PepsiCo Confirms 35 Driverless Trucks Operating on Public Highways Across Arizona and Texas

The Analysis

I read the PepsiCo driverless-truck number and the first thing I did was stop picturing a truck. I pictured a server. Thirty-five of them, running on public roads in Arizona and Texas between bottling plants, warehouses, and the back doors of Walmart and Dollar General, in real commercial service, not a pilot. That detail matters more than the headline, because the moment you model a truck as a server in a linehaul queue, the whole story changes shape. The driver was the constraint. Take the driver out and you have to ask what constraint is left.

The report, from the Wall Street Journal on 13 June 2026, makes PepsiCo the first big consumer-goods company to confirm driverless trucks on revenue lanes at that scale. Before we celebrate the future, let me put the system on the table and find the bottleneck, because removing one constraint never removes the next one, and the next one is where the real cost lives.

Slice the fact down to what we can stand behind. PepsiCo has said it is running thirty-five autonomous trucks in actual commercial service on public highways and city streets across Arizona and Texas. These trucks move product between bottling plants, warehouses, and the stores of Walmart and Dollar General. This is not a test loop. The company calls it active commercial revenue lanes, and it is the first time a major consumer-packaged-goods shipper has publicly confirmed driverless operations at that size. The fleet runs with technology partners and plugs into PepsiCo's existing transportation planning systems. Those are the hard points. Everything about 'the future of trucking' around it is interpretation, and interpretation is what I want to pressure-test.

Treat the linehaul as a queue, because that is what it is. A shipment arrives as a job. A truck is the server that moves it. The driver is the part of the server that used to cap everything: hours-of-service rules, the need for sleep, the fact that one person can only run a truck so many hours a day. Remove the driver and the server's capacity constraint shifts from human-hours to vehicle-hours, maintenance windows, and road time. The bottleneck migrated off the person and onto the asset and the schedule. That is the actual event underneath the press release. Not 'no driver.' A constraint moved from a regulated human to a machine and a plan.

Now why this, why now, why Arizona and Texas. The trigger is not that self-driving suddenly works everywhere. It is that PepsiCo built a set of lanes with low variance. Bottling plant to warehouse to a known store dock, on a fixed schedule, on roads the partner has mapped and re-mapped. Low variance is the whole secret. An autonomy stack is a server that handles a well-behaved queue beautifully and a messy queue badly. PepsiCo handed it the well-behaved queue. Add to that the two states' permissive frameworks letting these trucks run on public roads, and you get the crossing: the service is reliable enough to carry revenue, and the regulator allows it in public. The cause is lane discipline first, technology second. If you flip either, the thirty-five trucks stop making sense.

Layer the impact and say who feels what. PepsiCo itself saves driver cost on these lanes, but thirty-five trucks are a rounding error against a national fleet that runs in the thousands, so the direct saving is real and small and mostly a learning investment. The technology partners, the autonomy firms behind the wheel, get the proof point they needed to go sell the next shipper, and that is where the bigger value sits for them. The drivers on exactly these thirty-five lanes are displaced, but the number is tiny against the national driver pool, so the labor story is symbolic this year, not structural. Wall Street gets a re-rating of autonomous-freight names, which is why the stock tickers moved more than the trucks did. And for the importer or exporter reading this from the other side of an ocean, the relevant transmission is narrow: this touches US domestic linehaul and drayage cost structure, not your China-to-US ocean rate. Keep the scope honest or you will mis-price the risk.

Time the transmission. These thirty-five lanes are live now, in mid-2026, so the operational learning accrues this year. But scaling to a material share of any shipper's fleet needs more states with permissive rules, more lane types that behave, and insurance priced for the new normal. The related Volvo Autonomous Solutions target of three hundred plus trucks by the first quarter of 2027 tells you the industry's own clock: meaningful fleet presence in roughly twelve to eighteen months, broad cost impact on US trucking maybe 2027 through 2029. For our clients the near-term, next-twelve-months effect is limited to US inbound drayage and domestic distribution inside the Arizona-Texas corridors. Ocean freight, the long haul that actually moves your import, is untouched by this number. The clock is real but it is a slow clock.

Run a number so the saving is not a slogan. Suppose one of these lanes runs a round trip of roughly six hundred miles a day, and a human-driven truck is capped by hours-of-service at maybe five hundred miles of productive movement with the rest lost to the driver's mandatory rest and the single shift. An autonomous truck on the same asset can, in principle, run closer to the vehicle's road-time limit across a longer window because no single person needs to sleep. Call it a rough lift from five hundred to near the road limit, say six hundred miles of movement a day on the same truck.

That is about a fifth more asset movement from the same iron, before you count any driver-wage saving. Quantify before you optimize: the gain is real, but notice the variable that actually moved is asset utilization, not headcount. The truck was always capable of more road time; the driver was the cap. Remove the cap and the asset breathes. That is the mechanism, and it is cleaner than any 'trucks without drivers' framing.

Now the part nobody in the headline wrote, and it is the part that should change what you do. The real story is not 'thirty-five trucks with no driver.' It is that PepsiCo turned a slice of its linehaul into a closed, low-variance loop, bottling plant to store on fixed rhythm, which is exactly the kind of queue you can automate. What the coverage misses is that this only works because the lane variability was engineered down first. The autonomy did not tame a wild network. It was handed a tame one.

So for an importer or exporter, the lesson is not 'go buy autonomy.' It is 'go make your own lanes behave.' Standardize your US domestic routing, fix the schedules, clean the dock appointments, kill the one-off irregular runs, and you become a candidate for the same cost curve. The bottleneck to your own automation is not the technology. It is your own lane discipline. Most shippers have not even measured how variable their lanes are, so they blame the tool when the tool refuses their chaos.

There is a second missed point, and it is the deeper one for anyone who runs a network. Removing the driver exposes the next constraint, and the next constraint is harder than the driving. It is dispatch orchestration and empty-repositioning. A driverless truck still has to be in the right place with an empty trailer at the right time, and that is a planning problem spanning the whole network, not a perception problem on one vehicle. The autonomy community is brilliant at the vehicle and quiet on the system. But the queue that actually limits throughput is the matching of empty assets to pending jobs across a day, a week, a region.

Automate the cab and you still starve if the dispatcher, human or machine, cannot solve the empty-trailer problem faster than the freight arrives. So the bottleneck migrated from the seat to the schedule, and the schedule is where the money and the pain both live. Anyone selling you 'just take the driver out' is selling the visible part and ignoring the part that decides whether the truck ever moves loaded.

A third thing left out: the thirty-five trucks still sit inside a risk and ownership structure that autonomy does not dissolve. There is still a remote operator or safety supervision per the regulations, and there is still a technology partner taking margin on the run. So the cost-per-mile saving is real but the organizational bottleneck, who owns the autonomy stack and who eats the liability when a load is late or a sensor fails, has not moved at all. The headline implies a clean removal of cost. The reality is a shift of cost and risk to a different line on the same budget, with a different counterparty. That is fine, often worth it, but call it what it is. The field is not flatter because a cost vanished. It is flatter because a cost moved to someone whose whole business is absorbing it.

Flip the premise and watch the conclusion turn. If PepsiCo's lanes were irregular, long-tail, unknown docks and peak surges, the same thirty-five trucks would fail on edge cases and the program would still be a pilot. If Arizona and Texas had not offered permissive frameworks, this would not be on public roads, it would be on a private loop, and the revenue claim would disappear. If insurance had not priced the risk down, the per-mile saving would evaporate against the premium. So the claim is true inside a narrow box: predictable lanes, friendly states, priced risk. Step outside that box and the same trucks, same vendor, same hype, stop proving anything. A conclusion with that many boundary conditions is not a forecast of trucking. It is a description of one shipper's unusually tidy corner of it.

So what do you actually do if you are the importer or exporter whose freight lands in these corridors. Map your US domestic lanes by variance before you dream about autonomy: count how many of your runs are fixed-schedule and known-dock versus one-off and irregular, because only the first bucket is autonomy-ready and most networks are surprised by how small that bucket is. Attack variability on purpose: standardize appointment windows, repeat routings, and dock readiness so the same lane runs the same way every time, because that repetition is the raw material autonomy needs and you can build it without buying a single truck.

Watch the asset-utilization lever, not the headcount lever: when you evaluate any domestic carrier or 3PL, ask how many revenue miles their tractors run per day, because the autonomy gain is utilization, and a carrier already running high utilization has less to give you than one that is leaving iron idle. Time your own exposure: the cost curve only matters inside Arizona and Texas drayage and distribution for the next year, so price those lanes harder in your 2026 bids and do not let a salesperson fold this headline into a trans-Pacific quote where it does not apply.

Track the empty-repositioning math, because that is the next bottleneck; ask any autonomy provider how they balance empties across your region before you sign, since a fast cab with no trailer in the right place moves nothing. Set a review trigger: revisit this in the first quarter of 2027 when the three-hundred-truck targets land, and only re-plan your domestic network if two or more carriers show live revenue lanes in your corridors, not just announcements.

Spell out the empty-repositioning bottleneck with a number, because it is the part the vehicle folks skip. Say a driverless truck finishes a delivery in Phoenix and the next job sits in Dallas, eight hundred miles away. With a human driver the truck deadheads and the driver is paid to move the empty asset; the cost is real but hidden inside the rate. With no driver the deadhead still happens, the asset still sits idle between jobs, and now nobody is being paid to move it, which sounds like a saving until you notice the truck earned zero revenue for that leg anyway.

The saving was always in utilization, and utilization is gated by how well you match empties to pending freight across the whole network and the whole day. Automate the cab and you still win or lose on that match. A network that cannot balance empties today will not magically balance them with the driver removed; the constraint simply becomes impossible to hide, because there is no person in the seat absorbing the idle iron with their own time.

Look at the risk side as its own queue. Every autonomous mile generates a claim-or-not event, and the cost of that event is priced into the insurance before the truck rolls. If the loss ratio on these lanes stays low, the premium falls and the per-mile case strengthens; if one bad incident hits, the premium jumps and the whole saving can vanish in a single quarter. So the cost curve is not a fixed line you lock in by buying the truck. It is a live variable fed by real-world outcomes, and the autonomy partner is holding it. For the shipper this means the 'saving' is partly someone else's risk position, and you want to know how long they intend to eat it. A vendor quoting a low per-mile rate while carrying the early-loss risk is subsidizing you now and repricing later. Budget for the reprice, because it always comes.

Keep the scope tight when you brief your own team. This number says nothing about ocean freight, nothing about your China origin, nothing about port congestion, nothing about the dray from the port to the distribution center. It says something only about repetitive US domestic lanes in two states. The mistake I expect to see is a logistics provider folding 'driverless trucks' into a trans-Pacific quote as if the saving reaches your container. It does not. The container still crosses the ocean on a ship and clears customs on a clock that autonomy on an Arizona highway cannot touch. So when a salesperson waves this headline, ask which exact lane the saving applies to, and if the answer is not a fixed US domestic run you already ship, the discount is a story, not a number you can put in a budget.

Measure the one thing this whole episode proves matters: lane variance. Pull your last ninety days of US domestic moves and sort them into fixed-rhythm known-dock runs versus everything else. If the first bucket is under a third of your volume, your network is not autonomy-ready and will not be for years, no matter how many truck makers hit their 2027 targets. The work that pays is shrinking the second bucket, standardizing the irregular into the regular, because that is the raw material the machine needs. A shipper who treats this as a vehicle purchase will wait for the industry. A shipper who treats it as a lane-discipline project starts capturing the same cost curve with the trucks they already have, by making their own queue behave before any vendor arrives.

One last boundary to hold, because the utilization story gets oversold. The gain has a ceiling set by the road and the load, not by the absence of a driver. A truck still cannot run more than the legal road hours allow, still needs fuel, still breaks, still waits at a dock that is not ready. Removing the driver lifts the cap from 'one person's day' to 'the vehicle's day,' but the vehicle's day was never infinite. So the famous 'truck runs twenty-four hours' line is partly true and partly marketing: it runs as long as the tires, the fuel, the maintenance clock, and the dock allow. The realistic lift is meaningful but bounded, and a shipper who underwrites a business case on round-the-clock operation will be surprised by the maintenance and the empty repositioning that quietly cap it. Quantify the bound before you count the saving, or the saving will count you.

Carry the queueing lens one step further into your own buying. When you request a domestic quote from a carrier bragging about autonomy, ask them the throughput question rather than the headline question. Not 'do you have driverless trucks' but 'what share of my lane's daily moves are fixed-rhythm known-dock, and what is your empty-to-loaded ratio on that lane.' The first number tells you whether their autonomy even applies to your freight. The second tells you whether they have solved the bottleneck that autonomy exposes. A carrier who answers both without flinching is running a real system. One who redirects to the truck is selling the visible part. You are buying the schedule, whether they say so or not.

The plain version. PepsiCo did something real, and the queueing lesson is the useful one: they removed a constraint by first making the queue behave, then letting the machine run it. The danger is reading 'thirty-five driverless trucks' as 'the whole industry changes tomorrow.' It changes for shippers who can make their lanes as tame as a bottling plant to a Walmart. For everyone else, the bottleneck just moved to the schedule, and the schedule was always the harder problem.

↑ Back to the key points

— By Ravi Chandran