Why a Leap Hour Is Coming to Save (or Break) Global Infrastructure

The Clock Is Ticking Earlier Than Expected

Japan isn’t asking the world to change how it tells time. It’s asking whether we should keep telling it at all.

At the International Bureau of Weights and Measures (BIPM), Patrizia Tavella sits on the time department’s steering committee. She has been calculating that there is a 30% chance the world will need a “negative leap second” by 2035—a second subtracted from Coordinated Universal Time (UTC) to realign atomic clocks with Earth’s spinning axis. That number terrifies the infrastructure operators she consults. When she asked whether a 10% risk of failure was acceptable, the answer came back uniformly: no.

So the proposal on the table is not incremental. It would replace the leap second with a leap hour—a single-hour correction, scheduled less frequently, giving system operators years instead of weeks to prepare.

The vote lands at the next General Conference on Weights and Measures in October. If passed, the change takes effect around 2029, which is earlier than the 2035 deadline the 2022 resolution set. Japan is pushing the acceleration. The reason is simple: Earth is spinning faster, and the gap between atomic time and solar time is closing in a direction nobody comfortable with the current architecture wants to face.

The Rotation Problem

A sidereal day—the time Earth actually takes to rotate once—averages 86,400 seconds. It hasn’t done that for centuries. The Moon’s tidal pull slows rotation by roughly 1.7 milliseconds per century. But superimposed on that long drift are shorter-term wobbles driven by the position of the Sun, the Moon’s orbit, core-mantle interactions, and possibly melting ice sheets shifting mass toward the equator.

In 2020, the planet began spinning faster than it had at any point since measurements started. By July 4, 2024, Earth completed a rotation 1.66 milliseconds faster than the nominal day. In the following year, several additional sub-86,400-second days followed. The trend has not reversed.

The effect is tiny in isolation. The problem is compounding: atomic time (TAI) has accrued 37 leap seconds above UTC since 1972, all positive. The system is calibrated to add seconds, never subtract them. A negative leap second would require every system that reads UTC to handle a 23-hour day. Most were not built for that.

Why Negative Leap Seconds Are Worse Than Positive Ones

A positive leap second inserts an extra second. Systems that expect exactly 86,400 seconds per day encounter an unexpected value. Cloudflare’s DNS outage on January 1, 2017, traced back to a leap-second-related bug, disrupted traffic for hours. Not catastrophic, but expensive and humiliating for a company whose selling point is uptime.

A negative leap second is worse. It means deleting a second from the calendar. Any system that counts seconds linearly—and nearly all network time protocols do—would see time skip backward. Clocks would read 23:59:58, then 00:00:00. Any transaction, log entry, or cryptographic nonce anchored to that missing second becomes ambiguous. Replication conflicts appear. Audit trails break. Scheduling systems reorder events.

The BIPM’s own risk assessment, cited by Tavella, treats negative leap second exposure as non-trivial. The consulting firms, financial infrastructure operators, and telecom engineers she contacted all said a 10% probability of the event was too high to accept. That consensus is what is driving the leap-hour discussion forward out of its 2035 timeline.

What a Leap Hour Actually Means

A leap hour inserts or removes 3,600 seconds at once. It would be far less frequent—likely once every few decades rather than irregularly every 18 months or so. The longer interval gives operators time to patch, retest, and rehearse. The larger adjustment reduces the chance of overlapping corrections and short-notice emergency scheduling that currently defines leap second preparation.

The trade-off is immediate. A leap hour creates a day with 25 hours (or 23, if subtracted). Every calendar system, every scheduling database, every financial trading window, every broadcast timetable that assumes a fixed 24-hour day must handle it. The difference is that it would be announced well in advance, rather than sprung weeks before implementation.

The leap hour is not a new idea. It was discussed in the 1970s when leap seconds were first standardized. It was set aside because operators feared that even a planned hour-long jump would disrupt systems. The logic now is inverted: the uncertainty of an unpredictable leap second is considered more dangerous than the disruption of a scheduled leap hour.

Who Wins and Who Loses

The winners from a leap-hour framework are the infrastructure operators who currently operate on panic cycles. Telecommunications companies, clearing houses, and cloud providers would gain predictability. They could schedule maintenance windows, update firmware, and run parallel tests with known lead times.

The losers are the systems that have already coded around the leap second and assume no further adjustment is possible. Some financial platforms treat a leap second as a hard boundary for trade settlement. Some broadcast and streaming systems embed UTC directly into content timelines. A leap hour breaks those assumptions outright.

Japan’s push matters globally because timekeeping is not a national project. UTC is defined by the BIPM, maintained by the International Earth Rotation and Reference Systems Service (IERS), and adopted by treaty across 193 member states. A change approved in Sèvres reverberates through every GPS receiver, every NTP server, every atomic clock disciplined to UTC. If Japan shifts the conversation, it shifts the architecture for the entire digital world.

The Infrastructure That Still Isn’t Ready

The leap second of 2016 was preceded by months of warnings. Linux patched its kernel. Windows updated its drivers. NTP daemon configurations were adjusted. Cloudflare published a postmortem. Yet the outage still happened.

A leap hour would be orders of magnitude harder to test. It affects every system that touches time, not just those that implement NTP. Flight schedules, shipping manifests, power grid synchronization, satellite maneuver planning, medical device logging, legal contract timestamps—all of these sit on UTC or derive from it. None of them have been stress-tested for a 25-hour day.

The 2022 resolution assumed operators would have until 2035 to adapt. The rotational data suggests that deadline is generous. If the Earth continues to speed up, the window between the vote and the first required correction narrows to less than a decade. Tavella’s 30% risk estimate is not a prediction; it is a probability range that already exceeds what the industry considers acceptable.

The Real Stakes

The leap-hour debate is not really about time. It is about whether the world’s critical systems can tolerate irregular corrections from a natural source that cannot be predicted more than a few years ahead. The alternative—letting UTC drift from solar time until the offset grows to minutes—is politically and culturally unacceptable. Noon would stop meaning what it meant for millennia.

A leap hour is a compromise between two intolerable outcomes: chaos from unpredictability, and chaos from abandonment. It preserves UTC’s link to the Sun while giving operators a horizon they can plan for.

The vote in October will decide which kind of chaos gets managed next. The Earth will keep spinning regardless.