Few climate stories swing between alarm and reassurance as violently as the fate of the Atlantic Meridional Overturning Circulation, the ocean system usually shortened to AMOC. In 2023, two physicists estimated it could collapse around mid-century [source: Nature Communications, 2023]. In early 2024, a modelling team reported the first full simulated tipping event and said the real ocean was "on route to tipping" [source: Science Advances, 2024]. Then in 2025, a different group ran 34 climate models and concluded a collapse this century was unlikely [source: Nature, 2025]. Same ocean, same year, opposite headlines.
That whiplash is the story. The AMOC is one of the most closely watched tipping elements in the climate system, and the science is genuinely unsettled — not because researchers are careless, but because measuring a planet-scale current is hard and the record is short. This piece tries to separate what is actually measured from what is modelled, and what is warned from what is known. It is about the state of the evidence, not a forecast.
What the AMOC actually is
The AMOC is the Atlantic branch of the ocean's global "conveyor belt." Warm, salty surface water flows north near the surface — the Gulf Stream is its best-known upper limb — releasing heat to the atmosphere over the North Atlantic. As that water cools and grows dense in the subpolar seas near Greenland, it sinks and returns south at depth. This overturning moves an enormous amount of heat and is a big part of why northwestern Europe is milder than its latitude would suggest.
The mechanism that makes it a tipping candidate is a feedback. If enough fresh water — from melting Greenland ice or increased rainfall — dilutes the northern surface water, it becomes too light to sink. Sinking slows, less salty water is carried north, the surface freshens further, and the slowdown can, in theory, reinforce itself until the circulation jumps to a much weaker state. Today the AMOC runs at roughly 15 to 20 sverdrups (one sverdrup is a million cubic metres per second). The debate is about whether, and when, that number could fall off a cliff rather than simply drift down.
What the observations show
Here the honest answer is: less than either camp's headlines suggest. Direct, continuous measurement of the AMOC only began in 2004, when the RAPID mooring array was strung across the Atlantic at 26.5°N. Twenty years is a short record for a current that varies enormously from month to month and year to year. The array shows a modest downward tendency, but researchers who work with the data stress that over such a short span the forced trend is very hard to separate cleanly from natural variability — the "signal" is buried in "noise" [source: Geophysical Research Letters, 2025]. The measurements neither confirm an imminent collapse nor rule out a long-term decline.
To see further back, scientists reconstruct the AMOC indirectly from proxies — ocean sediments, water-mass properties, deep-sea grain sizes. One widely cited reconstruction, stitching such records back to about AD 400, concluded the AMOC is now in its weakest state in over a thousand years, with a first weakening in the nineteenth century and a faster decline after the mid-twentieth [source: Nature Geoscience, 2021]. That is a striking result, but it is worth naming what kind of result it is: an inference from indirect indicators, not a direct flow measurement, and it carries its own uncertainties. "Weakest in a millennium" and "about to collapse" are different claims, and the proxy record supports the first far more than the second.
The collapse warnings
The studies that made headlines went further, using early-warning theory. As a system loses resilience and approaches a tipping point, it tends to recover more sluggishly from perturbations — a signature that shows up statistically as rising variance and autocorrelation, or "critical slowing down." In 2023, two researchers applied this to a sea-surface-temperature fingerprint of the AMOC spanning 1870 to 2020 and estimated a central collapse date around 2057 under continued emissions [source: Nature Communications, 2023]. The number traveled around the world.
Read past the headline, though, and the same authors are careful. Their estimate carries a very wide uncertainty range, spanning decades on either side of mid-century; it assumes their statistical model is approximately right; and they explicitly cannot rule out other mechanisms, or that any collapse would be only partial [source: Nature Communications, 2023]. Critics add that the temperature fingerprint may not faithfully represent the actual circulation. The 2057 figure is best read as "a data-driven estimate under strong assumptions," not a due date.
A second line of work is physical rather than purely statistical. A 2024 study produced the first complete AMOC tipping event in a complex climate model, watching the circulation crash from about 10 to near 2 sverdrups over roughly a century of simulated freshwater forcing, and proposed an observable warning signal — the freshwater transport the AMOC carries at the southern edge of the Atlantic [source: Science Advances, 2024]. By that measure, the authors wrote, today's ocean sits on the tipping side. Crucially, they declined to translate their model's timeline into a real-world calendar year, warning that the two cannot be directly equated. "On route to tipping" is a statement about direction, not a date.
The case for caution
Other groups, using the same physics, reach calmer conclusions. A 2025 analysis ran 34 climate models under extreme greenhouse-gas and meltwater forcing and found that in every case the AMOC weakened but did not fully collapse. The reason was a mechanism the alarm studies underweight: wind-driven upwelling in the Southern Ocean keeps pulling deep water up, sustaining a weakened overturning even under harsh forcing [source: Nature, 2025]. The authors concluded that a full collapse this century is unlikely.
A companion study built a simple physical model constrained by the twenty years of real observations and projected only "limited" weakening — 18 to 43 percent, or roughly 3 to 6 sverdrups, by 2100 even under very high emissions — again judging a 21st-century collapse unlikely [source: Nature Geoscience, 2025]. In this telling the AMOC is not a light switch poised to flip, but a large system that will sag under warming and yet keep turning.
Why careful scientists still disagree
If two teams can look at the same ocean and the same two decades of data and reach "on route to tipping" versus "collapse unlikely," something instructive is going on. Part of it is that the constraints are genuinely ambiguous. Remarkably, two 2025 studies both branded their approach as "observationally constrained" and still diverged: one projected the limited 18–43% weakening above, while another, correcting different model biases, projected about a 51 percent decline by the end of the century under a moderate-emissions scenario — far more than the unconstrained model average [source: Science Advances, 2025]. Which biases you correct, and how, changes the answer.
The official synthesis reflects this humility. The IPCC's 2021 assessment judged it very likely that the AMOC will weaken over the 21st century — a high-confidence qualitative statement — while assigning only low confidence to the quantitative size of that decline, and only medium confidence to the conclusion that it will not collapse abruptly before 2100 [source: IPCC AR6 WG1, 2021]. In plain terms: near-certain that it slows, genuinely uncertain by how much, and unable to fully rule out an abrupt jump. Much of the doubt traces to known model biases in AMOC stability and to the fact that many models still underrepresent meltwater from a warming Greenland — an omission that, if anything, tilts models toward being too stable.
What a collapse would — and wouldn't — mean
Because the impacts are dramatic, they deserve careful framing: everything below is conditional, describing what models produce if the AMOC is made to collapse, not what is predicted to happen. In one 2025 simulation set in a 2°C-warmer world, an imposed AMOC shutdown drove profound European winter cooling — average London winters near 1.9°C with cold extremes around −19°C, Oslo averaging −16.5°C with extremes near −48°C, and winter sea ice reaching the coasts of Britain and Scandinavia [source: Geophysical Research Letters, 2025]. A separate model study found a collapse would also dry southern Europe [source: HESS, 2025]. The paradox is that a collapse would cool northwest Europe even as the planet as a whole keeps warming.
Two further points cut against the disaster-movie image. First, this is slow-motion: any real transition unfolds over decades, not a weekend. Second, a collapse is not purely a cooling event. One 2026 study estimated that a shutdown would trigger an oceanic carbon release raising atmospheric CO₂ by 47 to 83 ppm, adding roughly 0.2°C of global warming even after the ocean's own cooling is accounted for [source: Communications Earth & Environment, 2026]. Beyond Europe, the widely modelled consequence is a southward shift of tropical rain belts, disrupting monsoon systems that billions rely on, alongside faster regional sea-level rise on Atlantic coasts [source: Nordic Council of Ministers, 2026].
What the deep past tells us
The paleoclimate record is why scientists take the possibility seriously at all: the AMOC has collapsed before. Sediment proxies show the overturning was nearly or completely shut down during the coldest phase of the last deglaciation around 17,500 years ago, and sharply reduced during the Younger Dryas cold snap about 12,700 years ago, with abrupt Northern Hemisphere temperature swings tracking each change [source: Nature, 2004]. That history proves the system can tip and can do so relatively fast. But it comes with a caveat that cuts both ways: those past collapses were driven by enormous pulses of glacial meltwater under ice-age boundary conditions very different from today's, so the past is a warning about what is possible, not a template that can be read straight onto the coming century.
What to watch
So where does an honest reader land? Not on "collapse is imminent," and not on "nothing to see here." The defensible summary is narrower: it is near-certain the AMOC will weaken this century, deeply uncertain by how much, and impossible to fully exclude a more abrupt transition that the models may be too stable to capture.
A few things will sharpen the picture. Watch whether the RAPID array and its successors, as the record lengthens past two decades, resolve a trend that a shorter record can't. Watch whether the early-warning indicators strengthen or fade as more data arrive. Watch the next IPCC assessment, and whether modelling groups can reconcile the "resilient" and "on-course" camps by better representing Greenland meltwater and the ocean biases that push results in opposite directions. And watch the specific physical metrics researchers have flagged, such as the freshwater transport at the Atlantic's southern boundary, which turn a vague fear into something measurable. The most useful posture is neither panic nor dismissal, but attention — to the difference between what the ocean is measured to be doing and what the models fear it might.