Committed emissions from existing and planned power plants and asset stranding required to meet the Paris Agreement

Over the coming decade, the power sector is expected to invest ~7.2 trillion USD in power plants and grids globally, much of it into CO2-emitting coal and gas plants. These assets typically have long lifetimes and commit large amounts of (future) CO2 emissions. Here, we analyze the historic development of emission commitments from power plants and compare the emissions committed by current and planned plants with remaining carbon budgets. Based on this comparison we derive the likely amount of stranded assets that would be required to meet the 1.5 °C–2 °C global warming goal. We find that even though the growth of emission commitments has slowed down in recent years, currently operating generators still commit us to emissions (~300 GtCO2) above the levels compatible with the average 1.5 °C–2 °C scenario (~240 GtCO2). Furthermore, the current pipeline of power plants would add almost the same amount of additional commitments (~270 GtCO2). Even if the entire pipeline was cancelled, therefore, ~20% of global capacity would need to be stranded to meet the climate goals set out in the Paris Agreement. Our results can help companies and investors re-assess their investments in fossil-fuel power plants, and policymakers strengthen their policies to avoid further carbon lock-in.


Committed CO emissions and carbon budgets in the power sector
The power sector is expected to invest about 7.2 trillion USD in power plants and transmission and distribution grids over the next decade (IEA 2016). The average expected lifetime of generators can range from 20-25 years for solar PV up to 70 years and longer for hydroelectric generators (EIA 2011, IEA 2016. Coal-, gasand oil-powered generators have a typical lifetime of between 35-40 years (Davis and Socolow 2014). These lifetimes probably represent only economic rather than technical lifetimes, however, since many power generators operate long beyond their expected end of life. The relatively long payback periods for such assets expose investments to the risk of future changes in economic and regulatory conditions. Changes in input prices, the competitive landscape, or regulation can have large impacts on the profitability and economic viability of such assets, before they have a chance to pay their investment back (Caldecott et al 2017).
These long lifetimes mean that any investment made today in carbon dioxide (CO 2 ) emitting infrastructure will have a considerable effect on the ability to achieve required CO 2 emission reductions in the future-even if these desired reductions are many years away (Davis et al 2010, Rozenberg et al 2015. In recent years, therefore, the concept of (expected) committed cumulative carbon emissions (hereafter referred to as committed emissions) has been developed, and gained popularity within the scientific community (Guivarch and Hallegatte 2011, Davis and Socolow 2014, Pfeiffer et al 2016. Committed emissions are the cumulative emissions an asset would emit over its remaining lifetime under normal economic conditions, i.e. if it were to be operated at normal utilization (Davis and Socolow 2014).
To stabilize global warming at any level, not just 1.5 • C or 2 • C but virtually any level, anthropogenic emissions of long-lived climate pollutants (LLCPs) must eventually reach net-zero (Matthews and Caldeira 2008). Therefore, global warming can be seen as a function of the cumulative emissions of such LLCPs, chiefly CO 2 , rather than of annual emission rates (Matthews and Caldeira 2008, Allen et al 2009, Matthews et al 2009. The cumulative future emissions of currently operating and planned infrastructure, are therefore likely to be much more relevant to climate outcomes than the individual annual emissions of such assets .
In this regard, it should be noted that power and heat generation was responsible for ∼38% of total global emissions in 2014 (IEA 2016, Le Quéré et al 2016, more than any other sector. Committed emissions from power plants are, therefore, particularly important for climate policies. Davis and Socolow (2014) suggest a methodology for Commitment Accounting of CO 2 Emissions in the power sector, and find that, under standard lifetime assumptions, in 2012, assets in the power sector were committed to ∼307 GtCO 2 future emissions and that these commitments had been growing at ∼4% per year over the previous decade. Based on their results Pfeiffer et al (2016) calculated that 2017 would be the year when the global 2 • C Capital Stock for Electricity Generation was reached, i.e. when existing power generators would commit to enough CO 2 emissions, to consume the remaining generation-only carbon budget for a 50% chance for global warming below 2 • C. Other studies have since used the same or similar methodologies to calculate the CO 2 emission commitments of different assets or sectors and assess their impact on climate policies, investments and the consequent costs of achieving climate goals (Bertram et al 2015, Rozenberg et al 2015, Sanchez et al 2015. This paper updates previous efforts, especially those of Davis and Socolow (2014) and Pfeiffer et al (2016), by using an improved generator database and updating this data to late 2016. Moreover, for the first time, we include generators currently under construction, or in different stages of the planning process, to estimate the development of future committed emissions from the global pipeline of currently planned power generators. Finally, we use a significantly improved estimate of the currently remaining generation-only carbon budgets for different climate scenarios (compared to Pfeiffer et al 2016), and compare this new estimate with the emission commitments. This effort allows us to derive the likely cumulative amount of power sector stranding each climate scenario would imply.
The updated capital stock and budget numbers suggest that 2017, the previous estimate for the commitment year for the 2 • C capital stock (Pfeiffer et al 2016), might have been too optimistic-however, not by far. The cross-comparison with recently updated carbon budget figures (Millar et al 2017) suggests that the commitment year for a realistic chance to limit warming to only 2 • C was probably sometime between 2011 and 2016. Consequently, we find that the committed cumulative future emissions from currently operating power plants (∼300 GtCO 2 ) would now already surpass the currently available generationonly carbon budget for the average 430-480 ppm scenario (∼240 GtCO 2 ). In addition, plants in various stages of the planning process would add almost the same amount of commitments (∼270 GtCO 2 ) as those plants currently operational. Even if all currently planned projects are immediately suspended, up to 20% of global fossil-fuel generation capacity would still have be stranded (that is, prematurely decommissioned, underutilized, or subject to costly retrofitting) if humanity is to meet the climate goals set out in the Paris Agreement.

Data and methods
We calculate historic and current committed emissions from currently operating, planned and already retired power generators 5 . Since these are not typically reported in any publicly available source, we use existing databases on generator capacity vintages (in GW), combined with (historic) average annual utilization rates (in percent), heat rates (in mbtu per GWh), fuel emission factors (in tCO 2 /mbtu), and expected operational lifetimes. In the rest of this section we describe the databases and sources used (2.1), and how we calculate the committed emissions and how this differs from previously used methodologies (2.2).

Databases and sources
We determine generation capacities by merging all generators from the most recent versions of five databases: (1) CoalSwarm (Feb 2017); (2) Platt's UDI World Electric Power Plants (WEPP) database (Q4 2016); (3) Greenpeace's database of planned coal generators in China; (4) Sekitan's Japan coal-fired power plant database (Q1 2016); and (5) Kiko Network's Japan coal-fired power plant database (Q1 2016). We merge these sources by manually confirming unique power plant names, locations, current statuses, online years and capacity, using internet research as required 6 . The most recent data is used where matched generators 5 We differentiate between generator and plant. The generator is the device that generates the electrical power for use in an external circuit. A plant can consist of several generators. We calculate committed emissions on a generator level since generators within plants are often replaced, such that the remaining lifetime of a plant is less helpful than the remaining lifetime of a generator. 6 Most generators could be matched using an algorithm. Only generators that did not match were manually confirmed. have conflicting fields (for example different operating statuses). The resulting database effectively defines the locations of all the world's power generators, their ownership, age, fuel type, technology, expected lifetime and capacity. It is particularly current and comprehensive for coal-fired power generators, the most carbon-intensive assets 7 .
We use three additional sources to calculate committed emissions: (1) current and historic heat rates from the US Energy Information Agency (EIA) and the US Environmental Protection Agency (EPA) (EPA 2009, EIA 2017 Davis and Socolow's (2014) Commitment Accounting of CO 2 Emissions marks the first time a comprehensive methodology has been described to calculate future emissions from existing power generators. One criticism of their approach, however, applies to CARMA 10 , one of the databases they use. They make the 'arbitrary assumption that CARMA's emissions and energy data for 2009 (or, occasionally, 2004) are an accurate estimate throughout a plant's lifetime' (Davis and Socolow 2014). 2009 was in many respects not a representative year for global energy consumption and emissions-in fact, with the financial and economic crisis at its height, 2009 was one of the few years in recent decades in which global emissions decreased year-on-year (Le Quéré et al 2016). In this paper, we therefore refine the approach described by Davis and Socolow. First, a broader and updated (late 2016) base of power generators is used that completes known gaps in the Platt's UDI WEPP database, e.g. in microgeneration and in China (see section 2.1). Second, for missing online years 11 a similar, but more granular, methodology was used as the one described by Davis and Socolow. Most importantly, the estimation of online years and lifetimes is conducted based not only on technology, capacity and country, but also by taking account of generator and turbine type, online year (for lifetimes), and steam-type (e.g. subcritical vs. supercritical). 7 See appendix A.2 for additional limitations of the final database used. 8 The EIA and EPA provide data on current and historic heat rates for different generators, turbine types, and fuels. Historic EIA data on technology level goes back to 2001 and aggregated data for all fossil fuels back to 1949. 9 Datasets obtained from the EIA contain emission factors for different fuels, i.e. the amount of CO 2 in relation to the energy content of e.g. coal, lignite, oil, etc. 10 CARMA: Carbon Monitoring for Action (CARMA 2010). 11 The online year refers to the year in which the generator started operations.

Approach and methods
Third, actual lifetimes were simulated by using a Poisson distribution around the expected lifetimes of the power generators. This simulates managerial discretion as to when power generators are retired, and accommodates the fact that generators are rarely retired in the exact year of their estimated end of life. As expected lifetimes we use the median end-of-life age of already retired generators with the same fuel and technology, and similar nameplate capacity. Expected lifetime represents the economic rather than the technical lifetime (taking the maximum lifetime of similar already retired generators would come closer to the technical lifetime).
Fourth, instead of applying the CARMA database for the annual emissions of these generators, a different approach was applied. Generator-specific technical data was combined with (year-specific) heat rates and fuel emission factors from the EIA and EPA to calculate annual maximum emissions per generator. When multiplied by the simulated lifetime of each generator, and the historic (average) utilization rates, from the IEA, this results in an estimate of actual historic, current and expected electricity generation and emissions. By using historic average utilization rates over many years (ten years between 2004-2014) instead of a point estimate (CARMA uses 2009 utilization rates), a more realistic estimate of future utilization can be achieved. A detailed description and discussion of this methodology and in particular the use of historic average utilization rates can be found in appendix A.1 available at stacks.iop.org/ERL/13/054019/mmedia. The described approach results in an emission commitment estimate of ∼300 GtCO 2 in 2016. This estimate is ∼14% lower than an extrapolation of Davis and Socolow's results suggests, and implies that currently operating capital stock commits to significantly less future emissions than expected only four years ago. While the methodological differences explain some of the variance, the real-world explanation for this is that, in recent years, since Davis and Socolow's paper, the growth rate of committed emissions was much lower than expected. Between 2012 and 2016 emission commitments grew only by 2.1% p.a. globally instead of the 4% p.a. as expected based on Davis and Socolow's results. In some regions, emission commitments even decreased significantly in this period (see section 3.1). These results are particularly sensitive towards generator lifetimes and utilization rates. We discuss these sensitivities in section 3.4.

Findings
We find that currently operating generators would already commit to more future CO 2 emissions (∼300 GtCO 2 ) than would be consistent with the remaining generation-only carbon budget in the median 430-580 ppm scenarios. For a good chance for warming below 2 • C (430-480 ppm scenarios) ∼20% of currently operating capital stock would have to be stranded. Instead, the pipeline of currently planned generators would add another ∼270 GtCO 2 to the capital stock.

Committed emissions of generators operating in late 2016
In late 2016, a global total of ∼161 000 generators in our database were labelled as operating, idle, stand-by, or with a similar status indicating that a power generator was still in operation (table 1). This comprises ∼6200 GW of installed capacity, which has, on average, operated since 1997, and which had a remaining lifetime of 18 years in 2016 (see appendix C.1 for a full table with descriptive statistics).
Overall, this capacity, if operated over its full remaining lifetime at current utilization rates, could generate ∼537 k TWh (∼23 years of generation at current levels) 13 and would emit ∼300 GtCO 2 over the coming decades (i.e. ∼7 years-worth of current total global CO 2 emissions) 14 . These committed emissions are largely locked-in by coal generators (∼71%) and located in Asian countries (∼64%) 15 . Figure 1 shows the development of committed emissions over time by technologies (panel a) and regions (panel b). After the present day, we show decreasing commitments as they 'realize' into actual emissions. Committed emissions from coal decreased by 1.4% between 2000 and 2003, presumably because coal capacity was replaced by gas, which grew by ∼26.8% in the same period. Most of the growth in committed emissions after 2005, however, comes from coal-fired generators which accounted for 59% of total committed emissions in 2005 and account 12 Cumulative CO 2 emissions that can be expected from the future operation of these generators over an expected economic lifetime under standard economic conditions. 13 According to the IEA the 2014 global electricity generation was 23 808 TWh. 14 According to the Global Carbon Budget Project, total CO 2 emissions (Fossil Fuel and Industries and Land-use) in 2015 were 41 GtCO 2 (Le Quéré et al 2016). 15 See appendix C.2 for the full regional split. 16 Our definition of Asia includes all non-OECD Asian countries (i.e. most Asian countries except the Middle East, Japan and countries of the former Soviet Union. for 71% today. In recent years, Asia 16 has seen an especially strong increase in commitments. In 2000, committed emissions in Asia accounted for approximately one quarter of the global total but this share had increased to almost two thirds in 2016. Especially after 2004, most of the overall growth in emission commitments has come from the addition of fossilfuel-powered generators in Asia. Figure 2 provides more details on annual growth rates of committed emissions. Countries of the former Soviet Union and OECD countries (since 2004) have seen a decrease in overall commitments from power generators, indicated by negative annual growth rates (panel c). This development indicates that annual retirements or realizations of committed emissions are larger than additions to the capital stock. Panel a confirms the previous finding that the growth rates of gas capital stock between 2000 and 2003 crowded out coal infrastructure (negative growth rates) but were subsequently replaced by coal again. The overall annual growth of coal capital stock remains strong in 2016 (∼2.1% p.a.) while all other CO 2 emitting capital stock has decreased over the last couple of years.
We find that, on average, emission commitments from electricity generators grew by 3.2% per year between 2000 and 2016, and that most of that overall growth came from coal generators (3.9% p.a.) and happened in Asia (9.1% p.a.). The only technologies with stronger or similar committed emissions growth rates to coal were bioenergy (4.3% p.a.) and waste (3.6% p.a.). Growth in these technologies took place from a much lower base, however, and was hence negligible for overall committed emissions growth. Besides Asia, countries in Latin America (3.1% p.a.) the Middle East and Africa (2.9% p.a.) experienced committed emissions growth in the analyzed period, while OECD countries (−2.1% p.a.), and countries of the former Soviet Union (REF) (−3.5% p.a.), decreased their remaining committed emissions from electricity capital stock.

The pipeline of planned electricity generators in early 2017
In addition to the previously described operating generators, in early-2017 ∼24 000 further generators were either under construction (845 GW in ∼5200 generators) or in some stage of the planning process (2597 GW in ∼18 900 generators). Overall, this pipeline of generators would add ∼3440 GW to the global capital stock and add ∼270 GtCO 2 to the committed future carbon emissions.
In table 2 we split this pipeline of committed emissions by technologies and regions. Consistent with the development in recent years already illustrated above, by far the largest share of planned committed emissions is occupied by coal (∼78%) and is planned in Asia (∼65%). Just by finalizing all planned coalfired generators, the world would add an additional five years of total global CO 2 emissions at current levels. Gas-fired generators follow with 18% and are expected to add ∼50 GtCO 2 to the global capital stock (∼1.2 years of current total emissions).
If all current plans and construction projects for carbon emitting power generators were to be stopped, however, the remaining committed emissions in 2050 would amount to ∼20 GtCO 2 . If, however, all planned generators were to be built and come online, then remaining commitments in 2050 would be 4-5 times higher (∼90 GtCO 2 ). Figure 3 (panel a) illustrates this. Panel d shows the regional split of the current pipeline. In Asia, almost as much polluting capital stock is planned (119 GtCO 2 ) or already under construction (57 GtCO 2 ) as is currently operating (198 GtCO 2 ) 17 .
These findings consider all generators that were either planned or under construction in early 2017. This should include most changes, especially cancellations in the global coal pipeline, that were made before    Kriegler et al 2015). The analysis of the scenario outputs from these two The dent between 2018 and 2020 in panel (a) stems from the methodology that has been used to assign online year to generators that are delayed (after construction start) or deferred (before construction start) in 2016.
databases suggests that the median remaining carbon budget available in 2005 for a good chance for 1.5 • C-2 • C warming (430-480 ppm scenarios), was 1333 GtCO 2 . According to the same scenarios ∼14% of that budget in 2005 was earmarked for electricity generation, leaving a net generation-only carbon budget in 2005 of ∼187 GtCO 2 . In addition to this net budget, the median 2005-2100 cumulative electricity generation from bioenergy with carbon capture and storage (BECCS) in these scenarios was ∼1330 EJ (∼370 000 TWh). This amount of BECCS generation would remove ∼110 GtCO 2 from the atmosphere by the end of the 21st century 21 , thereby increasing the carbon budget for electricity production. Using the same calculation method for 480-530 ppm and 530-580 ppm scenarios, respectively, and updating these numbers over time with realized annual emissions (Le Quéré et al 2016), results in the annual remaining generation-only carbon budgets illustrated in figure 4. For better comparison, we also include carbon budget estimates from a recently published study which finds that the remaining post-2015 carbon budgets for a 50% chance for 1.5 • C or 2 • C warming were 817 GtCO 2 and 1524 GtCO 2 , respectively (Millar et al 2017) 22 .
We compare these remaining generation-only carbon budgets over time with the development of commitments from operating generators and find that the year in which built infrastructure committed us to enough emissions to reach the 1.5 • C-2 • C budget was in 2011, and hence six years earlier than previously estimated (Pfeiffer et al 2016). In 2014, emission commitments exceeded the remaining carbon budget for 480-530 and 530-580 ppm scenarios.
The above suggests that, if the climate goals set out in the Paris Agreement (UNFCCC 2015) are to be reached, some of the existing and planned power plants will need to be underutilized, retired early, or retrofitted with expensive CCS or efficiency upgrades, or-in short-stranded. Figure 5 illustrates, for different climate goals, all combinations of stranding (in percent of normal utilization) between old (currently operating) and new (planned or under construction) power generators.
We find that, in different combinations, only 42% (430-480 ppm) to 49% (530-580 ppm) of the total capital stock of both operating and planned generators can be utilized. Even if every single currently planned project was cancelled, the generators that are already operating now would still have to see reduced utilization resulting in∼20% of capacity becoming 21 83 MtCO 2 /EJ (Kriegler et al 2013). 22 To calculate generation-only carbon budgets we also multiply these total carbon budgets with 14% and add 110 GtCO 2 BECCS carbon removal.  stranded (∼80% of normal utilization) to meet the 430-480 ppm climate target.
Taking the remaining total global carbon budget as exogenous, the two most important factors in this analysis are the share of this budget that can be allocated to power generation (∼14%) and the additional generation-only budget that is added by BECCS generation. Changing these numbers would change the results of this analysis considerably. Should the share of generation-only budget be one percentage point smaller (∼13%) or bigger (∼15%) the stranding estimates for the 430-480 ppm scenario would change by 1.6 percentage points. In a scenario in which power generation would have only 13% of remaining total carbon budget left (under unchanged future BECCS generation) the 1.5 • C-2 • C compatible average utilization for currently operating and under construction or planned generators would drop from ∼42% to ∼39%. Should generation-only budget be 15% of total instead of 14% possible utilization would increase to ∼44%. The total amount of GHG captured by BECCS is also important. If BECCs turns out to be entirely unable to remove carbon from the atmosphere, the utilization rate of power generators compatible with 1.5 • C-2 • C would drop from 42%-22%.

Sensitivity of findings
Our findings regarding the committed emissions of currently operating and under construction or planned power generators are particularly sensitive towards simulated lifetimes and target utilization rates.
Realised lifetimes of power generators depend on a variety of factors that affect the economic viability of the generator, such as electricity, fuel, and carbon prices, regulation, and technological change (see appendix A.1). At the same time, the future lifetime for a currently operating (or planned) generator has a considerable effect on its remaining emission commitments. Based on a 42 year lifetime for coal generators, every additional year of lifetime would increase the original emission commitments of currently planned coal generators (210 GtCO 2 commitments) by 5 GtCO 2 (+2.4%). For currently operating coal generators (220 GtCO 2 commitments remaining in 2016), each additional year of lifetime increases committed emissions by 10 GtCO 2 (+4.6%) 23 . For gas-fired generators currently operating (emission commitments of 66 GtCO 2 in 2016) every additional year increases emission commitments by 3.3 GtCO 2 (+5%) 24 . For new gas generators each year would increase the emission commitments from 49 GtCO 2 to 50 GtCO 2 (+2%).
Utilization rates, as well, have a significant impact on our results. For instance for coal we apply a global average utilization rate of 61%. Reducing (or increasing) this utilization rate by one percentage point would result in a reduction (or increase) of committed emissions by 4 GtCO 2 (1.6%). For gas the applied utilization rate is 39% and every percentage point change hence a 2.6% increase or decrease in committed emissions. For a further discussion of utilization rates please see appendix A.1.

Discussion of findings
We analyze the expected (business as usual) cumulative carbon emissions from currently operating and planned power generators around the world and find that this capital stock would likely emit more CO 2 than compatible with the median scenarios that would meet current climate goals. Moreover, we estimate that commitments reached the remaining carbon budget for 1.5 • C-2 • C warming in 2011; with the carbon budget for 2 • C-3 • C warming being breached in 2014. Despite making similarly conservative assumptions with regards to decarbonization in other sectors, this finding updates an earlier estimate in which we identified 2017 as the year in which operating capital stock would commit us to 2 • C (Pfeiffer et al 2016). The changes compared to the previous finding come from updated, and more accurate, carbon budget figures as well as an update of the previously used power generator database. Supplemental databases add power generators (especially in China) and close known gaps, thereby improving the representation of the global generation capital stock.
The updated findings suggest that much of the global electricity generation capital stock would need to become stranded if the world were to meet its climate goals. This result postulates that power generation is assigned the same share of the overall carbon budget as in the median pathway and that future BECCS generation can add the expected atmospheric space. Under these conditions, some stranding would occur (∼10 to 20% of operating capacity) even if all current plans and construction projects for additional power generators would be suspended. This stranding would likely have the strongest impact on the coal sector in Asia, where 64% of current and 65% of planned committed emissions are located, most of it in coal-fired generators.
Committed emissions depend on future lifetimes and utilization rates of existing and newly build power plants. Shorter realised lifetimes or lower utilization rates would reduce remaining emission commitments of operating and planned generators considerably. Indeed, developments in recent years point towards decreasing utilization rates, at least for coal-fired power generators. In the context of this analysis, lower utilization rates would constitute stranding.
The stranding of power generation assets can have several causes and materializes in different ways (Caldecott et al 2016b). Among the most important causes for stranding in power generation are changing regulations (e.g. emission standards), higher input costs (e.g. rising prices for coal, gas and CO 2 permits), and changing market conditions (e.g. falling wholesale prices). Regulatory and technological efforts to keep within carbon budgets compatible with the Paris Agreement will result in significant stranding of both operational and planned fossil fuel power generation. The extent to which this affects existing assets, or those currently in planning, is largely a market and policy question. Regardless of where the stranding occurs, however, it will generate significant social and political economy impacts. Power plant owners, operators, connected communities and investors will be affected, but so too will producers of coal and gas upstream. These different groups, whether directly or indirectly, might have the political power to block policy reforms (Caldecott et  Options to avoid stranding if carbon budgets are inflexible are limited: the carbon budget 'allocated' to the power sector could in principle be expanded, but the power sector appears to be the one that is technically easiest to decarbonize (Clarke and Jiang 2014, Audoly et al 2017. Another radical solution around the issue of stranding coal power plants could be to relax climate goals (Guivarch and Hallegatte 2013), but that would be at odds with the Paris Agreement and result in elevated climate risk for the most vulnerable countries (Stern 2007, IPCC 2014. Our findings may help investors and companies to consider stranding risks and materialization scenarios in their capital allocation decisions. In recent years, the interest within the financial community for such evaluation frameworks and scenario assessments has increased (CTI 2013, Caldecott et al 2015, Carney 2015. Some recent developments in the global power generation sector, such as the cancellation of ∼130 GW of planned coal-fired generators in China, might have been motivated in part by the realization that said capacity could be at risk of becoming stranded if renewables continue to grow at high rates. The substantial pipeline of fossil-fuel powered generators, however, suggests that these risks are still not sufficiently considered (or considered sufficient). Furthermore, the trade-off between the stranding of currently operating and yet to be built generators imposes challenges for investors with broader generation portfolios. Under the constraint of a carbon budget, the optimization of such portfolios might include the stranding of old in favour of new, more efficient, generators, extended lifetimes for old instead of building new generators, the retrofit of some generators with efficiency enhancing or CCS technology, and the shifting of future capacity additions towards lowcarbon technologies (such as renewables and, maybe, gas).
Our findings may also help policymakers improve the set of economic incentives for different types of generation infrastructure. Any further additions to the polluting generation capital stock increase the cost that will need to be paid to achieve the agreed climate goals in the. Efficient and effective policies would incentivize investors to optimize their portfolios to meet carbon budgets, and shift current and future investments towards low-carbon technologies. In the meantime, regulation, such as emission standards, coal moratoriums, and emission levies could help to avoid any further carbon lock-in in the electricity sector and to un-commit some of the budget by decommissioning old and, particularly, dirty generators. Longer lifetimes, and maybe even subsidies for existing and relatively clean generators, on the other hand, could also help reduce the need for additional dirty infrastructure.

Conclusion
Current carbon emission commitments exceed the remaining carbon budget for the electricity generation sector if the world is to meet its climate goals. Nonetheless, the sector will see large amounts of carbon emitting infrastructure being added to its capital stock over the next few years. Investors should re-assess their investment decisions in dirty infrastructure, and policy makers should design their policies to avoid any further carbon lock-in that will prove costly in the future when emissions must decrease to meet climate targets. While long-term policies are not yet in place, some short-term measures, such as emission standards, coal levies and moratoriums, and even lifetime extensions for relatively clean fossil-fuel powered generators could help to avoid further dirty investments.