Critical Infrastructure

Hello everybody. Thank you for attending today's webinar, logistics for the AI era, how to move the world's most critical infrastructure. My name is Konstantina Georgaki and I'm the general manager here at Flexport leading the Northwest. We have a great webinar for you today and before we begin, I have just a few housekeeping items. So up to the right, you'll see uh on your screen a sidebar on the right of the main stage with a couple of things. Uh one is a chat where my colleague Carol will be dropping some useful uh information for you. Next to it, there is a document tab where we uh will share the document of this presentation and other useful resources like Flexport's tariff simulator and tariff calculator. And then you'll see on the right-hand side Q& A where you can drop your questions and uh we will manage the pace today so we can answer lots of them and if we don't get to all of them, we'll be sure to answer via email. One more callout, you'll see a button above the slides here that says supply chain analysis. If you click on it and you have to click on it twice, uh you will fire off a request to the Flexport team to request a supply chain analysis and during this presentation, I will walk you through uh what this entails. All right. One more housekeeping note, a legal note, the contents of this webinar are made for general information. They are not guaranteed to fit your specific circumstance. So we always advise our audience to reach out directly to discuss your individual business needs. Now, let's dive in. Joining me today is my colleague Alexis Botte, our vice president of and global head of air freight here at Flexport. I am particularly excited for this webinar because I live here in uh San Francisco and walking to the office today I saw about a dozen AI billboards. So, it's really palpable what's going on. You can feel the energy. And Alex and I are very excited today to peel back the curtain a little bit and show you what we're seeing on the ground and hopefully share some very useful content with you. So, here's how we're going to spend the next 45 minutes. First, I'm going to walk you through some of the key numbers and statistics of the AI infrastructure supercycle and translate it into freight. From there, I'll speak a little bit about supply chain challenges and opportunities specifically for shippers and what we're seeing in the market today. And next, Alexi will take the lead and take us deep in the logistics playbook. What to look for your logistics provider, how to move your supply chain, and go deep in the different levers you have at your disposal. From there, I'll take the mic back and walk us through cost mitigation strategies, how to look at your supply chain as a portfolio. And as you can see, I'm speaking really fast because we have a lot of content and we want to leave good time for Q& A. All right. Let's dive in. The AI infrastructure supercycle. We are now seeing the largest infrastructure buildout in the United States history and probably globally. The interesting comparison here, if you think about from the past, the largest infrastructure buildout previously was the United States Interstate Highway System. In today's dollars, that was 635 billions. And that was deployed over 36 years. Now, we are looking at an investment on AI and digital infrastructure that at the most conservative estimate would be at least 3 trillion. Some analysts even put it at north of 5 trillion. Half of that will be deployed right here in the United States, and it will be deployed in the next 5 years. So, that's a very different order of magnitude and very different playbook on how quickly this will unfold. Now, let's look at some of the numbers underneath that. About 600 billion of these most conservative 3 trillion would be capital expenditure by the large hyperscalers. — [snorts] — The 600 billion is just in 2026. It's up 36% from the year before, and if we take all of that investment combined, we're going to reach globally about 200 gigawatts of global capacity. That will be double of what we have today. Today, most of this capacity is deployed for AI training. For all intents and purposes, we're in the early innings. You'll see the models getting better every day. So, a lot of the data center capacity is still dedicated to training the models. In the next 5 years, we expect that all of this will

unfold really quickly. And by that time in 2030, most of the capacity will be dedicated to inference, so actually running the models. Now, let's start translating what this means in terms of freight. What we saw in 2025 is that data centers accounted for about 1. 4 million kilograms of air freight, and that's a conservative estimate. And if you start squinting a little bit on the graph to the left, the biggest exporting countries, unsurprisingly, Taiwan and China. Now, let's see what it means in terms of growth. So, AI infrastructure was about 15% of the air demand growth in 2025. Only second to e-commerce still riding from the previous boom. But if you look at the number here, it's been growing 49%. So, if we project that out in the next couple of years, it will most likely supersede e-commerce as the biggest driver of air over growth demand. So, that's significant. So, um because that is concentrated in a few corridors, that has created crunch in some on direct capacity in some of the AI highways. So, namely Taiwan to the US, China to the US, Vietnam India as an intra-Asia lane, and Thailand to the US. So, we're seeing some crunch on the direct capacity on these AI lanes, and particularly Taiwan and Vietnam have an extra layer of infrastructure challenge as maturing that capacity. And some measure of geopolitical volatility. So, what a shipper should do in this with environment? So, one is explore creative rulings all the way from very fast options all the way to deferred options because the market is really expensive right now in these highways. And then for time-sensitive shipments, so if you have like a project that needs completion, or line down situation, essentially build a very really resilient car speed mix, have many airlines on hand, and tap into fixed space commitments. With that, I'm going to call my colleague Alexi on stage to walk us through what it takes to execute on the AI in the supply chain. Good morning, everyone. So, Alexi Bouchet here, VP and global head of air freight at Flexport. So, in this section, we are going to discuss first what component typically move for AI company and the typical logistics challenge they face. Then we'll go through execution best practice. So, I will deep dive on how to execute for AI supply chain. And lastly, moving this infrastructure can be very costly. So, we'll explore what are the cost mitigation level that you can use. And throughout this presentation, I will do a mix of best practices and I will share real case study from our customer to make it a bit more lively and not be too theoretical. So, let's now look at the typical AI supply chain. So, I'm going to do an oversimplification, but just to anchor with everyone what is an AI supply chain. So, first consideration is there's a wide variety of components of different value moving on the supply chain from hundreds of suppliers, sometimes thousands. So, it can be any of the subcomponents in that picture all the way up to the full server rack that you see that you can see. And you have company that move a lot of these components, almost all of them all the way up to server rack, and you have company that only specialized in subcomponent, for example, chips or cooling system. For the sake of this presentation, I will assume a company of the highest complexity. So, that's a that has a broad array of component from low value component all the way up to server rack. That way we will explore all the scenario and how to execute for different type of company. Second important thing when we talk about an AI supply chain is to understand the typical two modes they use. So, they would use either a mode which is an inventory transfer to regional distribution center and then they distribute directly to client. But, sometimes you would ship directly to a project site. So, that's called a direct shipment. And these direct shipments then have higher visibility need because now suddenly you need to coordinate with a site manager or potentially someone from the client of your AI client. Uh and we will dive into these visibility

challenges. Last little thing, uh the very high value components for AI company typically have shock watches, tilt watches, or tracking device. So, you need to move the physical supply chain of these goods, but you need also to be able to cater for all the signals that come with these devices. So, you need to be able to see when a shock watch has been activated, when a tilt record that, and handle this exception. All right. So, these are the like high-level picture of the supply chain. So, now let's talk about the logistics challenges now moving these AI components. So, very high-level again, there's three type of challenges we need to address. First thing is that this is these are extremely high value and fragile components. To give an idea, a modern AI cluster can cost up to $1 million per rack. They are extremely sensitive to vibration, and a single road bump could just unseat a GPU. Second element is that these are large, heavy crates, and there's a lot of dangerous goods. So, it really And the dangerous good, why? Because there's battery packs, there's liquid in the cooling system, so wide array of dangerous goods. So, it really require deep expertise on all your operation step. For example, your trucker actually need to know how to segregate the dangerous good so that he loads the dangerous good properly and he can also flag that hey, this DG actually is only allowed for ocean freight so it shouldn't have been added to this air freight shipment. So, deep expertise for all actor in the chain, the trucker, your own team at a logistic provider is really required to move this good. Um and the third thing it requires extreme speed and reliability. So, in 2025, 57% of data center project had a delay and a large part of this was due to logistic challenges. So, you if you step in as a logistic provider, you need to help with this challenge. And when you do a direct shipment that will go directly to a site manager, it's very important that you have not only speed but precision. It needs to arrive at the right time, not before, not after and everyone needs to have signals that hey, it's arriving on that date and you need to keep everyone coordinated. So, these are the high level challenge and why it makes AI logistic fundamentally different risk category to general freight. All right, now let's discuss a little bit the playbook to deliver good execution for AI company. So, I have five 50 No, I have 15 slide of content with detailed pointer on execution. So, my intent through this webinar is really to if you're a logistic manager to give you a checklist of what you should look for in term of assessing new provider or evaluating your provider or just assessing how you are doing in your current supply chain. Uh, so that's the that's that should be easy and then giving you good tips to source your provider. If you're not operating an AI supply chain, there's a lot of this checklist that you can still reuse for your most critical component. So I'll discuss that. The six pillar of execution on AI infrastructure and then I will also discuss cost mitigation strategy. How do you make sure that you execute on an extremely high value supply chain but don't overpay for it. So what are the thing you should look at? So let me go through six pillar of execution. So first you need to build a fast and resilient air network. Second, you need to have the lowest amount of handling throughout the chain and lowest amount of end up so that you keep your cargo secure and moving fast. Third pillar is you need to have very specialized tracking expertise and my advice is to have a dedicated tracking team serving a client so that they build reps on your client. Third key pillar of execution, you need to provide live visibility for all party. If you're delivering into a site, then it's very important that you are able to provide visibility to the site. To put it very plainly, the typical can excel reporting is not good enough when you're delivering on a

an AI supply chain. Fifth pillar of execution, you need to be able to do live exception recording and resolution and the last pillar of execution, you need to provide fast highly compliant clearances and low inspection. I have slightly detailed here of this. We're going to go through this together. On the cost side, Uh, don't want to overpay. So you need to be able to segregate components and really have different execution paths with your forwarder for low versus high value components. And this is critical so that you don't end up like putting everything on the costly highways, basically. Second thing you need to look at, in particular if you have the wide variety of uh supplier coming from the same country, is you need to look at your consolidation program. Are you doing multi-supplier consolidation for air? Are you doing buyer consolidation for ocean? Or are you leaving money on the table here? And uh the last thing is you need to constantly evaluate uh your transit time trade-off. Uh with the Middle East crisis right now, for example, all the Middle East uh transit uh was blocked. Uh prices have gone significantly up from uh Southeast Asia going into Europe. Maybe it's a good uh opportunity to explore sea-air via US to Europe, for example. So, you need to be dynamic on uh your uh your supply chain and be able to re-evaluate the option when on how the market evolved and do that with your forwarder. All right. So, these are the general principle. Now, we're going to deep dive on every single of the point on this slide uh with uh detailed explanation. Let's go. All right. So, first, how do you build a fast and resilient air network? Here are the things you should look at when you assess your provider or your current seller. So, you need to have daily guaranteed uplift on allotment, block space, or charter on the core lane where you move the majority of your volume. The second thing that's very important is to have an integrated demand planning process with your forwarder. For example, uh if you are shipping as a full uh server racks, it's pretty important that you try to cut your PO at four racks because four racks uh is what you can load on an airline ULD all together. So, it's one full unit of freight that's going to move together as a single unit. And it's provide you the best agility in your supply chain. And also, typically, your trucker has 2 million of insurance. Four racks were each rack is 300 K to 500 K. It's going to move all in one truck. So, that So, this type of consideration with your forwarder are very important. Like, you really plan your PO to get the maximum speed and execution through your supply chain. Third point, you want to have a multi-airline procurement strategy. So, you need to ask your forwarder, "Hey, how are you going to move my freight? " Because what you want to do is to isolate from geopolitical or weather events. If suddenly there's a weather event in Anchorage and all the freight going to North America via Anchorage is blocked in Anchorage, you don't want that to stop your supply chain. So, you need to have an outlet via another corridor to be able to move your freight. Or if there's a crisis in the Middle East and you were reliant solely on Middle East airline to move your freight, that is not good. You So, you need to assess the resiliency of your portfolio to this type of event. Fourth thing that is very important, you need to have to assess that your provider has access to large freighter capacity. So, if you look at the typical server racks, they used to be around 8 ft, so it was barely fitting in a 777 aircraft. Now, what I'm seeing is more and more very big crates that can be 9 ft tall. Unfortunately, 9 ft tall, then it's going to have to go on a 747 aircraft. That's a subset of the freighter aircraft that exist in the market. And if you have a provider that do not have access to large capacity on this aircraft type, you will be stuck. You will be stuck because you're not going to be able to move this large freight, or you're going to have to re-cut your PO to adjust to to what is there. So, that's one thing to look at. And of course, when you have rushes, you need to have a provider that has priority access to extra charter capacity. So, that is that has deep relationship with the charter operator uh company like Atlas Air, for example, or National, uh and that can like really ramp up capacity for you in a heartbeat to deliver on your supply. So, this is how you build a resilient network. Like, these are the key principles. Now, I'm going to deep dive into a very specific example of a client

to just make it very real about how you do that in practice. Uh so, this client is one of the world's largest uh company by market cap. They are leader in networking solution and AI infrastructure. So, this is to um to let you imagine the type of client that would be. They ship uh from Hanoi in Vietnam, and they have their dis- a regional distribution center in Guadalajara in Mexico. And this is the solution we put uh together for them to build resiliency. So, what you see is we have So, that's at the bottom of my uh little drawing. We have direct shipment going from Hanoi uh to Mexico uh using Qatar Airways and Cargolux. Uh so, that's one execution chain. We also uh and so, Cargolux and Qatar Airways have freight freighter aircraft. Uh we also have a second execution chain, which is Hanoi uh to LA, uh and it's moving on uh Nippon Cargo Airline KZ and Cathay Pacific 747 uh, aircraft, but also on UA that only has passenger aircraft for uh, the components that are not the racks. It goes to LA bonded truck to Guadalajara. So, that's the second execution chain. And we have a third execution chain which is going from Hanoi to Shanghai on uh, the airline CK and China Eastern and then goes on our own uh, freighter Flexport freighter. So, we run our own plane from TVG to uh, to LA. And it's a 747 aircraft. — [snorts] — So, this is how the chain were set up. And we also have a truck going to PVG. So, in case for example, there's fuel shortages in Hanoi, jet fuel shortage in Hanoi, we can still truck the cargo all the way to PVG. So, now let me talk about a real example in the Middle East crisis. Okay, so, we are in March of this year and we start to have tension from Iran. The airspace is blocked over Middle East. So, suddenly Cargolux and Qatar Airways, they can't operate anymore. Well, we still add our Hanoi to LA solution plus trucking to Guadalajara. And then we started to have jet fuel shortage in Hanoi as a threat. It has not yet fully materialized on international cargo, but it's a threat. Even if that would be to happen and we could not refuel in Hanoi in Vietnam, I could still truck all the way to Shanghai and move it to my charter. So, the chain of execution would still remain uninterrupted. So, that's what I mean by building a resilient chain of execution. Make sure that no matter what happened in the world, you have uh, multiple options to route your freight. And for that particular client, despite the crisis, we maintained 98% plus on time performance. So, that's a little example on really how you build your chain to for resiliency. Let me move to uh the second pillar of execution. So, you want to have the lowest amount of handling uh for speed and safety. I mean, you don't You want to make it very simple, the least amount of forklifting that will trigger the shock watch or potentially damage the cargo is the best. Um So, what you're going to do, you're going to do only built-up unit because that BUP in the forwarder lingo uh from the forwarder warehouse at origin to forwarder air warehouse at destination. There is no loose handling at the airport terminal. Uh at origin, you want to have a warehouse that is on airport so that you don't have to truck the cargo from a warehouse at the airport to the airport uh with late delivery cutoff in this warehouse at origin so that the cargo literally doesn't sit at the warehouse. You know, it gets delivered uh boom boom. So, like four crates are built on the PMC and it goes straight to an airplane. At destination, you would want to have runway pickup uh where possible. Not all airports allow this, but where possible, you want to make sure that the cargo uh as you've seen this picture is unloaded from the plane directly into a truck and go straight to the forwarder facility so it doesn't go through an airport terminal. If that's not an option, at least you want to have what is called a dedicated bypass door. So, basically, it gets from the plane into a little tractor and a tractor go uh put it on a belt that goes straight to your truck. It's second best option uh in that case. And you want to have dedicated trucking for pickup and delivery to not mix this cargo with other cargo. That's typically how you would handle the supply chain. Okay. This is uh an example of our setup in Hong Kong. I'll go very quickly through that. So

for example, in Hong Kong, you can deliver cargo all the way up to midnight. So, if you think about it, the supplier would finish the production of the day, get it to a truck, ship it to our So, they produce in the case of Hong Kong in South China, ship it to a truck, deliver before midnight at our warehouse, and it would go on the early morning flight straight up into the United States. Our warehouse is straight in the terminal. So, there's no double-checking or document by the forwarder at the warehouse and the terminal. It's one entity. It's Actol in the case in Hong Kong. And we have direct access to the airside. So, really like we put the cargo in a ULD and fly it straight to to the airplane. And last thing, we have CCTV recording of all steps. So, if ever we were to activate a ShockWatch with a forklift, we can immediately record the event, see what happened to the cargo, show a picture to the client, say, "Are we okay to move it? " So, that's Leo's case study of like our setup for example at origin in Hong Kong. Okay. Next important thing in term of execution is the tracking. So, let me walk you through like the key thing you want to look at on tracking. It's kind of your checklist for the tracking. So, you want to have the pre-booked truck at a major shipping location. Literally, a truck on standby ready to go at your major shipping location. You want to have the same team of driver on your large distribution location. Very important that they are trained on your SOP as a client. And it's always the same person. They really know all your procedures. You need to have high-value tracker with over 2 million of insurance. And why is that it so important is if you're moving a very high value components like server racks that are 500k a piece, I told you some of them can be up to a million a piece or more, well, you don't want to have a truck run with just one server rack. So, you want to be able to have on the truck as much cargo as possible. So, they need to have very high value so that you don't have to get PO to adjust to the value of your trucker or to have to take side insurance to cover the value that is an insured. Third thing, you want to have GPS localization and active monitoring on all truck. If a truck stop on the route, which they should never, you need to have the trucker dispatch immediately call the driver, make sure that everything is okay, that this is not tentative of theft. You want to have absolutely a shock-free handling and driving. So, air-ride trucks, airbags around the server racks. High security hardware, so you would have bolts in on all truck and container. What we were doing for some customer also is hardware cloth, so it's like chicken wire on some of the pallets was part of their SOP to make sure that there could not be theft. And last thing that is important, you want to have driver that are trained on dangerous goods, that know how to segregate dangerous goods to not create a truck fire, and that know how to say, "Hey, this particular type of dangerous good cannot be loaded on air freight, so I'm not going to accept this particular piece because it should go on an ocean shipment. " So, they act as a further safety net on your supply chain, and they are part of your execution chain. Next important thing, secure facility and verification of cargo at each step. So, you're going to go through bonding facility. You want this facility to uh adhere to the highest security standard. Typically, it would be TAPA certified. Uh you want to have cloud accessible CCTV surveillance 24/7 so that you can see at all time what happened to your cargo. Make sure everything is scan in scan out. You know, that's one of the ask that you should have to your forwarder. I want all pieces to be scan in scan out so that there's always a clear inventory of the pieces. And picture at of all the high-value shipment at every single touch point basically through the chain. Next, let's talk about the team. So, the way we organize the teams that support this account is we have one global account manager. Uh and this is typically a very senior leader with deep operation expertise. So, that these are the type of calibers that we would put on this account. And then, you have an account management team 24/7 with dedicated uh supervisor in all geography with deep expertise on the account. Uh and the team is organized as a fully empowered structure. So, let me talk about this in effect. So, for example, this team is allowed to bypass

the procurement team uh in my organization and call a charter operator to uh go find a charter for the client. So, they can take this decision and just bypass everyone in the organization to do what's right for the customer. Of course, this is because the individual we put on this team uh have a different level of uh of experience and they are able to make this type of decision. And the last thing, all the client on top of like a team that is very expert uh that builds rapport on the account, um we digitize all the client SOP uh so that basically they get to see what's needed to serve the account, what exception pop in real time versus the SOP of the customer, and they can act immediately. I think actually I have an example of this in the next slide. Yeah, so this is our SOP configuration platform. So basically, we can ingest any client SOP in native language running through our platform, and we retrieve real basically rules of how you execute on this account. And this rule, we call that create work item. So what an operator need to do to execute on this account. And this work item can be executed automatically by your system. For example, automatically send a shipment notice to this email address 5 day before the arrival to the site. Or it can flag an action to be done by your team. Hey, now you need to call the site manager because we are 12 hour to delivery, for example. So this is how it work, and this is a screenshot of our system with in this case a demo account. All right, sorry like my slide movement was a little slow. Next thing that is important is to have live visibility for all parties. So not only the logistics team. So traditional calendar reporting is not going to cut it for this type of execution. What you want is to be able to provide full visibility to the logistics team and to potentially a site manager or the client of the AI company, so that everyone knows when a specific PO and can search in this case in our system when a specific PO will arrive and can coordinate all the subsequent action. So [snorts] if you look at what you should be able to provide, uh, purchase level, uh, purchase order level live tracking is paramount. You need to be able to share, uh, the shipment status with the final client or the site manager. Uh, you need to provide centralized communication and source of truth for all actor. Uh, you need to make sure that all transit exception and immediately surfaced and resolved. All tilt and watch, uh, tilt and shock watch events are fed instantly into the forwarder system and, uh, is are visible to the client and we can take decision on, "Hey, do we keep shipping on this particular shipment or do we hold it? " And, uh, um you you want the same level of visibility on the account teams, the client, and the site manager for a direct shipment so that they can all take decision together. Also, this is an example, for example, uh, of how in our system we have an ability to share a public link to either a site manager or the client of our, uh, AI client so that they can track the shipment but with control on the visibility. So, for example, they would could not be able to see the commercial invoice. They can only see, like, the actual shipment milestones, the expected delivery date, the information that are relevant to them and it's configurable. And they could receive it, uh, through email notification or through mobile notification if they are on the way. Okay. Last piece of execution. Uh, you wouldn't want to have, like, a super fast, uh, ground and air supply chain and just get held in custom clearances. So, you want to avoid any type of, uh, custom delay. So, the way you do it is you want to have the product portfolio stored in your forwarder system. You want to pre-screen all your HS code and you want to a system that is able to flag when a specific HS code has expired. I think this you will see it in my presentation actually how we actually flag it in our system. Prior to flight landing, you want to make sure that all the clearances are prefilled with an AI audit that run to make sure that we don't do not have any error. And

so that basically you are able to maintain 99% plus on time and compliant clearance. And the last thing is as a shipper, you want to be very careful to maintain your CTPAT or equivalent certification. Which allow you to lower the inspection rate. For example, when you cross the border. So the CTPAT is basically a certification that you participate in the good third party program of the United States. If you have this, your inspection rate crossing the border to the United States will be lower. So these are the thing we always look at with our client, what certification do they hold and we recommend them which ones they should actually have that they don't have. With that, I've talked about best practice in execution. We have roughly 8 minutes in this webinar, so I'm going to hand over to Constantina to discuss the cost mitigation strategy before we actually close up. All right. Thanks, Alexi. Okay, I will pick it up on the cost mitigation side of things and that's actually something that comes up fairly often with customers that have large supply chains that are exposed to a lot of air freight. In AI, I'll actually also be on the air. So, what's important here is that we take a look at the cargo as a portfolio. So, it's pretty tempting to look at a large infrastructure project and say everything needs to move really fast, but that can get very expensive very quickly. So, then you need to look at the portfolio and really segregate how fast like a 300k to half a million rack should move and with what kind of like protections and white glove delivery versus like a very low value component like a widget that's also like a not time critical for an infrastructure project. So, when you work with a forwarder, you want them to give you like the full gamut of options. So, all the way from like a next-flight-out solutions, which would be like the fastest way you could possibly move, all the way down to deferred air. And not only that, also have some like hybrid instruments on hand for you. So, CR solution to Europe, for example, or like a a fast boat into the US. And that extends all the way into tracking. So, in air freight, we frequently say that speed is won on the ground. So, I will change this to say that speed is modulated on the ground. So, you could have a tracking solution that goes slow or that goes fast. So, here on the left-hand side, you'll see an example of flows that inbound into LA, and then they get tracked down into Mexico. So, depending how time critical the flow is, it could have like a slow option, regular clearance at the border, takes a couple of days, two, three days, or you could have a fast option, direct bonded truck that goes all the way to destination there. All right. So, that gives you a little bit of a lay of the land of what you should expect and what you should uh demand from your logistics provider. To close us out, I wanted to spend a little bit of time on a topic that has come up uh fairly frequently um lately, and that's on bias console. And uh that's relevant to the little um action bar at the top uh where if you click it twice, you would be requesting supply chain analysis. So, uh this is an example of a supply chain analysis we did for a large tech company — [snorts] — where we analyzed their ocean flows, and then we uh discovered more than 10% optimization potential um of the cargo. So, we did that by looking at their LCL volume and at the FCL volume, and how much of that could combine into fewer containers. So, um the idea is simple, but the modeling is hard. So, we've seen many instances where uh customers have taken a crack at it with some junior analyst that didn't understand the constraints of physical movement and cross-border and what needs to wait for what. Um so, if you have that analytical capacity on your team, uh it's a great exercise for them. Uh if not, work with your logistics provider, or Flexport is happy to help you with that. So, that is the modeling part. Now, uh you have that analysis, and you say, "Okay, I looked in the rearview mirror. That looks great. How do I put it in practice? " So, what we have done at Flexport is we've have embedded it into our platform. With this, there is a module in our platform where the model runs continuously in the back, and then depending on your business rules, how sensitive the different components are

to transit, and how far you would allow them to be fetched at origin just to put like in the in simple terms, then you either give like a pre-approval uh for us to consolidate your cargos into fewer containers, or you'll say, "I want to be uh in the driver's seat every time. You'll have to present me the opportunity and I want to be the one authorizing that. And it will always give you a summary of how much money you saved, what trade-offs you actually made in terms of transit time. And then we present to you with your essentially a cockpit of your consolidation engine. All right. What did we do today? So, today we reviewed um your AI playbook for logistics. So, it runs on two parallel tracks. One is execution excellence, what Alexi took us deep into, how you ensure that the added work that's built around your supply chain is fast and resilient, how we minimize different risks to your cargo both from a safety perspective and also from like integrity of the cargo. And on the other side, cost mitigation strategies, you can run up a pretty hefty bill if you're not careful with how you segregate your supply chain and different instruments you have at your disposal for that. And with that, let's go through Q& A and we also have opened up the survey. So, please give us feedback. What would you like to see next? How can we make this webinar more useful for you? — [snorts] — Alexi is um a question you would like to take a pass at? Uh I can take the first question that was posted. Uh oh, there are a lot of questions. Okay. So, the first question is not specific to AI supply chain. It's about AI to assist airlines. So, let me read it back. So, how can AI assist airlines to reduce the operational issue with moving the good or splitting it, is Flexport working with airline to achieve it? Uh so, yes, we are working with airline. We also have our own plane, so we have a necessity, for example, to work with Atlas Air, who is the operator on our own plane to uh improve uh the quality of execution. Um high level uh here is how AI can uh help you execute better on an airline. So, the first thing it can allow you to digitize uh the SOP of your forwarders that are your client in the case of an airline and understand exactly what they want. And in the same way forwarders surfaced an execution failure uh for a shipper, well, as an airline, you can surface execution failure on a forwarder SOP. So, that's number one. Uh it's not AI, but it's predictive analytic. Uh you can use predictive analytic to drive a better uh overbooking strategy. Uh you can use also predictive analytic to uh better predict uh the available cargo capacity you will have on a plane uh if it's a passenger plane by predicting how much baggage uh the passenger will bring, for example. If it's a cargo plane, by uh looking at historical data on the weather and say, "Oh, I expect a lot of uh headwind on that route and therefore the uh the payload will reduce in the coming day. " Uh so, it can allow you to drive your available cargo capacity. Uh where AI where we're implementing AI right now uh not predictive analytic, but AI as well is uh detection of maintenance needs. So, maintenance is a very, very hard thing uh for um uh for the airline community because it's not very deterministic, you know? Each airplane is its own beast, so you need to combine the data on what elements are susceptible to fail with the human expertise of this technician that knows the airplane and say, "No, I think this airplane is going to be okay. " Or no, I think we need to change the piece a little bit earlier. So, AI is very good at this at surfacing to humans the action they should take or check. Uh and the last thing where we're implementing AI with airline and our own fleet is just anomaly detection. Hey, this client typically booked this much and today they have not booked. Maybe we should call them and make sure that the usual bookings they have is going to come, otherwise we should just replace the cargo. So, these are just example and I can go a bit deeper, but I don't want to take all the time in this uh in this webinar. Constantina, you want to take the next

question in line? — Yeah. Uh I'll read them out. Uh do you think the demand for this particular type commodities to stay as a structural change for air cargo? How far it could move to sea freight in the future? And I'll combine it with the next one. Right now, it seems the movement is concentrated on a few lanes. Do you see this changing any soon? Um So, there are sort of like two kinds of thought. One says, "Okay, AI is is here to stay, which is where with the one I subscribe. " And the one thinks this is a lot of like um a excessive investment that will decrease in the future. So, just to acknowledge that there is sort of like descending thoughts on this topic. I for now, given how quickly the infrastructure needs to be deployed, I do see it being a freight dominant. However, because it is so expensive, There's already a lot of push to move parts of the supply chain into ocean, and then ocean itself has enough They have a certain amount of like fast boat options that help bridge the gap. So, there's already some amount of the infrastructure going to ocean, but not the sensitive cargo. That's way too risky. And then on the other question, if it's going to be the same corridors or others, this is very interesting and it came up recently in a conversation. So, um the US is a dominant destination. However, what drives the build-out of the sites is suitability of the sites. And the suitability of the sites right now is access to behind the mirror, electricity, and water. So, you'll see sort of like if I were to pull like a US map, the concentration of data centers is driven by that. A Now, the interesting part is that this is a bit of opportunity for countries, right? So, a lot of countries are putting incentive in place to start attracting some of the data center build-out investment. So, I have customers in their portfolio that are serving this build-out um in India, for example, or other Southeast Asian countries, which it might seem niche, but it's like for smaller modular data centers because everybody is trying to diversify their portfolio and give a boost to their country. So, it's a very interesting uh environment, and but I expect this to take like at least a couple of years before you see like Tai Taiwan to the US being dethroned or like Thailand dethroned. Alex, maybe we have time for like a couple more? I think just one more because we are overrunning. So, maybe take another one. I Okay, I'll take a quick one. Does the Flexport system feed customers ERP system or only receives purchase order information and provide shipping information for customer to track PO status in the Flexport platform? Quick answer, it does both. So, if you have an ERP solution, the Flexport system connects to your ERP solution and writes into it your purchase order updates and also stores it in the platform. So, then you can store in your ERP system, keep it as a system of record, but if you want the solution to be able to chat with your providers in your ecosystem, you can do it on the platform and they're both synced and they will show the same information. And typically, our platform is a bit more intuitive when it's like a mobile service team or site managers that is not from logistics, you know, they can literally just search the name of the project or piece and actually see it in our platform. So, that's the advantage of our platform for this. And I think we're done. Thank you very much. Please make sure to leave us feedback in the survey. It's a very rich topic. We would love to do a follow-on according to the interest of our audience. Okay.