Energy and climate impacts of digital technologies
[Music]
[Applause]
[Music]
first of all a big thank you to the
organizers at climate change ai for
inviting me and for all their hard work
in putting this event together
today i’ll be speaking about the energy
and climate impacts of digital
technologies drawing on our work at the
iaea
and all the great research being
conducted by colleagues all around the
world
first a bit of context on where we are
today in terms of our greenhouse gas
emissions
and where we need to go some of you
might already be familiar with charts
like this one which show just how
quickly emissions have grown over the
past few decades
this one shows carbon emissions from the
energy sector this includes everything
from oil and gas production to
electricity generation
to the energy used in transportation
buildings and industries
these are all the sectors that we cover
at the iaea
this year carbon emissions are expected
to fall by around seven percent because
of the covet crisis
or around two and a half billion tons
which would be the biggest
annual drop in history but what’s really
daunting is that emissions need to fall
at almost the space every year for the
coming decades for us to reach our
climate goals
for around 1.5 degrees that means net
zero by 2050.
and for around 1.7 degrees that’s net
zero by 2070.
that means we need big structural
changes to policies and infrastructure
since the kind of behavioral and
economic changes we saw in 2020 just
aren’t sustainable and came at an
incredible cost to human life
and to society
with government spending trillions in
recovery packages we really are at a
critical turning point
if governments invest in clean energy we
still have a chance at a 1.5 degree
world
but if they choose to invest in
polluting technologies and industries
emissions are likely to bounce back like
we’ve seen after other drops in the past
and our chances that even a two degree
world could slip away
so you might be wondering which sectors
and activities are most responsible for
greenhouse gas emissions
about three quarters come from the
energy sector and industrial processes
the rest come from agriculture forestry
land use and waste
within the energy sector emissions come
from many different sectors and services
that use fossil fuels
industries buildings transportation coal
and gas power and fossil fuel production
now with that bit of background let’s
focus on digital technologies and how
they impact
the climate digital technologies have a
very real carbon footprint
from the energy used to power the data
centers data networks computers and
smartphones that we use every day
and how they’re used to affect emissions
from other sectors for example how gps
and smartphones have enabled new
transport services like uber and lyft
and in turn these changes can have
secondary effects on other sectors and
services in our built environment
let’s first look at the direct energy
use and emissions from digital
technologies
one thing about digital technologies is
that they’re growing very quickly
compared to other economic and energy
indicators
since 2000 our population has grown by
around a quarter
gdp has almost doubled electricity use
is up almost 70
while the number of internet users has
grown by more than ten times
and global internet traffic has grown by
more than two thousand times
so it kind of makes sense that there is
growing concern over the
energy and environmental impacts of
digital technologies
here’s a headline from the guardian a
couple of years ago saying that a
tsunami of data could use a fifth of
global electricity by 2025
and these kinds of headlines have been
around for a long time
here’s one from 1999 that predicted that
the internet would use half the u.s
electricity supply by 2010
and for that reason the article argued
that the u.s should build
more coal-fired power plants
but in reality the latest research shows
that energy use and emissions from the
ict sector have actually been flat over
the past decade
researchers in u.s published this paper
in science earlier this year showing how
global data center energy use has been
flat since 2010
since 2010 internet traffic is up 12
times
and demand for data center services is
up seven and a half times
but data center energy use has been flat
at around 200 terawatt hours
which is around one percent of global
electricity use
data transmission networks use another
one
there are really two main drivers that
explain how data demand and energy use
trends have diverged
first the energy efficiency of computing
and data transmission has been doubling
every two to three years
and for data centers there’s also been a
massive shift from smaller less
efficient corporate data centers to much
more efficient cloud and hyper-scale
data centers
and these big data centers are getting
even more efficient a few years ago
deepmind applied machine learning to cut
energy use for cooling by 40
so the industry is doing a lot in terms
of energy efficiency to reduce the
energy that it needs
it’s also doing a lot to make sure that
the remaining energy needs are supplied
with clean electricity
the big tech companies are the biggest
corporate buyers of renewable
electricity in the world
apple and google have reached 100
renewables for a few years now
and facebook is expected to reach 100
renewables later this year
what that means is that they’re buying
enough renewable energy each year to
match the amount of electricity that
they’re using in their operations
so for every terawatt hour of
electricity that they’re using
they’re buying a terawatt hour of
renewables that doesn’t mean that their
operations are fully powered by
renewables
all the time data centers run 24 hours a
day but
solar and wind aren’t available all the
time
that means that they still need to rely
at least some of the time on the grid
and that
it that it means it really depends on
where they operate
you can see on this map from google that
some of their sites like in oklahoma run
almost exclusively on clean electricity
but in other places like singapore it’s
mostly gas
overall they’ve estimated that they’re
at around 60
clean energy on an hourly basis
google recently announced that they’re
aiming to get to 100 carbon free
electricity
24 7 by 2030.
that means they’ll need to get enough
zero carbon electricity at each site for
every hour
of the year that might mean using
machine learning
to better forecast when and where
renewables will be available
so that they can shift some workloads to
different times of the day
or even different sites to take
advantage of renewables being generated
so far we’ve focused mostly on the use
phase of digital technologies
but there are impacts throughout the
life cycle of all products from
raw materials extraction and
manufacturing all the way to end of life
for smartphones for example the use
phase is actually a small part of its
overall environmental impact
around 80 percent comes from material
extraction in the manufacturing phases
and there are some other environmental
impacts beyond energy use and greenhouse
gas emissions
like impacts on land air water
biodiversity and of course electronic
waste
so we’ve talked a bit about data centers
networks and devices
but what are the specific impacts of
individual digital activities
you might have seen headlines like this
one over the past year
which said that watching a half hour
show emits as much
co2 as four miles of driving
it turns out the numbers in those
headlines were 40 to 80 times too high
of course the numbers depend on how your
electricity is generated
and whether you’re watching on a big
screen tv or a smartphone
since a big tv uses about 100 times more
electricity than a phone
but on average a half hour show probably
emits around 20 grams of co2
which is around the same as driving 100
meters in a combustion engine
car the main takeaway here is that
emissions from streaming video are
actually very small compared to other
everyday activities
and definitely much lower than driving
to a movie theater
or buying or renting a dvd
but of course it’s become a lot easier
to watch more
so there’s a big rebound effect we also
need to take into account
so going back to this could this
headline actually come true over the
next five years
that number actually comes from a 2015
study that estimated
ict energy use and emissions out to
the big challenge with long-term energy
projections for digital technologies
is that the technologies and their uses
change quickly
and they’re very hard to predict that
same author’s latest projections
published this year
are four times lower than his original
projections that were quoted in the
guardian
in just the past 10 years we’ve seen
smartphones become mainstream
new generations of mobile networks much
more streaming media
and huge efficiency improvements to keep
energy demand in check
and now we’re seeing emerging
technologies like blockchain machine
learning and 5g
and it’s really hard to predict how new
applications that are built on top of
these technologies could impact data
demand and energy use in 10
20 or even 30 years
let’s shift gears a bit and talk about
some of the effects that digital
technologies could have on other sectors
a good example here is teleworking
something more people are doing these
days
because of kovid generally the direct
energy use of digital technologies is
about the same whether you’re at home or
in the office
so most of the net effects are actually
going to come from using more energy at
home for heating or cooling
offset by energy saved from not
commuting
the net effects really depend on your
situation
if you live in a big home and it’s
winter time you’re probably going to use
a lot of energy for heating
and the amount of energy you save from
not commuting will depend on how far you
travel
and whether you drive take public
transit or walk or cycle
and whether the office saves energy
really depends on how often you telework
because an office that’s half full is
still going to be using a lot of energy
to keep it running
there are also longer term effects that
are difficult to predict
somebody that can telework all the time
might choose to move outside the city
where houses are cheaper to buy and that
might mean that they can buy a bigger
home or get a car
which means they’re probably going to be
using more energy overall
well the net effects of day-to-day
teleworking really depend on the
situation
one thing that’s for sure is that
replacing a business flight with video
conferencing can save a lot of energy
and emissions
a couple years ago we published a report
looking at how digitalization could
impact the energy sector
we found a whole range of applications
that could cut costs improve efficiency
and help reduce emissions
in buildings these are things like smart
building controls and thermostats that
can help optimize energy use
in transport we looked at automated
vehicles and how they could slash energy
use if they’re shared in electric
but could also drastically increase
energy use and emissions if they’re just
if they displace public transport and
are not shared or electric
in industry we looked at how 3d printing
and digital twins could accelerate
innovation and
reduce material and energy use
in electricity sensors and automation
are already being used to improve
efficiency and reduce maintenance costs
machine learning is being applied to
forecast solar and wind and better match
supply and demand from
increasingly decentralized sources
and in oil and gas i just saw a paper
published the other day that applied
machine learning to cut the costs of
detecting methane leaks
i think it’s worth highlighting here
that how digital technologies are
applied in these sectors will really be
shaped by climate policy
it’s important to remind ourselves that
digital technologies are agnostic to
climate
while we hope that they’re applied only
to reduced emissions they could just as
easily be applied to increase emissions
like through extending the lifetime of
coal plants or accelerating the
extraction of oil and gas
what we need are strong climate policies
that to make sure that digitalization
helps to combat climate change
and not make it worse
before i wrap up i just want to
highlight a couple key points about the
challenge ahead for decarbonizing the
energy system
first is that emissions today come from
many different sources
and that means we can’t pick and choose
which sectors to decarbonize or which
clean energy technologies to use
because we need to use all of them to
carbonize all sectors
as quickly as possible and the challenge
with the energy sector is that energy
infrastructure lasts
for a long time especially when we
compare them to digital infrastructure
and hardware
coal-fired power plants have a lifetime
of more than 50 years
and that means a coal plant that’s being
built today in asia could
still be running in 2070.
a new gasoline car that’s coming off the
assembly line could be driving around
in 20 years
the other point is that while dealing
with existing assets we also need to
rapidly scale up existing
and new clean energy technologies this
is an area where i think digital
technologies could play a huge role in
accelerating innovation cycles with
technologies like machine learning and
digital twins
to reach net zero by 2070 about a third
of emission reductions will come from
technologies that aren’t yet commercial
today
and to get to net zero by 2050 almost
half of the emission reductions come
from these technologies
and for hard to abate sectors like heavy
industry and long distance transport
three quarters of the reductions come
from these early stage technologies
so to wrap it up there’s a lot we can do
as individuals researchers and
practitioners to reduce the carbon
footprint of our digital activities
but there’s even more that governments
can do with strong
enabling climate policies to make sure
that digital technologies help to
accelerate climate action
as individuals we can replace our
devices less frequently
and watch our shows on smaller screens
when we can
and there’s a lot we can do with our
transport choices especially flying
and of course our diets but i think the
most important thing we can do as
citizens
is to support climate champions whether
that’s through voting
supporting ngos or buying from
responsible companies
as ml practitioners and engineers
there’s a lot you can do in your
day-to-day work
both in terms of reducing your own
emissions but also figuring out how your
skills could help tackle climate change
in other sectors
governments perhaps have the biggest
role to play through ambitious and
concrete action on climate change that
encourages tech
for good and finally there’s a lot that
tech companies can do
not just reducing their own emissions
from their data centers and products
but figuring out how to use their
cutting edge technologies and vast
resources to accelerate the global clean
energy transition
there’s a lot of potential for digital
technologies to help tackle the climate
crisis
and we need everyone citizens experts
companies
and governments to do their part to help
ensure a more sustainable future
thank you