The State of Streaming Sustainability 2025

As we near the end of the first quarter of 2025, both our industry and the world in general seem to be at an inflection point. Clean energy and sustainability continue to be tarred with negative propaganda, while at the same time, the thirst for raw computing power—used to power such things as cryptocurrency, ma­chine-learning compute (aka AI), and our industry’s own edge compute and live-streaming growth—con­tinues to increase.

In this article, I’m going to look at a few of the un­derlying issues facing the industry when it comes to sustainability, starting with one of the biggest pow­er consumption challenges of our generation: the explosion in data center growth. But first, a quick recap of last year’s issues.

GREENING OF STREAMING GROWTH

Let’s start with the growth of the Greening of Stream­ing (GoS) sustainability forum. Benjamin Schwarz, who has recently taken over the day-to-day operations of GoS, had a clear message for the industry in his first meeting as its president. “Advocacy has been part of GoS’s DNA from the start,” he said, noting that GoS founder Dom Robinson has used the almost 5 years since the seminal “The Greening of Streaming” ar­ticle that he and I co-wrote to play the role of outspoken advocate.

“With our ‘No Greenwashing’ rule, we’ve been seen as a bit activist, despite avoiding public disputes,” add­ed Schwarz, stating that activism isn’t off the table. “As I take the helm, our focus remains on accelerat­ing core projects like REM or EYANG and promoting their results. Any decision on the organization taking a more activist stance will be considered later on."

While the membership of GoS covers both North America and Europe, its primary advocacy has been at a U.K. Parliament and European Union level. Given the headwinds that may be faced with clean energy and sustainability initiatives in the U.S. with the change in administration, I anticipate the activism will focus on the European side of the pond rather than North America for at least the next 3 years.

MEASURING

Another recap is around measurements. In The State of Streaming Sustainability 2024, I mentioned the lack of consensus around the erroneous findings—and later corrections—of The Shift Project’s projections of energy usage by stream­ing companies. While those numbers are still float­ed in the industry from time to time, it appears that we’re getting closer to more accurate measurements, thanks in part to efforts by GoS and its member com­panies, as well as the membership of the Streaming Video Technology Alliance.

Having said that, it’s worth noting that streaming viewership continues climbing in 2025, and live-event streaming viewership is climbing even faster. The fact that live streaming is the fastest growth in the mar­ket, as shown most recently by the “Live-Streaming Technology Trends 2024” survey brief, means that some of the gains made by caching on-demand con­tent during off-peak hours as a way to increase en­ergy efficiency have been blunted.

SUSTAINABILITY IN DATA CENTERS

This leads into the core challenge we face: power consumption in data centers. In the northern Virgin­ia area, about 6 hours from my home, there’s a grow­ing sense of fatigue with data center development. The region, dubbed MAE East years ago, has expe­rienced a glut of data centers—often at the expense of open or public green spaces and agricultural en­deavors. This growth has been powered by a number of applicable computing uses, including well-known cryptocurrency and AI requirements as well as edge compute and live-streaming growth.

In the opposite direction and approximately the same distance from where I live, one of the larger ma­chine-learning data center endeavors is taking place in Memphis, Tenn. Again, the wide-open spaces of­fer plentiful flat land on which to build data centers, almost all of which are single-story buildings with massive footprints. The Memphis machine-learning projects, run by an (in)famous immigrant who is mak­ing his mark in Washington, D.C., also appear to sit on prior farmland, taking advantage of the delta-like areas that define the Mississippi River in the lower half of the U.S. as well as the water supply.

Stream Data Centers’ Chris Bair and Yvonne Deir write that data centers “typically use 3-5 million gal­lons of water per day (as much as a population of 30,000-50,000 people).” Meanwhile, in a case study, anthropologist Steven Gonzalez Monserrate says, “Like a pasture, server farms are irrigated. In many data centers today, chilled water is piped through the latticework of server racks to more efficiently cool the facility, liquid being a superior convective agent than air.”

There are many reasons for using liquid-cooling so­lutions, including the fact that areas that have higher-than-average humidity or temperatures (or a com­bination of both) require cooling tens or hundreds of thousands of servers replete with multiple CPUs, GPUs, NPUs, and a variety of other three-letter acro­nyms that make up modern raw compute capability.

So, what are the sustainability challenges of liquid-cooled infrastructures, whether they are the smaller compute-cluster solutions that we’re working on at Help Me Stream Research Foundation or the larger whole-data-center challenges? One of the biggest issues is a consistent liquid supply, the most preva­lent of which is a freshwater source. Yet these sour­ces tend to have limited capacity, tying conservation hand-in-hand with consumption.

“The most obvious way to reduce water consump­tion, for example, is to leverage closed-loop chillers,” according to Bair and Deir. “Those are more energy-intensive than evaporative cooling, but evaporative cooling requires more water.”

Another, perhaps less obvious, way is to house the data center in an environmental bubble, where the air temperature, humidity, and water consumption are tightly interleaved. I’ll say more about that later in this article.

ECONOMIC BENEFITS OF DATA CENTER SUSTAINABILITY

The economic development pitch for massive tra­ditional data centers is jobs that pay much higher than the average local base wage. I’m all for job cre­ation and even did a stint in economic development at a time when our local region was vying for a World Trade Center location, with the goal of introducing upper-end infrastructure and manufacturing jobs back into rural Appalachia.

The region around us, including the two states previously mentioned as well as North Carolina, has benefited greatly from the data center boom. In fact, Apple recently announced an expansion of the data center it had placed in a former soybean field in cen­tral North Carolina. Some U.S. states even offer tax incentives to data center companies if they employ a certain number of people and spend a certain amount on data center construction. For instance, local lead­ers have incentivized Google to expand its North Carolina data center, a $600 million undertaking.

Yet the “common knowledge” of high-paying data center jobs is also coming under scrutiny for not do­ing enough for the local economy. Google’s data cen­ter investment would create only 30 local jobs, and a planned $1 billion Microsoft data center project in In­diana would create only 200 over the course of 8 years.

A February 2025 article in The Wall Street Journal, The AI Data Center Boom Is a Job-Creation Bust, focuses on data centers in the U.S. Author Tom Dotan offers startling statistics on how few jobs data centers provide once they’re oper­ational relative to other types of businesses an area might attract. “In Abilene, Texas, some 1,500 people are building the first data center for the Stargate arti­ficial-intelligence venture led by OpenAI,” he writes, while stating that on completion, only 100 full-time employees will work there. “That total is a fraction of the number of people who might work on the same one million square feet if it were an office park, fac­tory or warehouse.”

How much different are the job numbers between a data center and a regular manufacturing facility? Almost 500% different. According to The Wall Street Journal article, a newly built food-packaging plant nearby is projected to employ 500 people.

Therein lies the disconnect between jobs and over­all investment when it comes to data centers. That disconnect has economic developers worried, since most tax incentives don’t kick in until a company has proactively employed at least 25 people. According to the common wisdom among economic develop­ers, without tax incentives, the data centers will go elsewhere. But without more high-quality jobs, the states, counties, cities, and other municipalities can’t afford to incentivize a data center, especially since their electricity-hungry business model often also pushes up overall electrical costs.

The even longer-term issue, though, is what hap­pens to these buildings when a data center is decom­missioned, either because of a technology advance­ment or—in the case of edge computing—a shift in geographic compute needs to another region or coun­try. To date, I’ve not seen any plans for repurposing data centers, as they’re often built as customized, be­spoke shells for the computing need du jour.

A MODEST DATA CENTER SUSTAINABILITY PROPOSAL

Earlier in this article, I mentioned the idea of an environmental bubble in which data centers may be able to thrive. While much has been written about the use of solar or wind power as clean energy sources for data centers, the fact remains that locations with copious sunlight—or enough sunlight to sustain so­lar farms to power even modest data centers—are often locations that have higher-than-average tem­peratures due to lack of cloud cover. Locations that use wind energy are a bit more conducive to lower-than-average surface temperatures, given the in­creased wind patterns, but the buildings themselves then have to be more robust to withstand wind shears.

The proposal, then, is this: Think smaller when it comes to data centers. And the corollary is to build data centers into existing environments. Don’t just take my word for it, though. On every single recent survey our team at Help Me Stream Research Foun­dation has done for Streaming Media, responses to questions asked about plans for mammoth data centers versus a smaller, regional variant have tilted toward the latter.

The focus on smaller data centers makes sense from a technical standpoint too, if one understands the context of telephony infrastructure: larger con­centration points (central offices, or COs) sending out trunk lines to smaller subscriber line interface circuit (SLIC) buildings, which then further branch signals down to neighborhood pedestals. The idea in telephony is to offload as much of the switching de­cisioning to the local area, leaving the CO to handle high-performance switching.

So far, I’ve focused only on building the packet-based equivalent of the CO, but from a sustainability standpoint, the real growth is in the SLIC and pedes­tal portion of the infrastructure. There’s no reason to push everything on the East Coast to MAE East, es­pecially if we’re talking about a live event that is 90% consumed within a 50-mile radius of the actual event.

Do these environmental bubbles exist? Yes, in mul­tiple places across the continental U.S., such as Ari­zona, Oregon, Virginia, and Wyoming. This is a much bigger industry discussion, one that I don’t have space for in this article, but I’ll cover it in a few of my Streams of Thought columns in 2025.

SUSTAINABILITY IN ON-DEMAND LONG-TAIL CACHING

Beyond the issue of data center sustainability, an­other aspect of sustainability that’s caught my atten­tion over the past year has been preservation and the immediate but efficient playback of long-tail content.

In 2023, I wrote a column called Evidence That Survives Time that’s garnered some attention, initially for chain-of-custody issues surrounding footage from the Russia-Ukraine war, but more recently closer to home with regard to foot­age from the last two U.S. presidential transitions. In that column and another one on a similar topic, I asked the industry the question of whether our tech­nology advances run the risk of slowing down justice.

It’s not just the sheer amount of content that has to be preserved, but also the ability to retrieve and play back this content a few years after an event. In researching those columns, as well as subsequent conversations, I found there’s an almost universal assumption that advances in technology always pro­vide backward-compatible paths for prior content.

Reality doesn’t support that view, though. In fact, most advances break compatibility with previous con­tent, almost as if the engineering assumption is that nothing prior to the current release of a media play­er or software application has to be preserved. If you need proof, try listening to a first-wave podcast (circa 1999) or even a second-wave podcast (circa 2007). In those two waves, there were a number of audio-centric podcasts (and a few webcasts, as we called them then) that used protocols to synchronize still images (slide­shows or even album art) alongside the audio track. These protocols often relied on proprietary content creation and distribution tools, but more than 85% of these tools are no longer in existence. Even for those that rose to become a de facto standard, often created by what we’d call today the MAANG companies (Meta, Amazon, Apple, Netflix, and Google), the tools no lon­ger support playback of original content. It’s one rea­son why we’re preserving the software stacks on par­ticular machines that are donated to Help Me Stream Research Foundation: A 2009-era iMac can play con­tent that a current-day iMac cannot.

CONCLUSION

Astute readers will have noticed a narrative thread running throughout this article: Sustainability also encompasses longevity. Whether it’s advocacy that in­dustry bodies like GoS and the Streaming Video Tech­nology Association generate in the form of best practices, significant challenges like the optimal use and reuse of data center buildings, or even the preservation of historically significant content—and I include culturally significant content such as memes and user-generated reels in this category—the end­game of sustainability should be planned in centuries instead of decades. Let’s proactively use the remain­der of 2025 to incorporate sustainability into the very core of the streaming industry.

The State of Streaming Sustainability 2024

How do others see us, and how do we measure our progress? New data on streaming power consumption will al­low the industry to target further reductions in power consumption while working toward longer-term solu­tions that reuse old technologies alongside current best practices as a way to extend the life of streaming tech for years to come.

18 Mar 2024

Evidence That Survives Time

From war crimes to war crime tribunals and war commemorations, how do we guarantee that content is available to play?

09 Aug 2023

The Greening of Streaming

The time has long since passed to turn our attention to making the streaming delivery ecosystem more environmentally friendly. Here are some of the key challenges, as well as some suggested first steps towards solutions.

26 May 2020