Avride Says It Passed One Million Deliveries. Nobody Can Point to Where It Said So.
Avride's trade-press milestone of one million U.S. robot deliveries has no traceable primary source, the opposite of Serve Robotics' SEC-filed count and Starship Technologies' dated newsroom figures, exposing an industry that has never agreed on what a completed delivery even is.

This week, a trade headline said Avride's delivery robots had crossed one million completed deliveries in the United States. I went looking for the document behind that number, the way I check every figure before it earns a place in the taxonomy I keep by hand. I never found one.
That is not the same as saying the number is wrong. It might be close. The real story is narrower and more useful than a fact-check: nobody, not the outlet that ran it, not Avride's own site, not me, can point to a primary document that actually says "one million," while a far less exciting figure, around 600,000, from the company's own largest shareholder, is the only number in this story anyone can trace to a source.
The biggest number in autonomous delivery this week is also the only one nobody can source.
Where the number was supposed to come from
The report crediting Avride with a million deliveries said the company announced the milestone "in a company blog post," without linking it, naming it, or dating it. Avride's own site has no newsroom or press-release archive I could find, no blog post matching that claim, and no page stating a cumulative delivery count at all. The closest thing to a sourced figure comes from Nebius, Avride's majority backer and former parent, which told shareholders in an August 2026 letter that the robot fleet, then about 500 units, had completed slightly more than 600,000 deliveries since launch.
Four hundred thousand deliveries is not a small gap to close in secret. This week's report also included a run rate: 180,000 deliveries in October alone, close to 7,000 a day. Run that rate backward from August and the math is not implausible. Close to a million by mid-November is a defensible extrapolation. It is just not the same thing as a sourced milestone, and the industry has stopped distinguishing between the two.
A number that had to survive a lawyer
Compare that to Serve Robotics, a publicly traded company that has to file its delivery counts with securities regulators rather than announce them in a blog post. Its most recent investor materials describe the fleet as "nearing 2 million" cumulative deliveries across sidewalk and hospital routes. Nearing. Not "has passed." Not a round number. A public company's disclosure counsel will not let an executive say "two million" in an SEC filing until the ledger actually clears it, because a wrong number in that document is a liability problem, not a headline problem.
That is the whole asymmetry in one comparison. The hedge is the tell. The filing that undersells its own progress is the one you can trust. The press release that rounds up cleanly to a milestone with no footnote is the one nobody has bothered to check.
Even the clean cases don't measure the same thing
Starship Technologies is the honest counterexample here, a company whose own newsroom documents a real, auditable trajectory: more than 10 million deliveries by its April 2026 release, a fleet above 3,000 robots, operations across more than 300 locations in eight countries. That is a number you can actually stand behind, built up publicly in increments from 50,000 to 8 million to 10 million over several years of its own dated press materials.
And yet even the best-documented pair in this category tripped over the same problem. Walmart said on May 29, 2026, that it had passed one million drone deliveries, a count of its own U.S. retail volume. Wing's own newsroom post weeks later cited the identical round number, "more than one million," as a lifetime, global figure that includes its long-running Australian operation. Two organizations, the same headline number, the same general window, and two different populations hiding underneath it. If Walmart and Wing, with real press offices and real dated releases, cannot agree on what a million means, the bar an unsourced Avride claim has to clear is lower than it should be.
Nuro makes the problem stranger still. Having pivoted away from running its own consumer delivery service toward licensing its self-driving stack to automakers, the company no longer reports a delivery count worth citing. Its most recent public figure, around 1.7 million autonomous miles as of April 2026, is not a delivery metric at all. It is a mileage count from an entirely different business the company barely runs anymore, surfacing in the same conversations as if it were comparable.
Nobody wrote the rule that would fix this
I checked whether any standards body actually defines what counts as one completed delivery for a fleet like this. There is no such definition. The closest document, an SAE best-practice guide for automated driving systems, is built to evidence a safety case for a deployment decision, not to count finished jobs. The Open Mobility Foundation's own data specification for delivery robots records whether a trip ended and whether a human, a semi-autonomous system, or a fully autonomous one drove it. It says nothing about whether whatever the robot carried actually reached anyone.
A delivery should be the easiest thing a logistics company can prove. It has instead become the easiest thing to repeat.
No regulator asks these companies to standardize how they count scale, because counting scale is not a safety question. No investor has forced the issue either, since a round, uncheckable number raises a round, uncheckable price just fine. The infrastructure for honest measurement simply was not worth anyone's time to build, and in its absence, a blog post nobody linked gets treated the same as a federal filing.
What I'm actually watching for
None of this means Avride's business is fake. Six hundred thousand verified deliveries by August, growing at a reported pace near 7,000 a day, is a legitimately fast curve, among the quickest in the category whether or not it has touched seven figures yet. What it means is that the industry does not currently reward whoever counts most carefully. It rewards whoever announces soonest, in the roundest number, with the least friction attached.
Running a platform that has to decide, every day, which of these claims belongs in a company's record and which belongs in a correction queue, I have stopped expecting the biggest number in a week's news to be the best-sourced one. The two are, if anything, inversely related: the harder a figure is to verify, the more cleanly it tends to round. The company that eventually publishes an actual, dated, auditable delivery ledger, the way securities law already forces Serve Robotics to and the way Starship's own newsroom bothers to without being forced, will be the one worth taking at its word the next time a milestone lands in an inbox. Until then, the safest assumption for any round delivery number this industry hands you is that it is a plausible estimate wearing the clothing of a fact, and the job of anyone keeping score is to say so before it hardens into one.
This column reflects the author's own analysis and opinion, based on primary company disclosures and public reporting. It is for general information purposes only and does not constitute investment, financial, or legal advice. Hero image: an Avride autonomous delivery robot operating on a residential sidewalk. Photo: Avride.











