The main public estimates of data center electricity use do not break out short-form video. Figures from the International Energy Agency (IEA), the U.S. Department of Energy (DOE) and platform engineers do show where that load sits, and why it is hard to separate from the wider rise in data center demand.

Data centers used about 485 terawatt-hours (TWh) of electricity in 2025, and the IEA projects that figure will roughly double to 950 TWh in 2030, around 3% of global demand, according to its 2026 report Key Questions on Energy and AI. The agency said consumption at data centers focused on artificial intelligence (AI) surged 50% in 2025.

In the United States, a Lawrence Berkeley National Laboratory report released by DOE in December 2024 found that data center use climbed from 58 TWh in 2014 to 176 TWh in 2023, about 4.4% of national electricity. It estimated 325 to 580 TWh by 2028, or roughly 6.7% to 12% of the U.S. total.

The IEA’s 2025 analysis of data center demand projected electricity use in accelerated servers, mainly driven by AI adoption, to grow 30% a year in its base case, against 9% a year for conventional servers. Accelerated servers made up almost half of the projected net increase and conventional servers around 20%. That breakdown is by equipment type, so video has no line of its own.

Where a viewing hour’s electricity goes

One public breakdown comes from a December 2020 IEA commentary by former IEA energy policy analyst George Kamiya. It estimated that streaming a Netflix video in 2019 typically used about 0.077 kilowatt-hours (kWh) per hour, with viewing devices taking 72%, data transmission 23% and data centers 5%. On those shares, the data center portion comes to roughly 0.004 kWh per viewing hour.

The mix changes on a phone. Because phones are so energy efficient, the commentary said, data transmission accounts for more than 80% of electricity use when streaming on one, and an hour of standard-definition video on a phone over Wi-Fi uses about 0.037 kWh. Those figures describe long-form viewing in 2019, so they are a reference point for short clips rather than a measurement of them.

What short clips add on the server side

Short clips also create work before anyone presses play. Engineers at Meta wrote in February 2023 that Facebook and Instagram Reels generate multiple bit-rate encodings for each upload, tailored to projected watch time, so a viewer’s app can pick the version that suits its connection.

For videos with high projected watch time, Meta said, its system might test seven resolutions at five quality settings on a 1080p upload, for 35 encodings. That first pass keeps only quality and bit rate data. A second encoding stage runs when projected watch time passes a threshold, so more compute goes to the clips expected to draw the most viewing.

Storage is the other standing cost. An ordinary personal profile such as mariasuvalova18 on TikTok can hold dozens of short clips, and each one has to stay stored and ready to stream. The IEA says storage systems account for around 5% of data center electricity consumption and networking equipment for up to 5%, while servers average around 60%. Cooling ranges from about 7% at efficient hyperscale sites to over 30% at less efficient enterprise facilities.

Compression cuts both ways. Meta wrote that each new video coding standard typically reduces bandwidth by about 30% to 50% at similar visual quality, but has consumed substantially more energy and compute than the last. After Meta deployed the AV1 codec, the average bit rate for Facebook Reels played on iPhones fell 12%, the company reported.

Why it matters

For utilities and grid planners, the IEA and DOE forecasts cover data centers as a whole and do not separate video from other services. Short-form viewing therefore cannot be read directly from those load projections. A change in viewing habits also would not map one-for-one onto data center demand, since most of a typical viewing hour’s electricity in the IEA’s 2019 estimate was used outside the data center.

For platforms, the levers described in these sources sit on the server side: how many versions of a clip to encode, which codec to use and how much to keep ready for delivery. Meta’s account treats those as trade-offs among compute, storage and network traffic rather than a single efficiency gain.

The caveats

The IEA’s per-hour estimate uses 2019 Netflix data, and its author wrote that the results are highly sensitive to viewing device, network connection and resolution. Meta’s post describes its own systems and does not report electricity use. None of the sources reviewed here gives an energy figure for a short-form platform.

Forecasts carry their own uncertainty. The IEA’s 2025 analysis said there is substantial uncertainty about data center consumption both today and in the future, and its 2035 cases run from around 700 TWh to more than 1,700 TWh.