1Department of Mechanical Engineering, Northwestern University, Evanston, IL, USA. 2Department of Chemical and Biological Engineering, Northwestern
University, Evanston, IL, USA. 3Energy Analysis & Environmental Impacts Division, Lawrence Berkeley National Laboratory, Berkeley, CA, USA. 4Faculty of
Management, Science & Technology, Open University of the Netherlands and Radboud University, Nijmegen, The Netherlands. 5Rocky Mountain Institute,
Basalt, CO, USA. 6Ericsson Research, Ericsson AB, Stockholm, Sweden. *e-mail: email@example.com
Bitcoin mining is an increasingly energy-intensive process1–3 for
which the future implications for energy use and CO2 emissions
remain poorly understood. This is in part because—like many IT
systems—its computational efficiencies and service demands have
been evolving rapidly. Scenario analyses that explore these implica-
tions can therefore fill pressing knowledge gaps, but they must be
approached with care. History has shown that poorly constructed
scenarios of future IT energy use (often a result of overly simplistic
extrapolations of early rapid growth trends) can spread misinfor-
mation and drive ill-informed decisions4–6. Indeed, the utility of an
energy demand scenario is proportional to its credibility, which is
demonstrated through careful attention to technology characteris-
tics and evolution, analytical rigour and transparency, and design-
ing scenarios that align with plausible future outcomes.
While we believe that Mora etal. had the right motivations in
developing Bitcoin CO2 emissions scenarios7, we respectfully argue
that their scenarios lack such credibility. We arrived at our conclu-
sion by replicating Mora and colleagues’ methods in detail, which
revealed key flaws in the design and execution of their analysis (as
documented in the Supplementary Information). We describe the
five most important issues below.
First, the use of transactions as the driver of future Bitcoin emis-
sions is questionable, given the tenuous correlation between trans-
actions and mining energy use. It is well established that energy use
is driven by the computational difficulty of the blocks mined1–3,
whereas the number of transactions per block can evolve (for exam-
ple, via SegWit)8 with no direct effect on block mining difficulty.
The authors themselves7 calculate Bitcoin energy use and emissions
in 2017 on the basis of block difficulty, not the number of trans-
actions (Supplementary Equation (1)). Without explanation, the
authors switch to transactions as the driver for projecting future
emissions, undermining their methodological consistency and the
integrity of their projections.
Second, all three Bitcoin adoption scenarios designed by Mora
etal. represent sudden and improbable departures from historical
trends in Bitcoin transactions; over the preceding five years annual
growth ranged from 1.3× to 2.3× (Supplementary Figs. 3 and 4)9.
Specifically, Mora et al. assume that Bitcoin transactions—which
totalled 104 million in 2017, representing a mere 0.03% of global
cashless transactions—would abruptly leap to 78 billion by 2019 in
the fast scenario (a 750× increase in only 2 yr), to 11 billion by 2020 in
the median scenario (a 108× increase) and to 8 billion by 2023 in the
slow scenario (a 76 × increase). All three adoption scenarios follow
steep logarithmic growth trajectories thereafter, which are conspicu-
ously inconsistent with historical trends (Supplementary Fig. 4) and
mathematically can only lead to large near-term emissions increases.
The authors base their scenarios on adoption rates of 40 arbitrarily
selected technologies, the social utilities of which vary widely. The
authors do not explain why such comparisons are valid, nor do they
justify the plausibility of the very abrupt changes in Bitcoin transac-
tion levels and growth trajectories that result from such comparisons.
Third, Mora etal. applied outdated values for mining rig efficien-
cies and electric power CO2 intensities, which inflated their estimated
2017 Bitcoin energy use and CO2 emissions values considerably. When
estimating the direct electricity use of Bitcoin mining, the authors
included in their selection pool many old and inefficient rigs that
were no longer economically viable in 2017 (Supplementary Fig. 5).
Furthermore, Mora etal. provided equal weighting when selecting
a rig from their pool as the sole rig type to mine a block, thus over-
representing slower, inefficient rigs and creating scenarios that require
physically impossible rig counts. When we excluded unprofitable rigs
in our replicated analysis, Mora and colleagues’ model produced an
estimate of 28 TWh in 2017 (Supplementary Fig. 6), which is one-
quarter of their original estimate of 114 TWh. Furthermore, they
applied 2014 CO2 intensities (in gCO2 kWh−1) to calculate 2017 emis-
sions, ignoring non-negligible grid decarbonization improvements
in the intervening years (Supplementary Fig. 7)10, despite sufficient
data being available at the time of their study for reasonable estimates
of 2017 power mixes11,12. Applying more reasonable 2017 electricity
use and CO2 intensity values in their model produced an estimate of
15.7 MtCO2e, far lower than their original estimate of 69 MtCO2e.
Fourth, by analytical design, Mora etal. applied 2017 per-trans-
action energy use and CO2 emissions values in all future years,
multiplied by annual transactions (Supplementary Equation (2)).
This decision effectively held both mining rig efficiency and grid
CO2 intensities constant for the next 100 yr (Supplementary Fig. 7).
This unprecedented choice ignores the dynamic nature of mining
rig and power grid technologies and violates the widely followed
practice of accounting for technological change in forward-looking
energy technology scenarios10,11. In acknowledging their static grid
intensity assumption, they point to at least one reference containing
credible grid intensity outlooks10 but failed to make use of them.
Estimating the future energy efficiency of mining is certainly more
difficult, but the authors do not explain why they simply ignored this
Implausible projections overestimate near-term
Bitcoin CO2 emissions
Eric Masanet 1,2*, Arman Shehabi3, Nuoa Lei1, Harald Vranken4, Jonathan Koomey5 and
arising from Mora, C. etal. Nature Climate Change https://doi.org/10.1038/s41558-018-0321-8 (2018)
NATURE CLIMATE CHANGE | VOL 9 | SEPTEMBER 2019 | 653–654 | www.nature.com/natureclimatechange 653