Introduction

As global warming continues, limiting the growth of atmospheric carbon dioxide (CO2) concentration has become an urgent priority. In 2024, global mean surface temperature (GMT) has exceeded 1.5  °C above pre-industrial levels1,2,3 signaling a potential long-term breach of this threshold within the next two decades jeopardizing the Paris Agreement goal of limiting long-term warming below 1.5  °C3. Therefore, beyond reducing anthropogenic emissions, such as following a carbon law to have a chance of achieving carbon neutrality4, it is crucial to quantify the impact of climate change on the natural carbon sinks on land and the ocean5. The ocean is a vital carbon reservoir, and has absorbed approximately one quarter of anthropogenic CO2 emissions over recent decades6, with recent estimates indicating about 26% for 2013–20227. While projected ocean and land carbon sinks increase, the proportions of absorbed CO2 are smaller in scenarios with higher emissions, resulting in a larger fraction of emitted CO2 remaining in the atmosphere8. Changes in major ocean circulation patterns in the future could have substantial implications for the global carbon budget9.

The Atlantic Meridional Overturning Circulation (AMOC) is considered to be one of the climate tipping elements that can have multiple equilibrium states10. This goes back to Stommel et al. (1961)11, who, using a simple box model, proposed that the AMOC has two different stable equilibrium states due to a positive salt advection feedback. Subsequent studies have further identified the bistability of the AMOC in modeling simulations, indicating the possibility of a stable AMOC collapsed state12,13,14,15,16. AMOC collapse has been shown to have a profound effect on climate17,18,19,20,21. Paleoclimate data provide strong evidence that AMOC changes caused abrupt climate changes during glacial periods22,23,24. Although the mechanisms of glacial climate changes are different from those caused by anthropogenic emissions, model simulations show that global warming may cause a weakening and even a complete shutdown of the AMOC in the future25,26,27,28. Despite no significant signal from direct observational measurements29, some studies based on indirect proxies suggest that the AMOC is weakening, and is currently in its weakest period in a thousand years30, although the reliability of such reconstructions is still debated31. The IPCC AR6 concludes that, with medium confidence, AMOC shutdown is unlikely this century32. However, recent research suggests this risk may be underestimated, highlighting the potential for substantial, persistent global impacts if tipping points were crossed20,33,34. Modeling studies also show that the weakening or shutdown of the AMOC can affect the functioning of the marine carbon cycle and cause additional changes in CO235,36,37.

Most previous studies on the impact of an AMOC collapse focus on the induced climate changes under pre-industrial conditions20,37,38,39 or specific CO2 levels18, finding that it causes a substantial drop in overall global surface temperatures, especially in the Northern Hemisphere, with a slight warming in the Southern Hemisphere. Several studies have investigated the transient AMOC response to ongoing CO2 forcing (e.g., SSP8.5 scenarios or transient CO2 experiments), indicating a weakening with increased CO2 forcing and a potential collapse at higher concentrations. Here, we define collapse as a rapid weakening of the AMOC to a nearly complete shutdown with the maximum strength below 5 Sv. These studies consistently show that AMOC weakening or collapse results in a similar global cooling pattern40,41,42. Several modelling studies for glacial and interglacial periods suggest that AMOC weakening or shutdown may result in increases or decreases in CO2, depending on model structure and boundary conditions, reflecting the complex interplay and compensating effects among ocean circulation changes and terrestrial carbon response37. As a result, the CO2 response to an AMOC shutdown remains highly uncertain. On the other hand, varying equilibrium CO2 concentrations substantially influence AMOC stability, with higher CO2 levels potentially corresponding to stronger equilibrium AMOC strength27,43,44, yet the consequences of AMOC shutdown on the climate and carbon cycle under these conditions remain understudied.

To simulate the impact of an AMOC collapse on climate under different global warming levels, we use the fast Earth System Model CLIMBER-X because it is computationally efficient and can systematically analyze the impact of an AMOC shutdown on climate and the carbon cycle45,46. CLIMBER-X includes a 3D ocean model, a statistical-dynamical atmosphere model, a thermodynamic-dynamic sea ice model, an ocean biogeochemistry model and a land surface model with dynamic vegetation and can simulate the global carbon cycle and the evolution of atmospheric CO245. The climate sensitivity of the model is  ~ 3 ° C and as such comparable to other major General Circulation Models8. We focus on the impact of an AMOC collapse at different equilibrium warming levels, and the resulting changes in the ocean and land carbon cycles and their impacts on the global climate via changes in atmospheric CO2. We designed experiments to quantify the role of ocean and land carbon cycles by setting up: (1) a coupled system with fully interactive CO2 (fully coupled, including the full land and ocean carbon cycle), (2) a version with the land carbon cycle disabled, leaving an ocean-only carbon cycle (land-carbon-off), and (3) a fixed-CO2 version to estimate the pure climate response following the AMOC shutdown (fixed-CO2). To force an AMOC collapse we apply freshwater hosing from model year 1000 to 2000, followed by continued simulation until year 7000. We add the freshwater to the Atlantic surface between 50 °N and 70 °N with intensities of 0.1 Sv and 0.2 Sv and compensate by adding a spatially uniform salt flux to the tropical Pacific between 30 °S and 30 °N to conserve global mean salinity (Fig. 1). To investigate the effect of an AMOC collapse at various warming levels we perform experiments starting from the climate and carbon cycle in equilibrium with CO2 concentrations ranging from 280 ppm to 600 ppm. Before starting the hosing experiments we spin up the model for 10,000 years, guaranteeing that the background conditions are in equilibrium across different CO2 levels (See Method and Supplementary Fig. 1).

Fig. 1: AMOC response to freshwater hosing from an equilibrium state under different baseline CO2 levels in fully coupled model setup.
Fig. 1: AMOC response to freshwater hosing from an equilibrium state under different baseline CO2 levels in fully coupled model setup.
Full size image

a AMOC strength (maximum Atlantic overturning circulation) at 26∘N, where the gray shading indicates the period with freshwater hosing. The dashed lines represent a hosing strength of 0.1 Sv, while the solid lines correspond to a hosing strength of 0.2 Sv. Inset: In the hosing experiment, freshwater is added to the Atlantic Ocean surface between 50∘N and 70∘N (+FH) and compensated in the Pacific between 30∘S and 30∘N (-FH). The streamfunction of the AMOC at 450 ppm before, during and after collapse for the 0.1 Sv-hosing experiment is shown in (b-d), averaged over model years 100–300 (b), 1500–1700 (c), 6800–7000 (d).

Results

AMOC response to hosing at different CO2 levels

After conducting 10,000-year spin-up simulations under prescribed atmospheric CO2 concentrations ranging from 280 to 600 ppm, the system reaches equilibrium at each baseline CO2 level (Supplementary Fig. 1). In the pre-industrial (280 ppm baseline) experiment, the simulated AMOC (Supplementary Fig. 2c) is slightly deeper and stronger than in many CMIP-class models47 but remains within the range of observational estimates47,48. We also compared our pre-industrial global overturning streamfunction with that from CMIP5 models and found similar large-scale patterns (Supplementary Fig. 3), consistent as well with reanalysis-based estimates49. The equilibrium AMOC increases in CLIMBER-X with higher CO2 (Supplementary Fig. 4b), which does not contradict the weakening of AMOC in transient CO2 experiments50. Willeit and Ganopolski (2024)43 show that in the CLIMBER-X model, higher CO2 levels enhance North Atlantic surface buoyancy losses primarily through thermal effects, which favors convection and deep water formation, thereby strengthening the AMOC. This is consistent with other atmosphere-ocean general circulation model simulations, suggesting that after an initial weakening due to transient warming, the AMOC may recover above its original strength due to positive salinity anomalies in the North Atlantic and long-term warming in the low-latitude Atlantic, which together enhance the meridional density gradient and strengthen the AMOC27. This interpretation does not exclude the well-established role of Southern Ocean wind-driven upwelling in setting the long-term equilibrium strength of AMOC, which requires compensating downwelling in the Atlantic or Pacific51,52.

Starting from these different equilibrium states, we apply freshwater hosing, utilizing the three model setups: fully coupled, land-carbon-off, and fixed CO2.

The results of the hosing experiments in the fully coupled model setup show the monostable and bistable characteristics of AMOC under different CO2 levels (Fig. 1a). For the pre-industrial CO2 level of 280 ppm the AMOC shows monostable behavior by collapsing under freshwater hosing but fully recovering once the forcing ends. At CO2 levels of 350 ppm or higher, the system shows bistable behavior, meaning that once the AMOC collapses, it cannot recover and stays in the off state. However, higher CO2 scenarios need a stronger freshwater hosing to cause this collapse. The corresponding results are also presented in the land-carbon-off and fixed CO2 model setups (Supplementary Fig. 5). Willeit and Ganopolski (2024)43 demonstrate using the CLIMBER-X model that atmospheric CO2 is a crucial control variable alongside freshwater forcing in determining AMOC stability. They show that when CO2 exceeds 350 ppm, the AMOC becomes bistable in the model, and if the AMOC collapses, this could be irreversible.

Here, we use the 450 ppm simulation as a representative example (which corresponds to the global equilibrium at 2.1  °C warming, see also Supplementary Fig. 4a) to illustrate the detailed evolution of temperature response during the hosing experiment. In our results, during the hosing period, the AMOC weakens until collapse (Fig. 1b-d), displaying AMOC streamfunction variations that are consistent with other Earth System Model simulations20. To investigate the impact of AMOC collapse, we only focus on a hosing rate of 0.2 Sv in the subsequent analysis, which ensures that all cases eventually lead to collapse.

Climate response to AMOC collapse

We investigate the long-term regional temperature responses to an AMOC collapse. The fully coupled model simulations show that, after reaching a re-established equilibrium following an AMOC collapse, the climate exhibits a strongly asymmetric hemispheric temperature response: pronounced cooling in the Northern Hemisphere and concurrent warming in the Southern Hemisphere, reflecting a global redistribution of heat (Fig. 2c). All three model setups show similar spatial response patterns (Supplementary Fig. 6). The Northern Hemisphere cooling simulated here, consistent across all CO2 baselines (Fig. 2d, f), aligns with previous modeling studies collapsing the AMOC under pre industrial conditions20,38,39,53. However, our simulations exhibit a much more pronounced warming in the Southern Hemisphere across all CO2 baselines than in pre-industrial cases in previous studies (Fig. 2c). In this scenario, the Arctic regions (60 °N − 90 ° N) cool by around 7  °C after the AMOC collapse (Fig. 2d), and the North Atlantic cooling is stronger than in other regions at comparable latitudes. AMOC collapse leads to a sharp reduction in the transport of heat to northern North Atlantic, which causes cooling strongly amplified by sea-ice-albedo feedback20,54. The cooling of the Arctic leads to a considerable increase in Arctic sea ice area, leading to an increase in albedo (Supplementary Fig. 7), which further causes cooling in the Northern Hemisphere. Cooling occurs in the Northern Hemisphere, intensifying over regions north of 60 ° N, which is similar to the pattern of temperature changes after the collapse of the AMOC in some simulations under pre-industrial10,20,39. In contrast, the Antarctic regions (60 °S − 90 ° S) warm by around 6  °C (Fig. 2f). Because our simulations do not include ice sheet modules, the result does not account for the additional effect resulting from melting ice sheets (e.g., those in Antarctica and Greenland). The magnitude of the long-term regional temperature changes induced by AMOC collapse is broadly similar across the different experiments of CO2 baselines (Fig. 2d, f). The spatial pattern is also similar, with persistent North Atlantic cooling in all cases. When the background greenhouse warming associated with a 450ppm baseline (Fig. 2a) is superimposed on the AMOC collapse response, the overall climate state (relative to pre industrial conditions) features intense warming in the Antarctic and pronounced cooling in the North Atlantic (Fig. 2e). Therefore, in some regions south of 60 °S, the combined warming effect may exceed 10  °C. The cooling is reduced over the Northern Hemisphere, where greenhouse warming partially offsets the temperature decline induced by the AMOC collapse.

Fig. 2: Changes in surface air temperature due to AMOC collapse for the fully coupled model setup in the experiments with a hosing rate of 0.2 Sv.
Fig. 2: Changes in surface air temperature due to AMOC collapse for the fully coupled model setup in the experiments with a hosing rate of 0.2 Sv.
Full size image

a Spatial distribution of temperature difference resulting from baseline CO2 level (450 ppm) above pre-industrial condition, calculated as the temperature difference between the baseline equilibrium state and the pre-industrial temperature. c Temperature change due to AMOC collapse at 450 ppm (ΔTAMOC), determined by subtracting the baseline equilibrium temperatures from the final state after AMOC collapse. e Total temperature response to the AMOC collapse at 450 ppm baseline experiment, calculated as the difference between the final equilibrium state after AMOC collapse and the pre-industrial state. Changes in mean surface air temperature due to AMOC collapse for the full coupled model setup under various baseline CO2 levels during hosing experiments (computed as the difference from each experiment’s initial value): (b) global mean surface air temperature anomaly (ΔGMTAMOC), (d) mean surface temperature change over 60∘N − 90∘ N, and (f) mean surface temperature change over 60∘S − 90∘S.

On a global scale, our results indicate that the AMOC collapse leads to a net increase in GMT (Fig. 2b), indicating an additional warming effect globally. If the AMOC stabilizes in an off state after the collapse, global temperatures eventually stabilize at 0.17–0.27  °C higher than before water hosing, across the different CO2 baselines (Fig. 3d). The transition from AMOC weakening to collapse is accompanied by characteristic temperature changes. After the start of hosing, the GMT first drops rapidly under the baseline GMT and ΔGMTAMOC remains below  − 0. 5  °C during the AMOC’s weakening and transition, then rebounds above the pre–hosing baseline temperature. (Fig. 3b and Fig. 2a). This closely aligns with the findings of Boot et al. (2024)42, where the AMOC collapse induces strong cooling around the transition point, with temperatures changing by approximately  − 1  °C before gradually recovering.

Fig. 3: Changes in atmospheric CO2 and global mean annual surface air temperature (GMT) after AMOC collapse down under various baseline CO2 levels.
Fig. 3: Changes in atmospheric CO2 and global mean annual surface air temperature (GMT) after AMOC collapse down under various baseline CO2 levels.
Full size image

(a, b) display the atmospheric CO2 concentration and GMT response to freshwater hosing over time in the fully coupled model setup. (c) Illustrates the difference in atmospheric CO2 concentrations between equilibrium states before and after AMOC collapse (\(\Delta {{{{\rm{CO}}}}}_{2}^{AMOC}\)) for the fully coupled and land-carbon-off model setups. \(\Delta {{{{\rm{CO}}}}}_{2}^{AMOC}\) is computed using the average over the final 500 years (6500–7000) minus the average over the first 500 years (0–500) of the simulation under each CO2 baseline. e Presents the corresponding ratio of \(\Delta {{{{\rm{CO}}}}}_{2}^{AMOC}\) relative to the baseline CO2 (pre-hosing atmospheric CO2 levels) for both the fully coupled and land-carbon-off model setups. In the fixed CO2 model setup, atmospheric CO2 concentrations remain constant throughout the entire simulation. d Illustrates the additional GMT changes due to AMOC collapse (ΔGMTAMOC) as a function of ΔGMTbaseline, where ΔGMTbaseline represents the difference between the initial GMT under each baseline CO2 level and the pre-industrial GMT. Specifically, ΔGMTbaseline is calculated using the first 500-year average GMT under each baseline CO2 level minus the first 500-year average pre-industrial GMT (Supplementary Fig. 4a). Panel (f) presents the ratio of ΔGMTAMOC to ΔGMTbaseline for the fully coupled, land-carbon-off, and fixed-CO2 model setups. All results in are based on experiments utilizing 0.2 Sv hosing to assess responses following AMOC collapse.

Having established a net additional warming following the AMOC collapse, we next examine results from different model setups to explore the mechanisms driving this response. We first assess the climate response to AMOC collapse under fixed-CO2 model setup, i.e. excluding the carbon cycle effects. In this model setup, both land and ocean carbon cycles are disabled and CO2 remains constant over time. Comparing the equilibrium GMT after the AMOC collapse with that before the freshwater hosing input shows a global mean temperature decrease of 0.2 to 0.3  °C (Fig. 3d). This suggests that oceanic physical processes associated with the AMOC collapse exert a cooling influence on the global climate. We then examine the roles of the ocean and land carbon cycles. When the land carbon response is disabled (land-carbon-off setup), the GMT increases by 0.41–0.51  °C after the AMOC collapse (Fig. 3d). Compared with the fixed-CO2 setup, where physical ocean processes alone induce global cooling, this warming response indicates that changes in the ocean carbon cycle contribute to a net additional warming. The stronger GMT increase in the land-carbon-off model setup relative to the full coupled setup further suggests that the land carbon cycle exerts a compensating cooling effect. Overall, the net additional warming in the fully coupled model is primarily driven by changes in the ocean carbon cycle during the AMOC collapse (Fig. 3a).

Carbon cycle response after AMOC collapse

We next examine how the atmospheric CO2 concentration changes over time after the end of the hosing experiment towards equilibrium in the simulations with interactive CO2. The results indicate that both the fully coupled and land-carbon-off model setups show a rise in atmospheric CO2 concentration as the AMOC begins to weaken, across all baseline CO2 levels (Fig. 3a). In the fully coupled version, the irreversible collapse of the AMOC would lead to an increase in atmospheric CO2 concentration by 47–83 ppm (100GtC to 175GtC in atmospheric carbon inventory) across the 6 experiments conducted under the baseline CO2 levels from 350 ppm to 600 ppm, with larger increases occurring at higher baseline CO2 levels. Under the 280-ppm baseline, our simulated CO2 rise during AMOC collapse ( ~ 30 ppm; Supplementary Fig. 2) falls within the upper range of previous estimates37, whereas higher-CO2-baseline cases in our study yield much stronger releases. In the land-carbon-off setup, where the terrestrial carbon cycle is disabled, AMOC collapse leads to an even larger CO2 increase, by 72–130 ppm (153GtC to 275GtC in atmospheric carbon inventory), as the compensating land carbon sink is absent (Fig. 3c). In contrast, in the fully coupled setup, the land carbon cycle functions as a carbon sink (Supplementary Fig. 8), absorbing part of the carbon released from the ocean following the AMOC shutdown. In all three model setups, carbon exchanges occur only among the ocean, land and atmosphere, with no exchange with geological reservoirs; hence, in the land-carbon-off setup, atmospheric CO2 increases entirely from oceanic release. Moreover, while the additional atmospheric CO2 concentration change increases linearly with the baseline background CO2 level, the ratio of this increase to the baseline CO2 level remains consistently around 20% to 22% for land-carbon-off setup, while the ratio is lower for the fully coupled setup at around 13% (Fig. 3e), illustrating the compensating effect of the terrestrial carbon sink (Supplementary Fig. 8). Our findings thus indicate that an AMOC collapse could substantially increase atmospheric CO2 concentrations.

Overall, our results indicate ocean dynamics induce global cooling, while ocean carbon release increases atmospheric CO2 resulting in a net global warming of about 0.2 °C in the fully coupled model setup (Fig. 3d), even though the land carbon sink offsets some of the CO2-induced warming. We compare the ratio of additional GMT change from AMOC collapse to the GMT increase due to baseline CO2 rise from the preindustrial level. In the fully coupled version, as baseline CO2 levels increase from 350 ppm to 600 ppm, this ratio drops from  ~ 19% to  ~ 8% (Fig. 3f), indicating that AMOC collapse contributes relatively less to GMT change at higher CO2 levels. This implies that the relative temperature response triggered by AMOC collapse is stronger and more detrimental at lower CO2 levels.

Southern Ocean convection following AMOC shutdown

As the AMOC collapse is accompanied by substantial oceanic CO2 outgassing, resulting in additional warming, we investigate associated ocean circulation changes and find that deep Southern Ocean convection plays a key role in driving this ocean carbon release. With the start of hosing, the AMOC weakens rapidly, reducing North Atlantic circulation depth (Fig. 1b–d and Supplementary Fig. 9a, b) while enhancing deep convection over the Southern Ocean (Fig. 4f-h and Supplementary Fig. 9a, b). CLIMBER-X has previously been shown to reproduce the main large-scale features of the present-day global air–sea CO2 flux pattern46, including a carbon source in the eastern equatorial Pacific and a pronounced sink in the North Atlantic, broadly consistent with observational estimates55,56,57. In our study, under warmer equilibrium climates with higher CO2 baseline conditions, the Southern Ocean exhibits a stronger carbon source than in pre-industrial run (280 ppm). Changes in global overturning circulation after AMOC collapse redistribute global ocean carbon sink and source patterns (Fig. 4). Both the carbon sink and source in the North Atlantic region weaken and the carbon sink distributed in the Northern Hemisphere shifts southward overall after AMOC collapse (Supplementary Fig. 10a). The mixed layer depth in the 70 °S–80 °S region deepens rapidly around the AMOC transition (Supplementary Fig. 10b–e), indicating the formation of deep convection in the Southern Ocean (Fig. 4g). This convection coincides with enhanced Southern oceanic CO2 flux to the atmosphere across the same latitude band (Supplementary Fig. 10a). As a result, the deep ocean is strongly ventilated (as indicated by the water-age tracer in Supplementary Fig. 10f–h), leading to a release of carbon in the Southern Ocean around Antarctica, which dominates the global oceanic CO2 outgassing (Supplementary Fig. 9f), and a large increase in atmospheric CO2. The deep water formation also keeps large parts of the Southern Ocean ice-free, resulting in a pronounced warming over the region (Supplementary Fig. 7). Overall, both the large increase in carbon flux from ocean to atmosphere and warming in the Southern Hemisphere in response to an AMOC collapse in our simulations are a consequence of widespread deep convection being eventually triggered in the Southern Ocean (Supplementary Fig. 10).

Fig. 4: The ocean carbon flux and global overturning circulation before and after AMOC collapse for fully coupled version with a hosing rate of 0.2 Sv.
Fig. 4: The ocean carbon flux and global overturning circulation before and after AMOC collapse for fully coupled version with a hosing rate of 0.2 Sv.
Full size image

During the process, the changes of net and cumulative ocean carbon flux to the atmosphere under different CO2 levels are shown in (a, b), respectively. The initial global air-sea CO2 flux patterns at 450 ppm is shown in (c), averaged over model year 100–300. (d, e) show the difference in air-sea CO2 flux between the average and the initial for model years 1100–1300 and 6800–7000, respectively. The positive values of CO2 flux indicate carbon flux into the ocean and negative values represent carbon release from the ocean. The streamfunctions of global overturning circulations before and after the AMOC collapse under 450 ppm, averaged over model years 100–300 (f), 1100–1300 (g), 6800–7000 (h) are illustrated respectively.

Support for this model response in the Southern Ocean comes from CO2 changes during Heinrich events observed in paleoclimate records. Large iceberg discharges from the Laurentide ice sheet during the last glacial period led to an AMOC collapse and proxy records from Antarctic ice cores show a large and abrupt increase of both Antarctic temperature and atmospheric CO2 some centuries after the AMOC shutdown, which have been suggested to be related to a possible start of convection around Antarctica at that time58,59. CLIMBER-X has recently been shown to realistically reproduce these features as a response to an AMOC collapse produced by the addition of freshwater flux in the North Atlantic to mimic a Heinrich event60. In that study, it was also shown that the start of convection in the Southern Ocean as a response to an AMOC collapse is a robust feature of the model. The mechanisms behind this so-called bipolar convection seesaw are described in detail in ref. 60 and basically involve a destratification of the Souther Ocean in response to an AMOC shutdown, which eventually leads to convective instability around Antarctica. As shown in this paper, this is not just a feature of cold climates, but could plausibly also occur in response to a future AMOC collapse (Supplementary Fig. 7 and Supplementary Fig. 10). It is this model feature that produces the exceptionally large CO2 response responsible for the overall global warming effect of an AMOC shutdown.

Discussion

Our CLIMBER-X simulations reveal the impact of AMOC collapse on ocean carbon storage and its effect on long-term global climate under different CO2 levels, ranging from pre-industrial levels (280 ppm) to more than doubling of pre-industrial CO2 concentration (600 ppm). These results highlight the role of the AMOC as a key regulator of global climate. The CO2 rise associated with AMOC collapse causes an additional global warming of about 0.2  °C. The substantial regional temperature responses, however, originate primarily from the direct physical effects of AMOC collapse on climate. Our results show that, at the equilibrium climate across all baseline CO2 levels, the AMOC collapse would lead to pronounced regional responses: a cooling by around 7  °C over the Arctic (60 °N − 90 ° N) and a warming by around 6  °C over the Antarctic (60 °S − 90 ° S) (Fig. 2). An AMOC collapse may cause the release of large amounts of carbon stored in the Southern Ocean to the atmosphere. While ocean dynamics during AMOC shutdown induce global cooling in the fixed-CO2 setup, the substantial ocean carbon release in the fully coupled setup increases atmospheric CO2 levels, and even with considerable land carbon uptake, the remaining CO2 offsets the cooling and contributes to additional warming. Our results indicate that under equilibrium warmer climate conditions with 350 ppm to 600 ppm of CO2, an AMOC collapse may release enough carbon to cause a CO2 increase by 47–83 ppm. Most of this CO2 rise is the result of convection being triggered in the Southern Ocean as a response to the forced AMOC shutdown.

Our results show that AMOC collapse induces much stronger CO2 increases at higher baseline CO2 levels (compared to 280 ppm), which is further supported by paleoclimate evidence. Recent studies show that CLIMBER-X reproduces critical aspects of millennial-scale climate variability during glacial times61. In particular, consistent with more complex models (e.g., Vettoretti et al. 202262; Malmierca-Vallett et al. 202463), it produces Dansgaard-Oeschger like events as part of internal variability of the climate system under mid-glacial ice sheets and low CO261. It is widely accepted that these Dansgaard-Oeschger events are associated with abrupt changes in the AMOC, involving transitions between weak and strong AMOC states. However, Dansgaard-Oeschger variability is not associated with a complete collapse of the AMOC and ice core data show only minor responses of atmospheric CO2 of  ~ 5 ppm during Dansgaard-Oeschger events64. This is consistent with what was found in model simulations (e.g., see ref. 65) and also with what CLIMBER-X produces60. As a response to a complete AMOC shutdown following freshwater input into the North Atlantic from Heinrich events during glacial times, CLIMBER-X produces a much larger response in atmospheric CO2 of  ~ 15 ppm, which is again in good agreement with ice core data during Heinrich Stadials60. The process behind this large CO2 increase is a sudden onset of convection in the Southern Ocean in response to an AMOC shutdown60. Several studies have suggested that deep Southern Ocean convection can promote the release of stored oceanic carbon37,66,67. Because CLIMBER-X reproduces past climate-carbon cycle variability linked to AMOC changes, we have confidence in its response to a potential AMOC collapse under warmer conditions. In our simulations, the comparatively larger atmospheric CO2 increase is primarily associated with the development of deep Southern Ocean convection, which ventilates carbon-rich deep waters and drives substantial outgassing to the atmosphere. The CO2 increase in response to an AMOC shutdown grows strongly with higher baseline CO2 levels (Fig. 3c). Higher CO2 baselines correspond to larger ocean carbon inventories (Supplementary Fig. 11), contributing to substantially more carbon to be released during a collapse, suggesting that the amplitude of past glacial CO2 responses is not a proper analog for the expected amplitude of CO2 increase in a warmer, higher-CO2 world. Consistent with this, other models show that in warmer climates a sudden Southern ocean convection onset can raise atmospheric CO2 by  ~ 40 ppm66, much larger than observed glacial-period changes. Our results suggest that the same Southern Ocean convection-driven mechanism could operate under future warmer conditions, producing a substantially larger CO2 response than during glacial times.

It is important to emphasize that our study focuses on the long-term climate response to the AMOC collapse induced by freshwater hosing in an equilibrium climate, which differs from studies of the transient climate change driven by CO2 emission scenarios (e.g., see refs. 42,68). Distinguishing responses to AMOC behavior under transient versus equilibrium background conditions is crucial: in CLIMBER-X, the AMOC weakens during transient warming, consistent with CMIP models42, but subsequently recovers and stabilizes in a stronger equilibrium state for baseline CO2 concentrations of 350–600 ppm (Supplementary Fig. 1). To isolate the AMOC effect in these warmer background climates, we therefore performed all hosing experiments under equilibrium conditions (280–600 ppm). In this case, higher CO2 baseline equilibrium states correspond to greater initial AMOC strength (Supplementary Fig. 1 and Supplementary Fig. 4b) and resilience, as evidenced by the stronger freshwater forcing required to trigger collapse (Fig. 1a). This suggests that our findings are conservative, in the sense that AMOC collapse with subsequent net-warming outcomes, occurs even with a stronger AMOC at the outset.

We focus on the integrated carbon response following AMOC collapse without decomposing individual mechanisms (e.g., solubility or biological pump effects69) in this study. Intermodel differences in AMOC sensitivity70, together with known biases in North Atlantic deep-water formation71, salinity, and meltwater runoff72, and background climatological states37 may influence the magnitude and thresholds of carbon responses to AMOC collapse across models. However, our results indicate that these factors play a secondary role compared with the dominant ventilation-driven carbon loss (Supplementary Fig. 9c–f). The total ocean carbon declines almost entirely through reduced dissolved inorganic carbon, and the dissolved inorganic carbon loss closely matches the cumulative ocean-to-atmosphere carbon flux (Supplementary Fig. 9c, d). Changes in dissolved and particulate organic carbon are comparatively minor (There is a difference in magnitude) (Supplementary Fig. 9e). As the system re-equilibrates after the collapse, the global mean net air–sea CO2 flux stabilizes near zero, whereas the sharp release during the collapse corresponds to pronounced Southern Ocean outgassing. In contrast, surface carbon flux changes in the Pacific and Indian Oceans are minor, while the Atlantic evolves from a temporary enhancement of the sink at the onset to a weakened and eventually nearly neutral state (Supplementary Fig. 9f). These surface carbon flux evolutions over different oceans explain the changes in atmospheric CO2. The sharp transient outgassing in our simulations is driven by deep Southern Ocean convection that ventilates carbon-rich waters.

Our experiments show that terrestrial carbon sinks may partially mitigate the impact of oceanic carbon release triggered by AMOC collapse and the induced deep convection in the Southern Ocean (Supplementary Fig. 8). Although the impacts on the terrestrial carbon cycle are not the main focus of this study, they are still important. Our results show that the absorption of carbon by the land markedly reduces the additional global mean warming from 0.41–0.51  °C to around 0.17–0.27  °C (Fig. 3). However, several model studies indicate that an AMOC collapse may lead to substantial terrestrial carbon loss, increasing atmospheric CO2 by up to  ~ 20 ppm, primarily due to tropical drying and reduced vegetation productivity73,74,75. At present, human land-use change has already converted large terrestrial carbon sources such as the Amazon rainforest from a carbon sink to a carbon source76,77. This means that while terrestrial carbon sinks remain important for global climate stability, their future evolution is uncertain and will depend on complex interactions between climate change and human land use.

In this study, the prescribed freshwater perturbations (0.1 and 0.2 Sv) are used as idealized hosing experiments to demonstrate AMOC bistability under different background CO2 levels and to evaluate the subsequent Earth system response, rather than to represent specific real-world forcing scenarios. Under high-emission scenarios, a near-complete loss of the Greenland Ice Sheet could generate roughly 0.1 Sv of freshwater input (0.1Sv = 105m3 s−1 ≈ 3, 156Gt yr−1)78. Thus, our idealized forcings fall within a plausible envelope for testing AMOC sensitivity. Finally, the model does not include dynamic ice sheets, so Antarctic meltwater and its teleconnections are not represented. Recent work shows that West Antarctic meltwater release can affect AMOC resilience79, and incorporating Antarctic meltwater effects remains an important avenue for future work.

In summary, this study comprehensively analyzed the possible impacts of AMOC collapse on the ocean carbon cycle and climate response under a series of CO2 baseline equilibria. The results emphasize that under the various climate warming equilibrium, the potential collapse of AMOC may lead to substantial ocean carbon release to the atmosphere, leading to additional global warming and producing pronounced regional responses. Although such long-term equilibrium responses may not be fully realized in the real world, they provide valuable insight into the underlying physical mechanisms and potential climate and ocean carbon responses. Importantly, while the additional global warming is modest, the substantial regional temperature responses, such as multi-degree warming over parts of Antarctica under a 450-ppm background, highlight the substantial climate risks associated with an AMOC collapse.

Methods

CLIMBER-X is a fast Earth system model45 that includes a 3D frictional-geostrophic ocean model GOLDSTEIN with 23 vertical layers80, a semi-empirical statistical dynamical atmosphere model SESAM45, a sea ice model SISIM45, and a land surface model PALADYN with dynamic vegetation81, which can simulate the global carbon cycle in Earth systems45. CLIMBER-X’s simulation performance of the Earth system under different forcing and boundary conditions is comparable to that of the most advanced CMIP6 models, especially the simulation performance of the current AMOC profile fits the CMIP6 model and climate observations well45. Moreover, CLIMBER-X also shows its ability to simulate the multi-stability of the AMOC system under different boundary conditions43,61.

To simulate the collapse of the AMOC and further analyze its impact on the global carbon cycle and global temperature, we use CLIMBER-X in a climate–carbon cycle set up including atmosphere, ocean, land with dynamic vegetation and carbon cycle, sea ice and ocean biogeochemistry. We do not use the ice sheet model and this paper does not involve the study of ice sheet changes. To further analyze the role of ocean activity and carbon cycle in the AMOC shutdown process, we designed experiments to quantify the role of each component by setting: (1) a fully coupled system (Fully coupled), which includes all the above modules, (2) a version in which the land carbon cycle is disabled, leaving only the ocean carbon cycle to interact with the atmosphere (land-carbon-off) with the influence of vegetation and soil carbon cycle processes being ignored and 3) a fixed-CO2 setup to estimate the pure climate response to the AMOC shutdown (Fixed-CO2) in which CO2 remains constant at the baseline CO2 level throughout the entire process and the temperature response comes entirely from ocean, atmosphere and land dynamics, among which the sea ice-albedo feedback is dominant.

A 10,000-year spinup run is first performed under each CO2 baseline to ensure the system reaches equilibrium (Supplementary Fig. 1). Subsequently, a 7000-year experiment is conducted for each equilibrated state, with freshwater hosing applied during model years 1000–2000. Freshwater hosing is performed at strengths of 0.1 Sv and 0.2 Sv in the North Atlantic in the range of 50 °N–70 °N. In our experiments, freshwater hosing of 0.1 Sv proved insufficient to induce AMOC collapse under atmospheric CO2 concentrations of 550 ppm and 600 ppm; consequently, the freshwater flux was increased to 0.2 Sv to ensure consistent AMOC collapse across all experimental scenarios. To keep global salinity constant, a compensating flux of the same magnitude and opposite sign was applied in the Pacific between 20 °N and 50 °N. We tested global compensation of the freshwater flux, instead of only in the tropical Pacific, and found little impact on the results. After the end of the hosing the run is continued until model year 7000 to ensure that the system returns to equilibrium. We investigate the carbon cycle and climate responses across various climate conditions by conducting the same spinup and freshwater hosing experiment under CO2 concentrations ranging from pre-industrial (280 ppm) to 600 ppm.