I remember being transfixed, watching the Beijing Olympics opening ceremony in 2008. Two thousand and eight drummers struck in perfect unison, like robots, and a scroll of ink and light unrolled across the stadium floor. It was clear to the world watching that China had arrived on the global stage and was capable of building a new future with speed and scale. Eighteen years later, China gave the world a preview of another future, a world in which the 2008 drummers could be slowly replaced with machines that can run faster than Usain Bolt.
On August 22, 2026, in that same city’s National Speed Skating Oval, built for the 2022 Winter Olympics, more than 2,000 humanoid robots, competed across 51 events and over 1,300 individual contests: running, soccer, table tennis, boxing, ballroom dancing. A robot built by the Tianzhuo team ran the 100 meters in 9.39 seconds on opening day, edging past Usain Bolt’s 2009 human world record of 9.58. That mark didn’t survive the week. Three days later, X-Humanoid’s Tiangong Ultra ran it in 8.86 seconds. The following day, the same machine ran it again in 8.64 seconds, breaking its own record for the second straight day as the Games closed, sparks flying from its waist as it slammed into the finish barrier. A year earlier, at the first edition of these Games, the winning time had been 21.50 seconds. In twelve months, the record fell by nearly thirteen seconds, a rate of improvement no human sport has ever produced, because no human sport is manufactured.
That same week, on our side of the pond, Elon Musk stood at the G20 Innovation Ministerial and predicted the future. Within ten years, he said, robot manufacturing and advances in AI software will produce a billion humanoid robots. This prediction was modest compared to his earlier predictions. At a Saudi investment conference in 2024, he predicted ten billion humanoid robots by 2040, more than there are humans on Earth, each costing $20,000 to $25,000, and each worth several humans’ worth of output. Morgan Stanley’s more conservative estimate predicts something like eight million humanoid robots in the U.S. by 2040, concentrated in healthcare, manufacturing, and logistics. The gap between Musk’s number and Morgan Stanley’s is either a rounding error or a species event.
Regardless, these are not just predictions, but provocations, claims about the future that are impossible to fact-check as they are parables to be unpacked.
READ: The Review by Ajay Raju | Hedonic Treadmill Paradox: Why we remain unimpressed with technological miracles (August 31, 2026)
The Sprint as a Parable
Watch the footage of that finish line again, the sparks, the small fire the officials rushed to put out. What we watched was not really a race, but a demonstration of compounding. A record that took human beings roughly a century to shave down by less than a second fell by thirteen seconds in the span of a single calendar year, because a machine’s limits are not a body’s limits, but a manufacturing problem. And manufacturing problems get solved by scale, capital, and iteration in a way that human physiology simply does not. To be clear, the same Games that produced those records also produced robots that tripped, collapsed mid-stride, caught fire on camera. The technology is not uniformly superhuman. Not yet, but the pattern is clear.
A machine which learns to sprint this year will learn to lift, to fold laundry, to change a bedpan, to clear rubble, to walk a warehouse floor, at a similarly compounding rate. And, within a decade, the humanoid robot will stop being a stadium spectacle and become a category of labor, the way the automobile stopped being a fairground novelty and became a category of transportation within a comparably short span.
Two Doors
Whether we have a hundred million robots instead of a billion doesn’t matter because these robots will build a different world. I think there are, broadly, two doors, and the machines walking through them may look identical. That is what unsettles me more than the sprint times.
Through the first door is the world Musk describes when he uses the word “abundance.” A robot in a warehouse doing the lifting that wrecks a human spine by fifty. A robot doing overnight home care so an aging population doesn’t have to choose between an institution and a family member’s exhaustion. A robot on a disaster site pulling rubble at 3 a.m. without needing rescue itself. A robot doing the harvest labor, the mine clearing, the deep-sea and off-world work that was always too dangerous or too remote for a human body to risk. In this world, a machine that runs a 100 meters in under nine seconds is not a curiosity, but a search-and-rescue unit that reaches a collapsed building before the window closes. The drudgery peels away from human life the way it did when the tractor replaced the ox.
READ: The Review by Ajay Raju | My city elders don’t trust Bitcoin (August 24, 2026)
The second door is the world of RoboCop and Terminator. Two thousand robots marching into a stadium in formation is a kind of readiness. If a billion of these machines exist, they are also, by definition, deployable at a warehouse or a checkpoint. My fear here is not that a machine will turn violent on its own initiative. Machines don’t have initiative; they have owners. The fear is that a robot built to lift a fallen patient is mechanically indistinguishable from a robot built to restrain a person the state has decided should be restrained, and that the difference between the two futures is not hardware, but governance. Who audits the firmware, who controls the update that ships at 2 a.m. to every unit in a metro area simultaneously, and who decides whether anyone outside a small circle of engineers and executives can see it happening at all? These are supply-chain and cybersecurity questions. Robots, already being sold for warehouses and homes, have shipped, in at least a few documented cases, with data channels nobody disclosed.
The RoboCop movies were written as warnings, not blueprints. But warnings become a footnote once the underlying technology stops being fiction. We are watching that transition happen in real time, at a pace that may outstrip our ability to legislate it.
What the Sprint Actually Measures
Back to that 100-meter record. Nobody in that stadium was measuring danger. They were measuring speed, because speed is legible, timeable, printable in a headline the next morning. But what about an equivalent scoreboard for the more pressing questions? How transparent is the firmware, who can audit it, what happens when a robot workforce that size loses power for six hours, who is liable when the fold-laundry robot and the restrain-a-person robot turn out to share ninety percent of their parts list? We are extremely good at building instruments that measure what’s easy to measure. We are much slower at building the instruments, legal, civic, technical, that measure what’s hard to measure and matters more.
READ: The Review by Ajay Raju | Real life ‘Widow’s Bay’: Wheelbarrowing frogs to safety (August 17, 2026)
That asymmetry will be the actual story of the robot decade, more than the sprint time or the billion-robot number. The engineering curve is compounding in public, on a track, with a stopwatch everyone can read. The governance curve, the rules about audits, ownership, exportability, kill switches, who is allowed to push an update to a machine standing in your kitchen, is compounding much more quietly, in agency filings and closed-door standards meetings, on a track almost nobody is watching, because it produces no equivalent of sparks flying off a finish line.
The Decade Ahead
A world with a hundred million well-governed robots, transparent in their firmware and narrowly scoped in their function, is a genuinely better world than the one we have now. Imagine more hands for the harvest, more eyes for the disaster site, more patience for the 3 a.m. shift nobody wants to work. But a world with a hundred million robots nobody can audit is a worse one, regardless of how fast they can run.
At the 2008 Beijing opening ceremony, the drummers moved as one, a demonstration of China’s capabilities to build and scale a new future. We are at a similar moment now, watching a preview we don’t yet know how to read. A sprint record, a billion-robot prediction, a small fire at a finish line that the officials put out and everyone moved on from within the hour. The machines will keep getting faster. That part is certain, but will they end up lifting the fallen or enforcing the peace? We have less than ten years to answer this question.