Rogue Planets: The Billions of Wandering Worlds Floating Through Interstellar Space

The shortest event in the MOA survey’s nine years of data lasted 0.057 days. Call it an hour and twenty minutes. For that long, a star in the crowded middle of the galaxy got a little brighter and then went back to normal, and that blip is, in its entirety, the evidence for a planet nobody will ever see.^[1]1. T. Sumi et al., Free-Floating Planet Mass Function from MOA-II 9-year Survey towards the Galactic Bulge, AJ (2023). Six events shorter than half a day; the shortest, 0.057 ± 0.016 days. No light from the planet itself. No host star. Just a brief bend in someone else’s starlight, caused by a world drifting between us and it.

Rogue planets are a lovely subject because the romance and the statistics pull in opposite directions. The romance says the galaxy is full of orphaned Earths sailing through the dark. The statistics say: probably, possibly, by a factor of a few in either direction, and we cannot yet tell most of them apart from planets on very wide orbits. I think both are worth taking seriously. But if you only remember one number from this piece, make it the error bar.

§How to lose a planet

Young planetary systems are not the orderly clockwork of a school diagram. They are crowded, and giant planets in them tug on each other and on everything smaller. Run that forward for a few million years and close encounters become likely. Most end with a nudge. Some end with one body flung inward and another thrown out entirely, with more than enough speed to escape its star forever. Planet–planet scattering is a slingshot with no one aiming it.

Our own solar system almost certainly did this. In 1999 David Stevenson pointed out that, during planet formation, some Earth-sized embryos would have either collided with Jupiter or passed close enough to be ejected from the solar system by its gravity.^[2]2. Caltech, Many life-bearing planets could exist in interstellar space, on D. J. Stevenson, “Life-sustaining planets in interstellar space?”, Nature 400, 32 (1999). Some fraction of the raw material that could have been extra rocky planets here is now somewhere out between the stars.

There is a second, quieter way to make a planet-mass object with no star, and it is why “orphan” is not always the right word. Stars form when clouds of gas collapse, and nothing in principle stops a small enough clump from collapsing into something planet-sized on its own. Those objects were never anyone’s children. In 2021 a team led by Núria Miret-Roig found between 70 and 170 free-floating planet candidates in the young Upper Scorpius and Ophiuchus region, up to seven times more than cloud-collapse models predict.^[3]3. N. Miret-Roig et al., A rich population of free-floating planets in the Upper Scorpius young stellar association, Nature Astronomy (2021). Between 70 and 170 candidates, depending on the age assumed for the region. Their reading was that ejection probably contributes about as much as collapse, which in turn means instabilities that throw out giant planets must be common in the first ten million years of a system’s life.

Note how they were found, too, because it quietly undercuts the other half of the hook. Those planets are only a few million years old and still warm from formation, glowing in the infrared. They were imaged directly. “Total darkness” is true of an old rogue in visible light. A young one is a faint ember, and that ember is how we count them.

§Weighing something that doesn’t shine

Old rogues are cold, so direct imaging fails and you need gravitational microlensing. The idea comes straight from general relativity: mass bends light, so an object passing almost exactly in front of a background star acts like a lens and briefly magnifies it. The size of the effect is set by the Einstein radius,

where $M$ is the lens mass and $D_L$, $D_S$ are the distances to lens and source. The event lasts as long as it takes the lens to drift across that angle, so its duration scales with $\sqrt{M}$. That square root does a lot of work. A Jupiter-mass lens produces events lasting a day or two. Earth is 318 times lighter, so divide by about 18 and an Earth-mass lens gives you a couple of hours. That is why surveys like OGLE and MOA point at the galactic bulge, where the density of background stars is highest, and sample it many times a night. A planet’s whole signal can come and go between two coffee breaks.

The catch is that a single light curve cannot separate mass from distance and speed. A light, nearby object and a heavier, more distant one moving differently can produce the same blip. For years every free-floating planet mass was therefore a statistical inference, not a measurement. That changed in January 2026, when a team led by Subo Dong published an event seen from the ground and from Gaia, more than a million kilometres from Earth. Two vantage points see the lensing at slightly different times, which is a parallax, which breaks the degeneracy. The object came out at about 0.22 Jupiter masses, roughly Saturn.^[4]4. S. Dong et al., A free-floating-planet microlensing event caused by a Saturn-mass object, Science (2026). The mass is 0.219 (+0.075/−0.046) Jupiter masses, and the authors are careful to say “unbound or on a very wide orbit.” It is the first rogue planet anyone has actually weighed. The paper still says unbound or on a very wide orbit, because microlensing cannot rule out a star far enough away that it never shows up in the light curve.

§The real number has an error bar you could drive a galaxy through

So, how many? The best-known figure comes from Sumi and colleagues in 2023: 21 free-floating or very-wide-orbit planets per star, with an uncertainty of +23 and −13, for masses between a third of an Earth and about twenty Jupiters.^[1]1. T. Sumi et al., Free-Floating Planet Mass Function from MOA-II 9-year Survey towards the Galactic Bulge, AJ (2023). Six events shorter than half a day; the shortest, 0.057 ± 0.016 days. Put plainly, the range runs from about 8 to 44 per star. The same analysis has them outnumbering planets in wide bound orbits by about nineteen to one, with most of the count at the low-mass end.

Now multiply. The Milky Way’s star count is usually quoted in the low hundreds of billions. Take 200 billion as a round number and Sumi’s range gives somewhere roughly 1.6 to 9 trillion objects. So the title is, if anything, too modest — provided you remember that the extrapolation rests on a few short-timescale events in one direction of the sky, and a mass function fitted through them.

The history here is a good reason for humility. In 2011, an earlier MOA analysis suggested Jupiter-mass rogues were nearly twice as common as stars, which made headlines. In 2017 Przemek Mróz and the OGLE team went back with a sample about six times larger, found no excess of the one-to-two-day events that such a population would produce, and set a 95% upper limit of 0.25 Jupiter-mass rogues per star.^[5]5. P. Mróz et al., No large population of unbound or wide-orbit Jupiter-mass planets, Nature 548, 183 (2017). The 95% upper limit is 0.25 such planets per main-sequence star. The dramatic number did not survive more data. What did survive is the hint of very short events, the Earth-mass ones, and that is what the current estimates are built on.

The same pattern played out with JWST in Orion. In 2023 JWST images turned up forty apparent Jupiter-mass binary objects, pairs of planets orbiting each other with no star at all, which would be very hard to explain by ejection. Within two years the claim was under heavy fire: one reanalysis argued many were background objects, and simulations argued such wide pairs could not form in the first place. Nobody yet knows what the forty double dots are.^[6]6. Leiden University, Large double planets without a star don’t actually exist (July 2025), on S. Portegies Zwart and E. Hochart, “Why wide Jupiter-mass binary objects cannot form,” Nature Astronomy (2025). Kevin Luhman’s 2024 reanalysis had separately suggested many of the pairs are background sources. I don’t have a horse in this race. I just notice that in this field, the extraordinary per-star numbers keep shrinking on contact, and the boring short-blip ones keep holding.

§An ocean with no sun

The question everyone actually wants answered is whether any of these places could hold liquid water. The surprising answer from the theorists is: plausibly yes, without any sunlight at all.

Stevenson’s 1999 argument goes like this. An Earth-sized embryo ejected early might keep a thick envelope of the primordial hydrogen it formed in. Without a star, its internal radioactive heat alone would leave it radiating like a body at about 30 kelvin. But molecular hydrogen under enough pressure absorbs far-infrared light, so a deep enough hydrogen blanket traps that trickle of heat the way a greenhouse traps sunlight, only more effectively. Under it, the surface could sit near Earth-like temperatures, with oceans at pressures like those at the bottom of ours. He estimated a body would need at least about half an Earth mass to hold such an atmosphere.^[2]2. Caltech, Many life-bearing planets could exist in interstellar space, on D. J. Stevenson, “Life-sustaining planets in interstellar space?”, Nature 400, 32 (1999).

The other route is the one we already know works in our solar system: a lid of ice. Dorian Abbot and Eric Switzer worked out in 2011 what it would take for geothermal heat alone to keep an ocean liquid under a frozen shell on a rogue planet. With Earth-like composition, about 3.5 Earth masses. With ten times Earth’s water fraction or a thick insulating atmosphere, as little as 0.3.^[7]7. D. S. Abbot and E. R. Switzer, The Steppenwolf: A proposal for a habitable planet in interstellar space, ApJL 735, L27 (2011). The ice is the insulation. The rock’s own decaying uranium, thorium and potassium are the furnace.

My honest take is that this is good physics and zero data. Both models are self-consistent; neither has an observed example, and microlensing will never give us one, because a lensing event tells you a mass and maybe a distance and then the object is gone forever. Abbot and Switzer note you could only characterise such a world if one wandered within about a thousand AU of the Sun. The interesting implication is the one about life’s energy budget. Every habitable-zone calculation assumes the energy comes from a star. These models say a planet can carry its own, slowly, for billions of years, as long as it is wrapped up well enough. That is not evidence that anything lives out there. It is a reminder that “the habitable zone” is a statement about one energy source, not about habitability.

§Roman is about to replace the guesswork with a census

The Nancy Grace Roman Space Telescope lifted off on a Falcon Heavy on 30 August 2026, bound for the Sun–Earth L2 point, with science operations expected to begin in early 2027.^[8]8. The Planetary Society, The Nancy Grace Roman Space Telescope has launched (August 2026): Falcon Heavy from Kennedy, bound for L2, science operations expected from early 2027 after about 90 days of commissioning. Among other things, it will stare at the galactic bulge from space for months at a stretch, without weather, daylight, or the atmosphere smearing the stars together, and catch microlensing events that ground surveys miss.

The forecasts are specific enough to be falsified, which is the best thing a forecast can be. Sumi’s mass function predicts Roman should find about 988 free-floating planets down to the mass of Mars, with the error bar again running from roughly 400 to about 2,800. A 2026 study by DeRocco and colleagues expects “several hundred” detections, enough to improve abundance estimates by orders of magnitude below Earth’s mass and tell competing mass distributions apart.^[9]9. W. DeRocco et al., Prospects for Reconstructing the Free-floating Planet Mass Function at the Population Level with the Nancy Grace Roman Space Telescope, AJ 172, 7 (2026). If the census lands anywhere near the forecast, most of the galaxy’s planets don’t have suns. If it comes in far below, that is the sentence to retract.

Either answer changes how ordinary our arrangement looks. We tend to picture planets as things that belong to stars, because every planet we have ever stood near does. It may turn out that the typical planet in the Milky Way is small, cold, a little warm inside, and alone, and that a world circling a star in a stable orbit is the exception that happened to produce someone to ask.