The Outsize Impact of AI Logistics

2025 July 5 Twitter Substack See all posts


Automated logistics will heavy impact one third of GDP.

AI's Impact on the Physical World

A common criticism of recent AI advances is that they only impact the digital world and not the physical one. Robotics can increase the impact, but which robots are the most important? Below, I'll make the case that logistics robots should have a dramatic effect on our economy and daily life.

Logistics In America

Logistics directly accounts for ~10% of the US economy and heavily impacts another quarter of GDP through the manufacturing, wholesale trade, retail, and construction industries. The usual ways to move goods are:

  1. Truck

    Trucking accounts for the majority of freight by volume and value due to its flexibility and speed. It beat rail due to the interstate system and improved truck technology. Costs are typically $0.10 to $0.30 per ton-mile for full trucks. Less-than-truckload can be 10x-100x more per ton-mile.

  2. Delivery Vans, Consumer Cars, etc.

    Mostly local, last-mile delivery. The cost tends to be astronomical, often $15 to $200 per ton-mile. Consumers also pay similar rates shopping at retail stores but internalize the costs by sacrificing time and adding personal vehicle miles.

  3. Freight Rail

    The US has one of the best freight rail systems in the world. Railroads focus on grain, coal, oil, or chemicals. Long-haul containers have also been a growth area. They charge $0.03 to $0.05 per ton-mile under normal conditions. Rail tends to be slower and less predictable than trucks and has high fixed costs. It is not competitive in higher-value shipments or short hauls.

  4. Air Freight

    Air freight is specialized for speed and costs more than $2.00 per ton-mile. Air freight still requires some form of last-mile.

  5. Domestic Barge/Ship

    Shipping on water costs $0.01 to $0.03 per ton-mile. It is mostly river barges in the US because of the Jones Act. It makes sense for bulk goods without time constraints or large items that can't fit on rail or truck.

  6. Pipelines

    Pipes are an absurdly inexpensive way to move liquids and gases in bulk.

The relative importance of different logistics systems shapes America. Firms build big box stores or select factory locations to minimize their logistics burden. Wal-Mart and Costco are directly downstream of truck dominance, for example.

Any advance in logistics has massive effects on our fabric of life and standard of living.

Breaking Down Cost Structures

Trucking and last-mile dominate total revenue share and matter most. Trains, pipelines, and water transport will continue to move large quantities of bulk goods. Traditional air freight will remain niche.

Trucks main cost categories are labor-related costs and fuel. They are >70% of semi-truck costs. The truck itself and maintenance account for the remainder.

Last-mile and sorting are the majority of the total logistics costs for parcels and deliveries. Labor costs are much higher because a driver carrying 500 pounds of packages costs about the same as a semi-truck driver carrying 30 tons of freight. Packages are unloaded one by one instead of with forklifts. The percentage of cargo weight to vehicle weight is low compared to semi-trucks. These inefficiencies are why many stores are warehouses where customers can load up and make the most of their trip. And why distribution centers for last-mile delivery are close to customers. The proximity allows semi-trucks to carry as many ton-miles as possible.

Labor, fuel, maintenance, and vehicle weight/cost are the most critical metrics to attack for improvement. The primary tension is that our existing transport technologies require scale and centralization for efficiency, but final customers are diffuse.

The AI and Electrification Opportunity

The first-order answer for cost reductions is clear. Remove the driver with software and reduce fuel costs with batteries. Automated driving is improving rapidly. LFP battery packs for autonomous semi-trucks are already theoretically cheap enough except in the US. Trucking would see significant cost declines. I think this result is already a given, especially since AI drivers can use small battery packs that charge more often than human driver hour rules allow.

The second-order optimization is much more interesting. The minimum cost of any size shipment plummets by removing the human driver and using a versatile electric powertrain. Haulers can use lower minimum costs to break up shipments into smaller chunks. That means more point-to-point, eliminating much of the distribution that today's truck logistics require. Prices for some categories, like less-than-truckload, could decline by 80% to 90%. These changes are extreme enough to reorganize goods-oriented businesses.

Full Trucks and Pallet Aggregation

The humble pallet is the default unit of transport in the US. Truck trailers are glorified pallet transporters. Only categories like parcels or e-commerce shipments still regularly stack boxes. The pallet is so productive because it allows mechanized transport via forklifts, eliminating break bulk labor. The risk of damaged or lost goods also falls precipitously. It is a modern miracle. Aggregating and disaggregating dozens of pallets in trucks happens to gain economies of scale in shipping.

The name of the game for minimizing truck logistics costs is to have full trucks drive point-to-point with no side stops. Less-than-truckload type pickups and deliveries incur last-mile inefficiencies. The distribution center, perfected by companies like Wal-Mart, unlocks full trucks point-to-point. A truck can take a load of grape jelly from the factory to the distribution center. Pallets of grape jelly are loaded on different trucks heading to stores. These trucks also carry other assorted items, like peanut butter, bread, or paper plates, that arrive on single trucks from their respective factories. The most efficient of these systems can "cross-dock" pallets in <24 hours. Sam's Club and Costco are the peak of this style of logistics, where pallets are not even unpacked but wheeled into aisles for customers to pick from.

The drawbacks of this system are that it only works at very high volumes, and distribution centers are expensive.

The extra handling is costly because one pallet might require 5-6 forklift moves to reach the shelf at Costco. And the intermediate goods to create those products require many times that. There are incredible efficiency and speed gains available for any technology that can cost-effectively end pallet aggregation.

Getting the Right Form Factor

Electric semi-trucks and vans are not optimal for hauling pallets.

Platform Requirements

Probably the most fitting vehicle for pallet hauling is one utilizing Tesla's CyberCab platform. The attributes of a high-utilization taxi are similar to dedicated pallet transport:

  1. Each CyberCab factory line is designed to have an output 5x-10x higher than today's automotive lines, unlocking affordable scale.

  2. Components like motors are made of inexpensive, available materials. They sacrifice top-end performance to meet cost and efficiency goals.

  3. The car will get more than 6 miles per kilowatt-hour or greater than 200 miles per gallon equivalent.

  4. The life is likely to exceed 1 million miles, and the engineers designed the car for minimal maintenance.

  5. The cost goal for the taxi is $25,000. A version without doors, climate control, seats, screens, airbags, or extensive roll cage/crash protection would be significantly cheaper.

A vehicle on a similar platform can alter the pallet transportation norm with its scale, cost, and reliability.

Charging Up

The scalability of the charging infrastructure is also crucial. Building truck depots that can handle trucks charging at one megawatt each is a challenge in any regulatory environment. Slow-moving utilities and other barriers add to the pain. A small pallet hauler would fit into the existing infrastructure better with more distributed charging.

The battery performance and charge rate requirements also change once driving is automated. Driver hour and rest regulations determine battery requirements for human trucks. Ideally, the driver can drive all day and only take one relatively short charging break to keep within their hour windows. Automated vehicles don't care when they charge or how many times they stop. The question is purely cost minimization.

The lifetime of the car will be limited to a few years, even at 1 million miles. Taking more charging stops and spending 25% of the time charging could save several cents per mile versus higher utilization from rapid charging. Charging quickly becomes expensive because of utility demand charges and the negative impact on vehicle battery life. Spreading miles over 30 months instead of 24 months won't have a significant cost impact.

Safety Considerations

Safety is a concern for autonomous semi-trucks because any error with a 30-ton behemoth can have serious consequences. Alternatively, a small vehicle with a flimsy shell weighing in at a few thousand pounds will do much less damage in a crash. If the only rigid portion of the vehicle is the "skateboard," then impacts with cars and pedestrians will be much safer because the stiff portion of the vehicle will be going under the car or chopping at the person's legs rather than impacting more vulnerable areas.

Better stopping performance and the lack of a hinged trailer are also helpful.

Regulatory approval should be easier, insurance cheaper, and driving performance won't stop improving.

Cost Estimates

A typical loaded pallet might weigh 2000 pounds, so the cost per mile and cost per ton-mile are the same. Tesla estimates $0.20/mile for the taxi service, but pallet hauling should be more competitive. The capital cost and maintenance should both be less because the pallet hauler has fewer sub-systems. Higher average speeds and utilization compared to taxi service will improve amortization. Insurance will be less without carrying people. Pallets spill fewer drinks and barf less than humans. $0.10-$0.15 per mile seems reasonable, making the vehicle competitive with today's trucks before considering systemic benefits. $0.05-$0.07 per mile could be a good long-term goal after insurance costs decrease and other incremental improvements add up.

A real-life example: I recently had a pallet delivered from ~450 miles away, with freight costing $376. A $0.10/mile pallet hauler would have been $45. A semi-carrying 26 pallets point-to-point would cost $40/pallet. A single-pallet carrier can be competitive with today's trucks at 1/25th the scale, even under the most generous assumptions for semi-trucks.

Eliminating Distribution, Forklifts, and Loading Docks

Handling, loading, and unloading are important considerations. The delivery cost I paid includes handling charges. Point-to-point estimates based on dollars per ton-mile do not. A rule of thumb is that each pallet touch costs $3-$5 plus storage costs of $12 per pallet per month. Those costs include the forklift, labor, paperwork, damage, and the warehouse.

The load/unload cost of the pallet carrier can decrease substantially if it can handle the pallet itself and simplify the process by traveling directly to the customer. The vehicle is small enough to drive into warehouses, factories, or stores to drop off or pick up pallets directly. Imagine dozens of bubble cars driving into Costco right after closing, gliding to the correct aisles, and dropping off pallets. They can drive into a factory and travel straight to the station that needs the parts. A touch like this could cost ~$0.25. The direct method could save $20-$30 per pallet compared to the traditional system by reducing the number of pallet moves and the cost per move.

The benefits extend further. It is common for trucks to wait hours to get loaded because the warehouse is busy. These problems are more likely at the median warehouse than with efficiency superstars like Wal-Mart. Not only do these delays idle trucks and drivers, but they make scheduling less predictable. Being able to drive directly to the load eliminates delays, improves predictability, and adds surge logistics capacity.

Finally, total delivery times can decrease because goods go straight to their destination instead of spending 12-48 hours in each distribution center they pass through.

The upshot is that a pallet hauler can further reduce costs by avoiding distribution bottlenecks and costs.

Instant Quoting and De-Bottlenecking Producers

A world where a producer can deliver a pallet directly to customers for less than Wal-Mart can get it to a Super Center's loading dock will change the relationship between producers and customers. Small direct orders are more feasible. Final prices can fall significantly in industries with inefficient distributors, like lumber. Delivery time with point-to-point drops from days to hours. A general contractor should be able to order lumber packages directly from the mill for next-day delivery and save 20%-30% versus today's distributor.

Producers can't just snap their fingers to make this happen, but several developments in AI and a switch to point-to-point logistics make handling higher volumes of small orders easier:

  1. Eliminating Human Resources Challenges

    Truck drivers and forklift operators are difficult to staff. Turnover rates can be as high as 50%. Managing a human resources challenge like this is often much different than operating a typical business. Staffing difficulties make it less desirable to expand internal capacity. Substituting AI for these "dull, dirty, dangerous" jobs makes expansion into direct-to-customer more feasible.

  2. Moving to Instant Quotes

    Overhead can drag down margins on smaller accounts and orders. Plus, the time of response with quotes heavily impacts the acceptance rate (and customer acquisition costs).

    Instant quotes are a forcing function and secret weapon to take advantage of point-to-point logistics. However, instant quotes only work with a clear model of the business, inventory, and supplier relationships. Again, AI lowering the cost of office labor and software makes this leap easier.

  3. In-housing More Functions

    A challenge with instant quotes is including subcomponents from vendors that don't have visible price and delivery information. There are also functions distributors do that a bulk producer would need to add. A revolution is brewing in more flexible production systems that use AI and software to make in-housing new processes more feasible (I plan to write about this soon).

  4. Reducing Delivery Complexity and Increasing Certainty

    Scheduling can lead to a lot of back-and-forth communication when the complex distribution path is opaque to customers. Point-to-point paired with instant quoting makes scheduling very simple. It can also make suppliers more reliable and regular, allowing them to enter the instant quote envelope.

  5. Educating Customers at Zero Marginal Cost

    Another function that distributors manage is customer education. Instant quoting systems can have built-in restrictions and information that convey to customers what is possible and what is affordable. At the highest fidelity level, these systems can display the hard edges of the production system with no labor from the producer, allowing better allocation of each party's resources. It is a version of cross-organization "design-for-manufacturing."

Inexpensive point-to-point logistics opens up the opportunity to eliminate middlemen and achieve double-digit cost reductions from supply chains. Structural reforms are necessary for most vendors to meet this demand successfully.

Crashing Last Mile Costs with Drones

Packetizing pallet transportation still does not solve the issue of getting goods into homes and businesses. Rapidly scaling drone delivery could break that link and reduce the horrific less-than-pallet premium.

Economics of Aerial Drone Delivery

Most companies seem to be pursuing aerial drones with these features:

These drones are efficient due to their small size, getting 20-30 miles per kWh (800 MPGe). And they are inexpensive. Some analysts estimate the eventual cost floor at $0.03/mile plus load/unload costs. Numbers like that would be remarkable because that is $6/ton-mile, not far off point-to-point air freight at 1/10,000th the scale. These figures are a fraction of traditional last-mile costs. Deliveries can eventually be less than a dollar.

The industry is still in its tumultuous early stages, and current costs can run anywhere from $4 to $60 per delivery. The drones are improving. The ratio of drones to human supervisors should increase, and organizations are making charging and loading drones more efficient. The cost leader, Manna, is currently at $4 deliveries and sees $1 as a reasonable goal. $4 is already competitive with typical route-based delivery. It is a fraction of the point-to-point grocery/food delivery cost.

Aerial drones should dominate time-sensitive deliveries. The drone's eventual $0.03/mile is cheaper than an autonomous ground vehicle, but it only carries a fraction of the payload. The aerial drone's killer advantage is in speed and flexibility.

Ground Drones

I've previously discussed lightweight, automated delivery vehicles here and here. They will make the most sense for larger/heavier deliveries. Their cost structure should be similar at the limit, costing <$0.10/mile.

Speed as a Service

Today's last-mile delivery has similar constraints to truck logistics. Sorting, distribution, and last-mile account for 70%-85% of delivery costs. Minimizing costs requires filling trucks at a local distribution center and then having them drive a route with stops as close together as possible. Scale drives efficiency through more distribution centers and high customer density.

A downside is that deliveries are locked in a daily cycle. A package must be at the distribution center, ready to load before the van leaves. Same for any outbound pickups.

Drones providing cheap point-to-point deliveries crack this paradigm. Deliveries (or pickups) happen when an item is ready at a fraction of the cost. Middle-mile trucks can sortie between distribution centers more regularly rather than leaving at the end of the day. Even better, companies can use pallet hauler vehicles instead of semi-trucks to increase service frequency between hubs by 25x. Package delivery time is roughly the drive time between two hubs, shaving days off the schedule.

The overall effect is profound. Packages have air freight delivery times and might be 70% cheaper than today's ground shipping. Packetized connections between hubs and $1 last-mile pickups and deliveries will lead to a new era of delivery efficiency.

Eliminating Break Bulk in Distribution

Unfortunately, moving away from a daily logistics cycle does not help the mess in distribution and sorting centers, nor does it solve the loading cost of drones. Incoming shipments require some combination of these actions:

  1. Unload pallet or loose boxes.
  2. Break down from pallets or unbox.
  3. Place on shelves.
  4. Pick from the shelves for shipping.
  5. Box for shipment.
  6. Load vehicles.

Some companies, like Amazon, have developed automation for many of these steps. These systems are still expensive to own and operate. Most operations do tasks by hand. Food delivery shoppers pick from regular store shelves. Employees have the load the coffee on the delivery drones in the parking lot.

Extending the Shipping Container Paradigm

One solution without messy robotics is to adopt small, reusable plastic containers of various sizes to replace boxes. In today's logistics, the cost of each delivery and pickup is high. Operators carefully orchestrate truck space, and picking up empty containers is a burden. A box adds $1 to a $4-$8 delivery and reduces system complexity, making them ideal for today's logistics.

That math might change under new conditions for several reasons:

  1. Point-to-point has more capacity.

    Return legs on point-to-point will often be empty, allowing vehicles to carry containers back with them. Today's delivery trucks, packed with boxes several layers deep, would struggle to take containers back.

  2. Drones and self-unloading vehicles eliminate labor.

    A human delivery driver working for a company that adopted containers would have to take the item out of the container. And then humans would unload the truck once it returns from a route, because the inside of most delivery vans is not accessible to robot arms or cranes.

    An aerial drone can drop the container wherever it needs to go. Ground vehicles can self-unload to allow access to robot arms or other drones.

  3. Low per-mile costs absorb lost volume and weight capacity.

    A break-bulk ship packed by expert longshoremen can probably maximize its volumetric and mass capacity better than a container ship. Containers are heavy and use extra space. But ships are so efficient at moving goods that the extra container mass and volume are worth it to improve loading and unloading efficiency.

    Vehicles with very low cost per mile make the inevitable backhaul of empty containers more affordable and an acceptable tradeoff for lowering distribution handling costs.

  4. Low-cost, generic container movers instead of complex automation.

    Warehouses with rollers, conveyors, automated boxing, and pick-and-place robots are capital-intensive. They require effort to maintain and operate.

    Container systems with standardized latch points enable lower-order automation. The same drones that deliver packages could swarm a pallet of containers dropped off by a ground vehicle and deliver them if the customer is nearby or place the container on the pallet for a ground vehicle headed to the next hub.

    Mini-hoist cranes could unload and load heavier containers onto mini-flatbeds that carry them to the correct pallet for their next destination.

    The net effect is that a distribution center could practically be in an empty parking lot or shed. The required level of automation already exists. And cross-dock times with 1/25th the packet size between hubs could be 15 minutes.

  5. Potential for lower system inventory makes warehousing less attractive.

    There might be a choice between two systems. The existing one can deliver to customers within a few days for most items and hours for the highest volume SKUs at a cost. Or a new system that can deliver any item at the speed of transport. Ground vehicles could provide 2-day shipping at a minimum to any point in the continental US. The vast majority of the population would have shipping measured in hours.

    The catch for the new system is that the item must travel in a container that enables rapid cross-docking at these extremely efficient transshipment hubs. The reward of drastically reducing logistics costs could be worth it.

A Day in the Life of a Package

XYZ company produces niche phone cases. Most of its orders come through online channels and are ultimately shipped to homes or in small batches to specialty retailers. Orders travel through Amazon's system or as parcels via UPS, FedEx, USPS, etc. Packages arrive at customers' homes in 1-5 days.

Logistics costs have several sources and dwarf production costs, which are less than a dollar. Each shift must have several employees in shipping/packaging to manage labels, pack boxes, or load pallets. The company has to pay Amazon or sell at wholesale to retailers. Besides the direct costs, there must be XYZ inventory in Amazon's warehouses to maintain 2-day delivery or at stores to sell effectively. There is also a daily inventory build-up in shipping as pickups for Amazon or a parcel service are lumpy. Like the proverbial box of corn flakes that costs the same to stock on the shelf with corn flakes inside or not, the customer isn't paying for the phone case but for everything it takes to get it to them.

Imagine that XYZ switches to the new logistics model. The factory is in the Mid-Atlantic. The majority of the US population is single-digit hours away. Orders can now be a true customer pull model where products are made within hours of ordering. XYZ uses injection molding, so there are still batch runs of certain colors or models. The low-cost direct logistics model makes it worth running smaller batches, using more flexible machines, and focusing on short setup times.

Employees load products straight into their containers, which are then stacked on pallets. Many items may not even require traditional packaging. A pallet hauling vehicle grabs the pallet at the end of the line once it is full and transports it to the nearest distribution hub, usually several times per hour. Drones swarm the pallet of containers once it arrives, quickly whisking the containers to the appropriate pallet for their next leg. Once the item reaches the hub closest to the customer then a drone will carry it to the customer's house.

We can compare pricing using Amazon's seller tool. A $9 phone case costs $2.59 for Amazon to stock. The customer pays a few dollars more for delivery directly or via subscription. There is also shipping to the Amazon fulfillment center, excess inventory, and packing/packaging at the factory. The total logistics costs are probably around $6 per phone case.

The direct shipping model could be less than a dollar end-to-end once fully developed. Assume a loaded pallet shipping the containers is mostly air because the standardized containers are large compared to a phone case. One holds 500 units. Each item moves through 4 hubs (2 local, 2 regional). For cost estimates:

4 transhipments are $0.17. The cost to the final distribution center 500 miles away would be roughly $0.10. A home 8 miles away from a hub would incur $0.48. The time elapsed from order to delivery is 12 to 13 hours, with the total cost at $0.75.

A system like this won't happen overnight. The most cost-sensitive portion is the aerial drones. They make the final delivery and move small containers in the transshipment hubs. Drone software needs to improve until one human can supervise thousands of drones. The vehicles require several iterations to reach the correct blend of efficiency, cost, and reliability. Loading, unloading, and picking/cross-docking must be autonomous. Factories must adjust to the new logistics paradigm. New companies need to compete with firms like UPS and FedEx. But the potential is certainly there.

Handling Food and Groceries

Food is unusual because of the mass and variety that a household needs. Getting 20 deliveries in a day instead of one grocery run is annoying. Speed is also at a premium, and delivery times typically need to be <15 minutes for takeout and within a few hours for groceries. There is no avoiding some intermediate distribution. It seems likely that change will accelerate long-running trends.

Meals prepared outside the home continue to grow as a share of food spending despite the high costs of restaurants and delivery. Direct delivery of pallets and rapid, low-cost delivery could unlock a new era of food production efficiency. Prices for high-volume items could rival home cooking costs. Pallet haulers could deliver pallets of ingredients straight to cooking stations, and a drone aviary at the end of the line could eliminate trips to the parking lot. Quality should increase compared to fast food. Productive central kitchens can afford well-paid cooks and quality control. Disposable containers aren't going anywhere in this application. Flimsy bags and containers are viable without transshipment. Presumably, in-person restaurants will cater even more to a pleasant and unique dining experience while delivery displaces more frozen meals and home cooking.

Grocery and warehouse stores should still exist with modifications. The in-person experience at grocery stores should push further towards beautiful displays and fancy items. And some people will still want to buy in bulk. However, there also needs to be versions like today's Amazon warehouses with picking robots that facilitate rapid fulfillment and loading of ground and aerial drones.

Food should taste better, have better experiences, and be more affordable than ever before.

Downstream Impacts on Distribution and Industry

The most disruptive force could be a stronger "gravity" model. A trip across the ocean on a plane or container ship today is a small portion of logistics costs. Warehousing and distribution dominate the total. Nor is transoceanic trade impacting delivery times because there are adequate inventory buffers in the system. Regional sourcing is more competitive under the new model because warehouse and distribution costs fall dramatically. That will push some items away from hyperspecialized global factories toward regional production.

Labor demand (for robots or humans) in manufacturing will grow as falling logistics costs stimulate demand and production shifts to less specialized factories.

The returns to large internal markets should also grow. Various paperwork and delays that are merely annoying today are large enough to eliminate the benefits of the new paradigm. That means the US and China should derive the most benefit. India and the EU could benefit by reducing internal barriers.

The End of Human Limited Logistics

Logistics robots will have a much bigger impact than production robots. The pure manufacturing cost in the phone case example might be $0.50. Most mass-produced products drive labor and capital expenditures to relatively small shares of production cost. A robot arm to load cases into containers or automatically change a mold might save a few cents per unit. It seems more likely that production robots will increase variety, shrink minimum scale, and compress development times. Low and high volume item costs can converge. Convergence has many interesting implications and benefits, but the cost floor will remain the same.

Similar to the production robots, the cost of shipping bulk goods via pipeline or ship won't change much. Humans are basically already out of the loop on these modes.

Intermediate goods traveling on pallets will have a significant cost reduction in transportation, even in the first-order scenario of automated trucks, forklifts, and electrification. Packetizing pallet movement with the "pallet hauler" bots will bring all shipments close to the cost floor without needing centralized logistics nodes.

The most dramatic changes come in the last-mile and parcel-size shipments. Human-centric logistics has always faced difficult tradeoffs between convenience and cost efficiency. Cost savings from slashing distribution centers and last-mile costs are two orders of magnitude larger than what production robots can do. More direct, frequent, and cheaper deliveries will be a massive expansion of the productivity frontier.

These improvements will disintermediate ~20% of current GDP in logistics, wholesale trade, and retail while reorganizing another 10-15% in manufacturing and construction. In the process, they will eliminate some of the jobs with the highest turnover due to their unpleasantness. Humans will be much better off.