Midnight does not arrive with a sound. Usually.
In a server room chilled to sixty-four degrees, the air smells faintly of warm dust and ozone, and the only sound worth hearing is the low, steady hum of cooling fans moving heat away from thousands of stacked blades. A wall clock—an ordinary quartz relic kept on the breakroom wall for people who still like hands—ticks along in its slow, predictable march. Nobody in that room watches the wall clock when something goes wrong. They watch logs scroll past in pale green and amber text, line after line, millisecond after millisecond, waiting to see if the machines agree on what now means. For another look, read: this related article.
I have spent enough years around distributed systems and real-time software to know that dread does not look like a movie explosion. Dread looks like a graph that suddenly turns into a jagged sawtooth while a pager buzzes on a nightstand at three in the morning. You rub your eyes. You reach for your glasses. You stare at a timestamp that claims an event happened before its cause, or that a cluster of servers spent six seconds frozen in a quiet, agonizing argument with itself over which clock was telling the truth.
For fifty years, human civilization lived with a strange little patch stitched onto the fabric of time: the leap second. Every now and then—twenty-seven times between 1972 and 2016—timekeepers around the world slipped an extra second onto the clock, holding the minute hand at 23:59:60 for a breath before letting midnight roll over. It was a gentle, poetic compromise between two different ways of counting the day. Then, in late 2022, delegates at the General Conference on Weights and Measures in Versailles voted to retire the leap second by 2035, setting a path toward allowing a wider drift between atomic time and solar time. Similar insight regarding this has been provided by The Verge.
On paper, that sounds like the driest bureaucratic footnote imaginable. A tiny adjustment. A second here or there. Something only physicists, astronomers, and grumpy systems engineers care about.
They are wrong. You feel it every time you tap a card, book a flight, send a message across an ocean, or trust a screen to tell you when something happened.
Two Clocks in the Dark
To understand why a single second could bring modern infrastructure to its knees, we first have to admit how confusing time actually is once you stop looking at the sun. Honestly, even after reading dozens of papers on metrology and orbital mechanics, my brain still pushes back a little at the sheer strangeness of it. How can a second—the thing we use to time a boiling egg or a sprint—be both a heartbeat of the universe and a moving target?
Let's use a metaphor here to keep our footing. Picture two runners on a track. One runner is a robotic pacemaker set to a metronome so steady it barely wavers by a fraction of a second over millions of years. The other runner is Earth itself: a four-and-a-half-billion-ton ball of rock, molten iron, sloshing oceans, shifting glaciers, and swirling winds, spinning through space while tugged by the Moon and Sun.
The robotic runner is International Atomic Time, or TAI. Since 1967, the international definition of the SI second has been anchored to the vibrations of cesium-133 atoms—specifically, 9,192,631,770 cycles of radiation corresponding to a transition between two hyperfine levels of the ground state. That is not a guess. That is a standard of astonishing precision, the foundation of Coordinated Universal Time, or UTC, which governs our civil clocks and digital networks.
The spinning planet is UT1, solar time tied to Earth's actual rotation relative to distant celestial objects. A solar day is supposed to be 86,400 seconds long. Except Earth is not a rigid, frictionless top. Tidal friction from the Moon gradually slows Earth's rotation over centuries, while earthquakes, seasonal shifts in atmosphere and water mass, and changes in the core make the planet speed up or drag its feet by tiny fractions of a millisecond from day to day.
Because the spinning runner falls slightly behind the atomic metronome over time, Coordinated Universal Time began adding leap seconds in 1972 so UTC would stay within 0.9 seconds of UT1. Keep the atomic ticks pure. Every so often, insert one extra second so the clock on the wall does not drift too far out of step with the turning Earth.
Simple. Elegant. Tidy.
Until the machines got faster than our patch.
When One Second Breaks the Machine
Here is a hypothetical scenario to make the stakes concrete, drawn from real failure modes that engineers have watched play out over the past two decades.
Imagine a character I will call Nadia, an on-call reliability engineer living in Toronto. Nadia is not a physicist; she is someone who likes cold brew coffee, terrible sci-fi novels, and weekends that stay unbothered. At 7:59:58 PM local time on a leap-second night, she is sitting on her couch while a financial clearing platform, a cloud storage service, and a hospital telemetry network hum quietly across three continents.
At 23:59:59 UTC, the leap second arrives.
To a human, one extra second is nothing. You can barely finish taking a sip of water. To a distributed computer cluster handling hundreds of thousands of requests per second, one second is an eternity filled with millions of tiny decisions. Worse, different operating systems and software stacks handle that extra second in wildly different ways. Some repeat the 59th second. Some smear the extra second across the preceding hours using a technique called "leap smear," nudging clocks slightly slower so there is no sudden jump. Some simply crash, panic, or log duplicate timestamps that confuse databases into thinking an update happened backward in time.
In Nadia's hypothetical night, two nodes in a distributed database disagree by a fraction of a second over whether a lock expired before or after a write completed. A queue backs up. Retry storms bloom like sparks in dry grass. Within minutes, latency spikes across services that have nothing to do with banking—a logistics tracker stalls, a dispatch board flickers, an alert fires on Nadia's phone.
Silence. Then buzz. Then buzz again.
Nadia sets her mug down on the coffee table. Her stomach tightens in that familiar, hollow way it always does when an outage starts with a riddle instead of a smoking gun. Nothing is on fire. No cable was cut. Yet millions of dollars of transactions or critical status updates are stuck in limbo because two pieces of software could not agree on what second it was.
We have seen echoes of this in the real world. In June 2012, a leap second triggered high-profile software glitches across major websites and operating systems, including Linux-based servers, causing spikes in CPU usage and service disruptions. On New Year's Eve in 2016, another leap second arrived, and engineers braced with months of preparation, custom patches, and sleepless shifts just to survive a single extra tick of the clock.
Think about the absurdity of that. We built fiber-optic cables across ocean floors. We bounced signals off satellites. We engineered processors with billions of transistors that execute instructions in nanoseconds. Yet every few years, the entire digital edifice had to hold its breath because a single extra second was being shoved into the calendar like an uninvited guest at a crowded table.
And then came the plot twist that caught even some seasoned observers off guard: Earth briefly started spinning faster in recent years, making days fractionally shorter than 86,400 seconds and prompting scientists to talk seriously about the possibility of a negative leap second—removing a second rather than adding one. Nobody had ever tested a negative leap second in production systems. The risk of breaking critical software was simply too high.
Why Now, and Why 2035?
So why cancel the leap second now, and why wait until 2035?
Let's look at the real-world tension behind the decision. On one side stand astronomers, geodesists, and navigators who care deeply about where Earth actually is in space relative to the stars and the sun. If you point a telescope or track a spacecraft or study Earth's rotation over decades, you want solar time and atomic time to stay close together. Without leap seconds, the gap between TAI and UT1 will widen slowly over time. Astronomers estimated it would take roughly fifty to a hundred years—or even longer depending on rotation trends—for the difference to reach about a minute, though larger drifts could accumulate over centuries.
On the other side stand telecom networks, satellite constellations, financial exchanges, power grids, and global navigation satellite systems like GPS. For these systems, uninterrupted, continuous time is life or death. A sudden jump in UTC can throw off synchronization protocols, disrupt high-frequency trading timestamps, interfere with telecommunications handoffs, and create cascading bugs across interdependent software.
The decision reached at the 2022 General Conference on Weights and Measures in Versailles—building on years of debate within the International Telecommunication Union and the Bureau International des Poids et Mesures—chose continuity for modern technology while promising a review by 2035 to determine how the maximum allowable difference between UT1 and UTC should be increased or managed.
Notice what that timeline tells us. Canceling the leap second is not a magic wand that makes the problem vanish overnight. It is a transition—a ten-plus-year runway to untangle decades of assumptions baked into code, standards, and hardware.
Consider what happens next: if we stop inserting leap seconds, Coordinated Universal Time stops being yanked back into lockstep with Earth's rotation every few years. Instead, the drift between atomic time and solar time is allowed to grow wider than the old 0.9-second limit. Eventually, if left completely untouched for centuries, noon by the atomic clock and noon by the sun would drift apart—not tomorrow, not next year, not even in Nadia's lifetime, yet real enough over deep time that future generations will have to decide how to reconcile the heavens with the servers.
It is easy to feel a little dizzy thinking about that. Most of us live in a world of convenient illusions: that a minute is always sixty seconds, that a day is always twenty-four hours, that the clock on our phone is simply right. Admitting that time itself requires negotiation between atoms, planets, and committees can feel unsettling.
Yet there is something oddly grounding in that uncertainty. Time is not a ruler nailed to the wall of the universe. It is a measurement we invent, refine, and argue over so we can meet at the same corner, ship goods on schedule, and keep lights glowing in the dark.
The Human Cost of Invisible Friction
Why should anyone outside a physics lab or a data center care about how atomic time and solar time get along?
Because invisible friction always finds a way to become human friction.
When a payment system stumbles during a synchronization glitch, a small business owner whose card terminal freezes during a Friday dinner rush does not care about cesium-133 hyperfine transitions. They care that a line of hungry customers is staring at them while a screen spins a loading wheel. When a hospital alert system or dispatch network experiences timing drift or queue stalls, the stakes are measured in minutes of human anxiety, not lines of log output.
Even in quieter corners of daily life, accurate, agreed-upon time is the quiet scaffolding of trust. Contracts rely on timestamps. Aviation and maritime navigation depend on precise timing signals. Power grid operators balance supply and demand across vast regions where milliseconds of phase and timing matter. When our timekeeping standard fights our software architecture, we pay a hidden tax in complexity, brittle workarounds, and engineering exhaustion.
Retiring the leap second removes one sharp, recurring shock from that system. Instead of forcing every server, router, and database on Earth to survive a sudden extra second on the same exact tick, timekeepers are moving toward allowing a broader tolerance before any adjustment—potentially a larger step like a leap minute in the distant future, if ever needed—while giving engineers years to prepare.
Still, let's be honest: removing one source of trouble never means zero trouble. Software has a long memory. Somewhere in a bank, an airline, a factory, or a telecom tower right now, there is almost certainly a piece of legacy code written twenty years ago that assumes a specific rule about leap seconds or clock offsets. Finding and fixing those assumptions before 2035 is real work, done by real people squinting at documentation at midnight.
What Remains When the Patch Comes Off
Think back to Nadia in our hypothetical scenario, sitting on her couch with a cooling mug of coffee. Imagine the future where the leap second is gone, where the calendar rolls smoothly from :59 to :00 without that awkward extra beat, and where the machines no longer trip over an inserted second every couple of years.
The relief is quiet. There is no parade when a bug does not happen. No headline celebrates a midnight that passed without a pager going off.
Yet beneath that quiet relief sits a deeper question that will never quite go away, no matter how clean our code becomes. We can measure atoms with staggering precision. We can build clocks that would barely lose a second over the age of the universe.
Still, we live on a restless planet—one whose oceans pull, whose crust shifts, whose winds push against the spin—turning faithfully beneath a sky that does not care about our spreadsheets.
The extra second is going away. The planet keeps turning anyway.