GMA's "Deals & Steals": A Data-Driven Look at the So-Called 'Savings'

The Silent Fleet: Why Are Self-Driving Trucks Vanishing from Public Roads?

Just a few years ago, the narrative was set in stone. The highways of tomorrow would be a seamless ballet of autonomous 18-wheelers, drafting in perfect, fuel-saving platoons. Companies with futuristic names like Aurora, TuSimple, and Waymo Via were raising capital in the billions, promising a revolution in logistics that was always just around the next quarter. We were shown sleek promotional videos of trucks navigating desert highways, their cabins conspicuously empty. The future wasn't just coming; it was already barrelling down the interstate in the fast lane.

And then, a strange quiet descended.

The constant drumbeat of press releases slowed to a trickle. Pilot programs that were once the darlings of tech media have been quietly "concluded" or scaled back. For the average person, the odds of spotting a truly driverless semi on a public highway remain statistically indistinguishable from zero. The silent fleet we were promised seems to have become just that: silent. Vanished.

This isn't the explosive, headline-grabbing collapse of a dot-com bust. It’s a slow, creeping fade from public view, a retreat from the front lines of innovation back into the cloistered world of R&D labs and closed test tracks. The capital hasn't entirely dried up, but the ambition seems to have been… recalibrated. So, where did the trucks go? The data suggests this isn't a temporary pit stop. It’s a fundamental rerouting, driven by a brutal collision with economic and physical reality.

The Discrepancy Between Projections and Pavement

Let’s first establish the baseline. The investor decks from the 2020-2021 boom were a masterclass in aggressive forecasting. We saw hockey-stick graphs projecting tens of thousands of autonomous trucks in operation by 2025, generating billions in high-margin, recurring revenue. One prominent player’s SPAC presentation (with a valuation north of $10 billion) was predicated on a 2024 commercial launch at scale. The story sold was one of inevitability and imminence.

The pavement tells a different story. Instead of mass deployment, we've seen mass attrition. TuSimple, once a leader, has imploded amid internal turmoil and a delisting. Other major players have pivoted their public messaging, shifting from bold declarations of "full autonomy" to the much more sedate language of "enhanced driver-assist" and "hub-to-hub" models. These hub-to-hub routes are the industry's big concession; they are essentially milk runs, predictable and repeatable trips between two fixed points on a map, often on sun-drenched, low-traffic stretches of interstate. It's a far cry from the original vision of a truck that could handle a complex delivery from a warehouse in New Jersey to a distribution center in downtown Los Angeles.

I've looked at hundreds of these corporate filings and earnings call transcripts, and the shift in linguistic tone is the most telling data point. The confidence has been replaced by caution. The swagger has been replaced by caveats. The talk of replacing drivers has been supplanted by talk of "augmenting" them. This is the part of the analysis that I find genuinely puzzling: Why the public reticence? If the tech is progressing, even slowly, wouldn't you want to maintain investor confidence with steady updates? The silence itself is a signal. It suggests the problems being encountered are not the small, iterative ones you can fix with a software patch. They are foundational.

This retreat from public highways isn't just about bad press from the occasional accident. It’s about unit economics. Every mile an autonomous truck drives with a human safety engineer in the cab (and often a chase car full of more engineers) is a capital-incinerating loss. The industry has spent billions to replicate the capabilities of a human driver who gets paid about 65 cents a mile—to be more exact, the national average is currently around 69.4 cents per mile. After years of testing, the return on that investment is still stubbornly, infuriatingly negative. So, what happens when the math doesn't work? You take the experiment off the expensive public balance sheet and back to the lab.

The Infinite Archipelago of Edge Cases

The core of the problem, the reason for the vanishing act, is what engineers call "edge cases." This benign-sounding term is, in my view, the single greatest poison pill to the dream of full autonomy. The industry sold the problem of driving as a 99% solvable equation. Get the truck to stay in its lane, manage its speed, and avoid obvious obstacles, and you're most of the way there. This turned out to be a catastrophic miscalculation.

The reality is that the last 1% of the problem represents 99% of the risk and complexity. Driving isn’t one big problem; it’s an infinite series of small, unpredictable ones. I think of it less like climbing a single, massive mountain and more like navigating an uncharted archipelago. The sunny, straight highway is the main island—easy to conquer. But the real world is the thousands of other islands: a tattered piece of a blown tire lying in the lane that looks like a shadow; a state trooper unexpectedly waving traffic through a makeshift construction zone; the glare of a low winter sun bouncing off wet pavement, completely blinding the truck’s optical sensors.

Each of these "islands" is an edge case. You can program a solution for one, but then another, entirely novel one appears. Imagine the truck’s AI, humming along perfectly for 500 miles, suddenly confronted with a sheet of plywood falling off a contractor’s pickup. Its training data has never seen that exact scenario. What’s the protocol? A human driver makes a split-second, intuitive judgment based on a lifetime of chaotic, real-world experience. The AI has to reference a decision tree based only on what it’s been explicitly taught. The cost of a miscalculation isn’t a bug report; it’s a multi-ton catastrophe.

This is where the entire model begins to buckle. How do you gather enough data to account for a near-infinite number of variables? The industry’s reliance on "disengagement reports"—where a safety driver has to take over—is a deeply flawed metric. It only measures outright failures. It doesn’t measure the countless "near-misses" or moments of hesitation where the system almost got it wrong. What is the true error rate when you account for these silent failures? And how can you possibly underwrite the insurance for a vehicle whose failure mode is so unpredictable? The "vanishing" isn't a sign of technological failure. It's a sign of a rational, risk-averse retreat from an economically and legally untenable position.

The Data Points to a Detour, Not a Destination

Let's be precise. The dream of the fully autonomous, coast-to-coast robot trucker isn't dead, but it has been placed in a medically induced coma. The silence we’re now observing on public roads isn't a mystery; it’s the sound of a market correction. The capital is quietly being reallocated from the moonshot of Level 5 autonomy to more terrestrial, profitable applications like advanced driver-assistance systems (ADAS), remote-controlled yard operations, and geofenced, short-haul routes. The technology hasn't vanished. It has simply retreated to the specific, contained environments where the math actually works. The silent fleet isn't coming to a highway near you anytime soon, because the data is clear: for now, the open road is a problem too big, too random, and too expensive to solve.