A version of this commentary was originally published in Carbon Brief.
Even before millions were confined to their homes by a global pandemic, improvements in internet connections and service offerings had led to an exponential increase in the use of streaming video around the world. With few options left for entertainment, streaming services are taking off. In this commentary, we examine the carbon footprint of these services.
Streaming services are associated with energy use and carbon emissions from devices, network infrastructure and data centres. Yet, contrary to a slew of recent misleading media coverage, the climate impacts of streaming video remain relatively modest, particularly compared to other activities and sectors.
Drawing on our analysis and other credible sources, we expose the flawed assumptions in one widely reported estimate of the emissions from watching 30 minutes of Netflix. These exaggerate the actual climate impact by 30- to 60-times.
The relatively low climate impact of streaming video today is thanks to rapid improvements in the energy efficiency of data centres, networks and devices. But slowing efficiency gains, rebound effects and new demands from emerging technologies, including artificial intelligence (AI) and blockchain, raise increasing concerns about the overall environmental impacts of the sector over the coming decades.
A number of recent media articles, including in the New York Post, CBC, Yahoo, DW, Gizmodo, Phys.org and BigThink, have repeated a claim that “the emissions generated by watching 30 minutes of Netflix [1.6 kg of CO2] is the same as driving almost 4 miles”.
The figures come from a July 2019 report by the Shift Project, a French thinktank, on the “unsustainable and growing impact” of online video. The report said streaming was responsible for more than 300m tonnes of CO2 (MtCO2) in 2018, equivalent to emissions from France.
The Shift Project’s “3.2kgCO2 per hour” estimate is around eight times higher than a 2014 peer-reviewed study on the energy and emissions impacts of streaming video.
That 2014 study found streaming in the US in 2011 emitted 0.42kgCO2e per hour on a lifecycle basis, including “embodied” emissions from manufacture and disposal of infrastructure and devices. Emissions from operations – comparable in scope to the Shift Project analysis – accounted for only 0.36kgCO2e per hour.
Looking at electricity consumption alone, the Shift Project figures imply that one hour of Netflix consumes 6.1 kilowatt hours (kWh) of electricity. This is enough to drive a Tesla Model S more than 30km, power an LED lightbulb constantly for a month, or boil a kettle once a day for nearly three months. At this rate, streaming all six seasons of House of Cards (61 hours) would consume as much electricity as five weeks of total consumption for an average European household.
With users collectively watching at least 165m hours per day, the Shift Project figures imply that Netflix streaming consumes around 370 terawatt hours (TWh) per year, which would be comfortably more than the annual electricity demand of the UK.
For comparison, these figures are 800-times larger than figures reported by Netflix (0.45TWh in 2019) and nearly double the estimated electricity use by all data centres globally (198TWh in 2018). It is clear that the Shift Project figures are too high – but by how much?
First, it overestimates bitrate, the amount of data transferred each second during streaming, apparently assuming a figure of 24 megabits per second (Mbps), equivalent to 10.8 gigabytes (GB) per hour. This is six times higher than the global average bitrate for Netflix in 2019 (around 4.1 Mbps or 1.9 GB/hr, excluding cellular networks) and more than triple the transfer rate of high-definition (HD, 3 GB/hr). Other typical transfer rates are 7 GB/hr for ultra-high definition (UHD/4K) and 0.7 GB/hr for standard definition (SD).
(In part, this difference stems from a stated assumption of 3Mbps apparently being converted in error to 3 megabytes per second, MBps, with each byte equivalent to eight bits.)
The chart below shows each of three ways that the Shift Project overestimated electricity use for streaming video – such as the bitrate – and one area where it underestimated the actual figure. These other errors are described in the text below the chart.
Second, I estimate that the Shift Project analysis overestimates the energy intensity of data centres and content delivery networks (CDN) that serve streaming video to consumers by 7- to 18-fold, relative to figures derived from 2019 Netflix electricity consumption data, Cisco traffic data and IEA analysis.
Third, my analysis shows the Shift Project overestimates the energy intensity of data transmission networks by 6- to 17-fold. This is the result of high energy-use assumptions for various access modes – for example, 0.9 kWh/GB for “mobile” compared to my estimate of 0.1-0.2 kWh/GB for 4G mobile in 2019. The Shift Project also assumes a higher share of data transfer through more energy-intensive mobile networks compared to WiFi, which it puts at one-third of the total compared to less than 10% based on Netflix data.
However, the Shift Project underestimates the energy consumption of devices by some 4- to 7-fold, because it assumes that viewing occurs only on smartphones (50%) and laptops (50%). According to Netflix, 70% of viewing occurs on TVs, which are much more energy-intensive than laptops (15% of viewing), tablets (10%), and smartphones (5%).
Taken together, my analysis suggests that streaming a Netflix video in 2019 typically consumed 0.12-0.24kWh of electricity per hour, some 25- to 53-times less than estimated by the Shift Project, as shown in the chart, below left. The results are highly sensitive to the choice of viewing device, type of network connection and resolution, as shown in the chart, below right.
For example, a 50-inch LED television consumes much more electricity than a smartphone (100 times) or laptop (5 times). Streaming through 4G mobile networks consumes about four times as much electricity than through WiFi.
Because phones are extremely energy efficient, data transmission accounts for nearly all the electricity consumption when streaming through 4G, especially at higher resolutions (Scenario D). Streaming an hour-long SD video through a phone on WiFi (Scenario C) uses just 0.037 kWh – 170 times less than the estimate from the Shift Project.
The carbon footprint of streaming video depends first on the electricity usage, set out above, and then on the CO2 emissions associated with each unit of electricity generation.
Powered by the current global average electricity mix, streaming a 30-minute show on Netflix would release 0.028-0.057kgCO2e (28-57 grammes, second column in the chart, below). This is some 27- to 57-times less than the 1.6kg figure from the Shift Project (leftmost column), which was compared with driving four miles (6.4 kilometres).
To put it in context, my updated estimate for the average carbon footprint of a half-hour Netflix show is equivalent to driving around 200 metres in a conventional car.
But as the chart above shows, this figure depends heavily on the generation mix of the country in question. In France, where around 90% of electricity comes from low-carbon sources, the emissions would be around 4gCO2e, equivalent to 20 metres of driving.
Using country average emission factors may still overestimate emissions, particularly from data centres. Technology firms operating large data centres are leaders in corporate procurement of clean energy, accounting for about half of renewable power purchase agreements in recent years.
The electricity mix is also rapidly decarbonising in many parts of the world. For instance, the emissions intensity of electricity in the UK fell by nearly 60% between 2008 and 2018. Compared to 2018 levels, global emissions intensity of electricity falls by around one-fifth by 2030 in the IEA Stated Policies Scenario and by half in the Sustainable Development Scenario.
Although the carbon footprint of streaming video remains relatively modest, it might still seem reasonable to expect the overall impact to rise, given exponential increases in usage.
However, there have already been major improvements in the efficiency of computing, described by “Koomey’s Law”. This law describes trends in the energy efficiency of computing, which has doubled roughly every 1.6 years since the 1940s – and every 2.7 years since 2000. A similar trend has been observed in data transmission networks, with energy intensity halving every two years since 2000.
Coupled with the short lifespans of devices and equipment, which hastens turnover, the efficiency of the overall stock of devices, data centres and networks is improving rapidly.
For example, increasingly efficient IT hardware (following Koomey’s Law) and a major shift to “hyperscale” data centres have helped to keep electricity demand flat since 2015 (chart, below right). Data centres worldwide today consume around ~1% of global electricity use, even while internet traffic has tripled since 2015 and data centre “workloads” – a measure of service demand – have more than doubled (chart, below left).
Many new video streaming and cloud gaming services have also launched in recent months. Particularly noteworthy is the rapid growth in video traffic over mobile networks, which is growing at 55% per year. Phones and tablets already account for more than 70% of the billion hours of YouTube streamed every day.
The ease of accessing streaming media is leading to a large rebound effect, with overall streaming video consumption rising rapidly. But the complexity of direct and indirect effects of digital services, such as streaming video, e-books, and online shopping, make it immensely challenging to quantify the net environmental impacts, relative to alternative forms of consumption.
Moreover, emerging digital technologies, such as machine learning, blockchain, 5G, and virtual reality, are likely to further accelerate demand for data centre and network services. Researchers have started to study the potential energy and emissions impacts of these technologies, including blockchain and machine learning.
It is becoming increasingly likely that efficiency gains of current technologies may be unable to keep pace with this growing data demand. To reduce the risk of rising energy use and emissions, investments in RD&D for efficient next-generation computing and communications technologies are needed, alongside continued efforts to decarbonise the electricity supply.
Streaming video is a fairly low-emitting activity, especially compared to driving to a cinema, for instance. As consumers, we can further reduce our environmental footprint by streaming at lower resolutions, using smaller devices and screens, and connecting through WiFi instead of mobile networks. Replacing devices less often can also help, since production accounts for more than two-thirds of the lifecycle carbon emissions of mobile devices, and electronic waste is a growing problem across the world.
Technology companies can continue to play a big role in reducing the environmental impact of streaming, including through further efforts to increase energy efficiency – both in the near-term with new technologies and developing next-generation technologies – and investing in renewable energy to power their data centres and networks.
Sustainable design and coding could also help, such as further improving video compression. A recent study explored the potential energy and emission reductions of shifting YouTube music videos to audio only when playing in the background.
It is important to keep in mind the scale of emissions from digital technologies compared to other sectors, with digital technologies accounting for around 1.5% of global carbon emissions.
All sectors and technologies are needed to help achieve the goals of the Paris Agreement and digital technologies are no exception. In fact, digital technologies, such as AI, could help accelerate climate action. But without sound climate policies, AI could end up just helping to make oil extraction cheaper or extending the lifetime of coal plants.
What is indisputable is the need to keep a close eye on the explosive growth of Netflix and other digital technologies and services to ensure society is receiving maximum benefits, while minimising the negative consequences – including on electricity use and carbon emissions.
Instead of relying on misleading media coverage, this will require rigorous analysis, corporate leadership, sound policy and informed citizens.
The analysis of the carbon intensity of streaming video presented in this piece is based on a range of sources and assumptions, calculated for 2019 or the latest year possible.
- Bitrate: global weighted average calculated based on subscriptions by country and average country-level data streaming rates from Netflix in 2019; resolution-specific bitrates from Netflix.
- Data centres: low estimate based on Netflix reported direct and indirect electricity consumption in 2019, viewing statistics and global weighted average bitrate (above); high estimate based on 2019 cloud data centre IP traffic from Cisco and energy use estimates for cloud and hyperscale from IEA.
- Data transmission networks: calculations based on Aslan et al. (2018), Schien & Priest (2014), Schien et al. (2015), and Andrae & Edler (2015), and weighted based on Netflix viewing data by devices.
- Devices: smartphones and tablets: calculations based on Urban et al. (2014) and Urban et al. (2019), iPhone 11 specifications (power consumption and battery capacity), and iPad 10.2 specifications; laptops: Urban et al. (2019); televisions: Urban et al. (2019) and Park et al. (2016), and weighted based on Netflix viewing data by devices.
- Carbon intensity of electricity: based on IEA country-level and global data, and 2030 scenario projections.