Revealing a hidden truth: Which consumes more energy, Bitcoin, streaming, AI, or social media?

In 2025, Bitcoin mining utilized approximately 171 TWh, which accounted for 16% of the overall energy consumption of data centers.

Globally, conventional data centers consumed between 448 and 1,050 TWh in 2025, with estimates differing among analysts. Gartner reported 448 TWh, while both Socomec and the IEA suggested a range of 600 to 1,050 TWh.

Projections from Gartner indicate this figure could rise to 980 TWh by 2030; however, the IEA also suggests we may surpass the 1,000 TWh mark this year (if we haven’t done so already).

Facilities focused on AI are estimated to have consumed between 82 and 536 TWh in 2025, representing 11-40% of total data center energy consumption. The broad range is influenced by the pace of AI deployment and the challenges in accurately monitoring usage data.

Consequently, traditional data centers, which encompass cloud computing, enterprise applications, streaming services, and social media, likely accounted for over 388 TWh in 2025.

2025 metric	Range (TWh)	Average used (TWh)	Notes
All data centers (excluding BTC)	448–1,050	800	Conservative working average for analysis
AI-focused data centers (derived from total)	88–536	350	Midpoint
Traditional / non-AI data centers (derived)	388–712	450	Total minus AI (800 − 350)
Bitcoin mining (electricity use)	138–204	171	Range spans Cambridge estimate (~138) to Digiconomist annualized estimate (~204)

According to Gartner,

“In 2025, AI-optimized servers are anticipated to constitute 21% of total center power consumption and 44% by 2030. By 2030, they will account for 64% of the additional power demand for data centers.”

Meanwhile, Socomec states,

“Data centers are projected to consume around 536 TWh of electricity in 2025, which represents about 2% of global electricity usage. This figure could potentially double to 1,065 TWh by 2030 as the requirements for AI computing power continue to rise.”

For this analysis, we will use an average of 1,000 TWh for all data centers in 2026, considering the extensive rollout of new infrastructure. Nevertheless, this may understate AI energy consumption by an equivalent amount to Bitcoin’s annual usage.

However, since there is no definitive consensus on the precise energy usage and distribution, I consider this to be the fairest allocation.

2026 projection metric	Share of total	Implied electricity (TWh)	Notes
All data centers (ex BTCl)	100%	1,000	Projected global data center electricity consumption
AI data centers	40%	400	AI share estimated at 40% of total
Traditional workloads	60%	600	Remaining share of total
Bitcoin mining (context)	–	150	Comparison benchmark accounting for difficulty drop

These projections place Bitcoin’s energy usage significantly lower than that of AI, video streaming, and social media.

I wonder how many members of the ‘Buttcoin’ community will take offense to this information while watching videos about how Bitcoin is a scam on YouTube or discussing it on Reddit?

Related Reading

Bitcoin is redrawing where cities and data centers rise as it competes for wasted energy, not cheap labor

From flared gas to curtailed hydro, off-grid power is redrawing the hashrate map and miner margins.

Nov 28, 2025 · Gino Matos

Energy mix for Bitcoin and traditional data centers

The energy profile for digital infrastructure indicates that Bitcoin operates on 52.4% sustainable energy (including renewables and nuclear), in contrast to the broader data center industry’s average of 42%, as per the Cambridge Digital Mining Industry Report 2025.

It is projected that AI data centers will consume 40% of total data center electricity in 2026, an increase from 14% in 2024. Traditional workloads will account for the remaining 45% of the global data center electricity footprint of 1,000 TWh, with Bitcoin covering the rest.

Bitcoin miners may face constraints in 2026 as AI firms compete for firm power supply, driving prices higher.

Related Reading

BlackRock warns crypto's love affair with AI is over as an energy war with Bitcoin miners begins

For years, Bitcoin mining has battled an optics war over energy consumption. Now AI is entering the same arena, presenting a very different argument: jobs, national competitiveness, and constant demand.

Jan 10, 2026 · Andjela Radmilac

At the conclusion of 2025, the network’s difficulty reached 148.2 trillion, with a slight decrease at the beginning of 2026 due to a decline in hashrate linked to Bitcoin’s falling price.

If efficiency enhancements offset hashrate growth, competition for inexpensive electricity could lower Bitcoin consumption to a minimum of 142 TWh by late 2026. In constrained scenarios, where AI infrastructure outbids mining operations, consumption could drop to between 100 and 140 TWh by 2030.

Bitcoin’s renewable energy composition currently stands at 43%, with hydropower comprising 23.12%, wind 13.98%, and solar 4.98% of the total energy profile, according to Cambridge Judge Business School.

Nuclear power contributes 9.8–10%, bringing the total sustainable energy percentage to 52.4%.

Natural gas has supplanted coal as the primary fossil fuel source, accounting for 38.2% in comparison to coal’s 8.9%, which is a decrease from 36.6% in 2022.

This shift in fossil fuel composition signifies a move towards lower-emission sources. The overall sustainable energy percentage surpasses both the global grid average of 40% and the data center industry standard of 42%.

However, Bitcoin’s per-user environmental footprint is approximately 2,768 kg CO2e annually per user, based on 30 million global users. Notably, an increase in users does not lead to a rise in energy consumption as seen with social media.

While this figure is 57 times greater than TikTok’s 48.5 kg per user and 46 times higher than the average social media user’s 60 kg footprint, it scales differently.

Scenario	BTC users	Total footprint (Mt CO₂e/yr)	Per-user footprint (kg CO₂e/user/yr)	Comparison notes
Baseline	30,000,000	83.04	2,768.00	≈57× TikTok (48.5 kg); ≈46× a 60 kg “avg social” benchmark
Social media equivalent energy per user	1,384,000,000	83.04	60.00	This is the BTC user count required if total footprint remains constant
TikTok-scale user count	1,500,000,000	83.04	55.36	At TikTok-scale, BTC per-user would be ~55 kg

Data center growth in 2026

Investment in AI infrastructure reached $400–450 billion in capex globally in 2026, with over half directed towards processors such as GPUs, TPUs, and custom ASICs, according to Deloitte Technology Predictions 2026.

The Stargate Initiative launched by OpenAI signifies a total investment of $500 billion, surpassing the inflation-adjusted $280 billion of the Apollo space program.

Google allocated $75 billion to AI infrastructure in 2025, which included the $4.75 billion acquisition of Intersect Power for data centers with integrated clean energy development.

Inference workloads now account for 66% of AI computing power in 2026, up from 33% in 2023, with training making up the remaining 33%.

This shift reflects the deployment phase of AI models, where ongoing query processing dominates energy consumption rather than one-off training sessions.

ChatGPT managed up to 200 million requests daily at 0.3 Wh per query for GPT-4o, totaling around 60 MWh each day. Earlier model iterations consumed as much as 2.9 Wh per query before optimization.

GPT-5 forecasts indicate 18.35 Wh per 1,000-token response, representing an 8.6-fold rise over GPT-4o’s per-query consumption.

If it processes 2.5 billion requests daily, GPT-5 could consume as much as 45 GWh per day, equivalent to the electricity needs of 1.5 million U.S. households, according to analysis from Windows Central and PatentPC.

Conventional data centers, which include social media platforms, streaming services, cloud computing, enterprise applications, SaaS, e-commerce, and financial services, are predicted to consume 400 TWh in 2026.

Current data does not differentiate social media and streaming consumption from broader traditional data center categories. These platforms are estimated to represent 15–30% of traditional workloads.

Category	Sustainable Energy Mix	Growth Rate
AI Data Centers	42%	~40% annually
Traditional Data Centers	42%	~9% annually
Bitcoin Mining	52.4%	Constrained by competition
Total Data Centers	42% average	2.5x from 2024

Meta reported a power usage effectiveness (PUE) of 1.09 for its data centers in 2025, marking industry-leading efficiency compared to the enterprise average of 1.5–1.6.

The company has avoided 16.4 million metric tons of CO2e since 2021 through enhancements in efficiency and renewable energy procurement.

TikTok’s parent company ByteDance reported approximately 50 million tons of CO2e in total company emissions annually. Per-user emissions were calculated at 48.49 kg CO2e based on third-party analysis of usage patterns.

Streaming energy usage

Netflix consumed 451,000 MWh annually as of 2019 data, sufficient to power 37,000 homes.

Energy consumption from streaming indicates that viewing devices account for 72% of emissions, data transmission 23%, and data centers 5%. Per-hour streaming energy was measured at 0.077 kWh in 2019, although efficiency improvements of approximately 20% annually since 2010 suggest that current consumption is lower.

The International Energy Agency stated:

“Contrary to a plethora of recent misleading media reports, the climate impact of streaming video remains relatively modest, especially compared to other activities and sectors.”

The Shift Project’s 2019 assertion that one hour of Netflix streaming consumed 6.1 kWh was corrected in 2020 to about 0.8 kWh.

This represented a seven-to-eightfold exaggeration that persisted despite the correction.

Current estimates from the Carbon Trust suggest streaming emissions are approximately 55g CO2e per hour on European grids. The IEA’s 2020 analysis calculated 36g CO2e per hour, with variance reflecting different grid carbon intensities and efficiency improvements over time.

Bitcoin benefits the energy grid unlike streaming or social media

Bitcoin mining facilities can reduce demand almost instantly, allowing participation in demand response programs and utilizing otherwise curtailed renewable energy.

Flexible loads like Bitcoin mining could contribute 76 GW to grid capacity, roughly 10% of peak demand, based on modeling from Duke University, as reported by CPower Energy.

In Texas, ERCOT incorporated Bitcoin miners as flexible loads following the 2021 blackouts, preventing an estimated $18 billion in gas peaker plant construction.

AI and traditional data centers require a steady, reliable power supply for service delivery, limiting their capacity to provide grid balancing services.

Data center occupancy rates reached 85% in 2023 and are forecasted to surpass 95% by late 2026, offering little flexibility for demand response.

Water consumption estimates for U.S. AI servers range from 731 to 1,125 million cubic meters annually by 2030, according to MIT News.

Bitcoin’s air-cooled ASIC systems consume significantly less water compared to liquid-cooled data center infrastructures.

Advancements in ASIC technology demonstrate that top-tier models from 2026 achieve 9.5–12 joules per terahash (J/TH), in contrast to legacy models from 2020–2023, which operated at 25–30+ J/TH.

The Antminer U3S23H delivers 1,160 TH/s at 9.5 J/TH, while the S21 XP Hydro achieves 473 TH/s at 12 J/TH.

These efficiency gains of 50–70% are facilitated by transitions from 7nm to 5nm and 3nm chip architectures. Overall network consumption remains stable or increases due to Jevons Paradox, where improvements in efficiency enable more mining activity at reduced costs.

The pattern repeats across all three sectors.

AI inference efficiency has improved tenfold from early GPT-4 estimates to GPT-4o, yet total AI consumption is anticipated to rise sevenfold from 60 TWh in 2024 to 420 TWh in 2026.

Streaming data center energy intensity has decreased by 20% annually since 2010, yet total streaming hours and absolute consumption continue to grow.

Efficiency improvements lower costs per unit, enabling increased consumption that frequently exceeds savings from efficiency enhancements.

Goldman Sachs estimates that 60% of the increase in data center electricity demand through 2030 will be satisfied by fossil fuels, adding approximately 220 million tons of CO2 to global emissions.

Natural gas acts as a “bridge fuel” during the period of 2026–2028 while renewable and nuclear projects are still under construction.

Tech giants like Amazon, Microsoft, Meta, and Google have contracted over 50 GW of renewable capacity, equivalent to Sweden’s total generation capacity. However, delivery timelines lag by two to five years due to development schedules.

Microsoft’s $10 billion renewable energy deal with Brookfield will provide 10.5 GW of capacity starting in 2026, comparable to the output of 10 nuclear power plants.

Google’s partnership with NextEra aims to restart Iowa’s Duane Arnold nuclear facility in 2027. Meta has collaborated with Oklo to develop small modular nuclear reactors to power data centers in Pike County.

Meta’s Louisiana data center represents a $10 billion investment with more than 1,500 MW of new renewable energy planned for grid connection.

Critical power capacity for data centers worldwide was 55 GW in 2023 and is projected to reach 82–96 GW by 2026, indicating a near-doubling of infrastructure within three years.

Regional distribution shows that the U.S. and China are responsible for about 80% of global data center electricity growth through 2030. The United States is expected to add 240 TWh, marking a 130% increase from 2024, while China will add 175 TWh, a 170% rise from 2024.

Currently, Ireland allocates 21% of its national electricity to data centers, with projections indicating this could rise to 32% by 2026 if current growth trends persist.

Grid connection timelines of two to five years in many areas, coupled with supply chain bottlenecks for transformers and substations, impede expansion rates.

Local utility capacity is nearing limits in several markets, while the availability of cooling water poses challenges in drought-prone regions such as Arizona, Nevada, and Texas.

Energy use across crypto

Ethereum’s shift to proof-of-stake on September 15, 2022, led to a 99.988% reduction in energy consumption, dropping from 23 TWh annually to