AI Race Facts.Facts. Not fear.
A clear, sourced look at the data behind America's AI infrastructure — data centers, power consumption, water usage, and how the United States compares with China in the global AI race.
Independent and ad-free. Donations directly fund hosting, research time, and citations.
Top Water Users in the United States (2020)
These three categories account for approximately 90% of all freshwater withdrawals in the United States. Data centers account for about 0.015% of total U.S. water use.
Thermoelectric power refers to water withdrawn for cooling at power plants — primarily coal, natural gas, and nuclear facilities. Most of this water is used in "once-through" cooling systems: it is drawn from a river or lake, circulated through the plant to absorb waste heat, and then returned to the source. A smaller share is consumed through evaporation in cooling towers. It is a withdrawal-heavy category, not primarily consumption.
Source: U.S. Geological Survey (USGS), Model estimates for water year 2020, released 2025.
Water Use citations
- [1] U.S. Geological Survey (2025). Estimated Use of Water in the United States in 2020. Thermoelectric power, irrigation, and public supply represent the largest freshwater withdrawals. water.usgs.gov
- [2] Shehabi, A. et al., Lawrence Berkeley National Laboratory (2024). 2024 United States Data Center Energy Usage Report. US data centers directly consumed ~66 billion liters (~17.4 billion gallons) of water for cooling in 2023. eta-publications.lbl.gov (PDF)
Why This Site Exists
I started building this site because the intense protests against new AI data centers didn't make sense. Other major industries — especially agriculture — use far more water and power, often with significant waste and exports. There appeared to be coordinated opposition, so I began digging into the numbers.
This is what I found:
AI data centers withdrew roughly 17.4 billion gallons of water in 2023 — about 0.015% of total U.S. freshwater withdrawals (~117 trillion gallons/year, USGS 2020). For electricity, data centers used ~176 TWh, around 4.4% of U.S. consumption (LBNL 2024). Compare that to other sectors:
- Thermoelectric power (cooling fossil/nuclear plants): ~41% of U.S. freshwater withdrawals — roughly 48 trillion gal/year (USGS).
- Agriculture / irrigation: ~42% — roughly 49 trillion gal/year, with alfalfa alone using ~15% of Western water (USGS; Richter et al., Nature Sustainability 2020).
- Public supply (residential & municipal): ~12% — roughly 14 trillion gal/year, ~82 gal/person/day at home (USGS; EPA WaterSense).
- Livestock & aquaculture: ~3% — about 3.5 trillion gal/year (USGS).
- Fashion & textiles (global): ~93 billion m³/year (~24.6 trillion gal) — a single cotton t-shirt takes ~2,700 gallons, and a pair of jeans ~1,800 gallons (UN Environment Programme; Ellen MacArthur Foundation).
- Golf course irrigation (U.S.): ~1.5 billion gal/day — over 540 billion gal/year, more than 30× annual AI data center use (GCSAA Environmental Profile).
- Bottled water industry (U.S.): ~15.9 billion gal/year sold in 2023 (Beverage Marketing Corp. / IBWA) — comparable to annual AI data center withdrawals.
Despite using a fraction of a percent of national water, AI data centers face outsized resistance — while the U.S. risks falling behind in the global AI race, particularly against China's aggressive buildout of its own infrastructure.
This site presents the facts, with clear sources from government reports and academic research. No hype. No agenda. Just data to help people understand what's really at stake.
Fashion's water footprint
The global fashion industry is one of the largest industrial water users on the planet — dwarfing AI data center water use by orders of magnitude.
Global water consumption by the fashion industry — roughly 24.6 trillion gallons annually.
Water to produce a single cotton T-shirt — enough drinking water for one person for ~2.5 years.
Fashion uses roughly 1,400× more water annually than all US AI data centers combined.
Global fashion industry water use, per year
Fashion vs AI data centers — annual water
Global freshwater withdrawals by sector
US annual water use — AI data centers vs agriculture & power
| Category | Annual Water Use | Notes |
|---|---|---|
| Global Fashion Industry | ~93 billion m³ (~24.6T gal) | UNEP, 2019 |
| One Cotton T-shirt | ~2,700 liters | WWF |
| One Pair of Jeans | ~7,500–10,000 liters | UNEP |
| Fashion Share of Industrial Wastewater | ~20% | UN Environment |
| US AI Data Centers (Direct Cooling) | ~17.4 billion gallons | LBNL, 2024 |
The global fashion industry uses roughly 1,400× more water per year than all US AI data centers combined.
Citations
- [1] UN Environment Programme (2019). UN Alliance for Sustainable Fashion addresses damage of "fast fashion". Cites the fashion industry as the second-largest consumer of water worldwide, producing ~93 billion cubic metres per year. unep.org
- [2] Ellen MacArthur Foundation (2017). A New Textiles Economy: Redesigning fashion's future. Reports ~93 billion m³ of water used by the textiles sector annually and ~20% of industrial water pollution from textile dyeing/treatment. ellenmacarthurfoundation.org
- [3] World Wildlife Fund. The Impact of a Cotton T-Shirt. It takes ~2,700 liters of water to produce the cotton needed for one t-shirt. worldwildlife.org
- [4] UN Environment Programme (2018). Putting the brakes on fast fashion. Cites ~7,500 liters of water to make a single pair of jeans. unep.org
- [5] Shehabi, A. et al., Lawrence Berkeley National Laboratory (2024). 2024 United States Data Center Energy Usage Report. US data centers directly consumed ~66 billion liters (~17.4 billion gallons) of water for cooling in 2023. eta-publications.lbl.gov (PDF)
- [6] FAO (2025). AQUASTAT Water Data Snapshot 2025. Agriculture accounts for ~72% of global freshwater withdrawals, industry ~15%, and municipal/domestic ~13% (2022 data). fao.org
The ~1,400× comparison divides 24.6 trillion gallons (UNEP fashion estimate, converted from 93 billion m³) by 17.4 billion gallons (LBNL 2024 US data center estimate).
Streaming and social media's footprint
Video streaming and social platforms are some of the largest drivers of internet traffic — and a steadily growing share of global electricity and data center demand.
Data transmission networks consumed 260–360 TWh in 2022 — roughly 1–1.5% of global electricity. Video streaming dominates.
Video streaming services (Netflix, YouTube, TikTok, etc.) accounted for ~65% of global downstream internet traffic in 2022.
Estimated electricity for one hour of Netflix streaming — roughly equivalent to running a low-energy LED bulb for 4 hours.
Estimated electricity per hour of use
| Activity / Platform | Energy or Footprint | Notes |
|---|---|---|
| Data transmission networks (global) | 260–360 TWh / year (2022) | ~1–1.5% of global electricity (IEA) |
| Data centers (global, excl. crypto) | 240–340 TWh / year (2022) | ~1–1.3% of global electricity (IEA) |
| Video streaming share of internet traffic | ~65% (2022) | Sandvine |
| Netflix HD streaming | ~0.077 kWh / hr | IEA revised estimate (2020) |
| YouTube HD streaming | ~0.15 kWh / hr | Carbon Brief / Shift Project |
| TikTok (mobile) | ~0.06 kWh / hr | Greenspector mobile study, 2021 |
| Instagram (mobile feed) | ~0.055 kWh / hr | Greenspector, 2021 |
| 1-min HD video upload | ~50–100 Wh | Carbon Brief |
Streaming video and social platforms together drive the majority of internet traffic and a meaningful share of global data center and network electricity — and that share is growing faster than AI inference.
Citations
- [1] International Energy Agency (2023). Data Centres and Data Transmission Networks. Data transmission networks consumed an estimated 260–360 TWh in 2022, about 1–1.5% of global electricity use. iea.org
- [2] Sandvine (2023). The Global Internet Phenomena Report. Video accounts for ~65% of all downstream internet traffic globally; Netflix, YouTube, TikTok, and Instagram are among the top applications. sandvine.com
- [3] Kamiya, G. — IEA / Carbon Brief (2020). Factcheck: What is the carbon footprint of streaming video on Netflix? Revised estimate: ~0.077 kWh per hour of Netflix streaming (down from earlier overestimates). carbonbrief.org
- [4] The Shift Project (2019). Climate Crisis: The Unsustainable Use of Online Video. Per-hour estimates for YouTube and other streaming platforms (note: subsequent IEA analysis suggests these are upper-bound estimates). theshiftproject.org
- [5] Greenspector (2021). Environmental impact of TikTok, Instagram, and other mobile apps. Per-hour mobile energy measurements for TikTok (~0.06 kWh) and Instagram (~0.055 kWh). greenspector.com
- [6] FAO AQUASTAT (2025). Water Data Snapshot 2025. Global freshwater withdrawals: ~72% agriculture, ~15% industry, ~13% municipal/domestic (2022 data). fao.org
Power usage: the real numbers
Data center electricity demand is growing — but remains a single-digit share of U.S. consumption. Residential and commercial buildings together still dwarf data centers.
U.S. data center electricity consumption — about 4.4% of total U.S. electricity.
Still less than residential + commercial buildings combined.
Used by residential buildings alone — far above all data centers.
U.S. data center electricity consumption (TWh)
AI infrastructure is essential industrial capacity — comparable to power plants and factories. Growth is real, but the share remains modest.
AI energy efficiency: rapid progress
While total demand is rising, the energy required per AI task continues to drop dramatically thanks to better models, hardware, and software optimization.
Energy for a typical text query (e.g. Gemini) — roughly 9 seconds of TV watching.
Performance-per-watt improvement across recent generations of AI accelerators (e.g. Google TPU).
Drop in median prompt energy reported by Google over a single year of model and serving optimization.
Energy per AI query vs. everyday tasks
Hardware efficiency gains: AI accelerator generations
Energy per AI task is falling faster than overall demand is rising in many workloads — efficiency is doing real work alongside new capacity.
What Tech Companies Are Doing
Major tech and AI companies are actively developing and deploying new technologies to reduce the water and energy demands of their data centers.
Microsoft
In 2024, Microsoft introduced a new data center design using closed-loop, direct-to-chip cooling that consumes zero water for evaporative cooling. New facilities in Arizona and Wisconsin will pilot this approach starting in 2026.[1][2]
Google uses a science-based approach to choose between air cooling and water cooling depending on local conditions. The company aims to replenish more freshwater than it consumes by 2030 and has supported over 100 water stewardship projects across 68 watersheds.[1][2]
Meta
Meta has supported more than 40 water restoration projects since 2017. In 2024, these projects returned over 1.59 billion gallons of water to stressed watersheds. Some newer data centers use extremely low amounts of water.[1][2]
Amazon (AWS)
AWS has improved its Water Usage Effectiveness by 40% since 2021. The company uses reclaimed wastewater at many sites and is working toward being water positive by 2030. They have also launched initiatives to use AI to help solve broader water challenges.[1][2]
Energy Efficiency Improvements
- Companies are rapidly adopting direct-to-chip and immersion cooling, which are far more energy-efficient than traditional air cooling.[1]
- Newer AI chips and optimized software are delivering significantly better performance per watt.[1][2]
- Hyperscalers are investing heavily in renewable energy and exploring nuclear power deals to meet growing demand sustainably.[1][2]
- AI itself is being used to optimize cooling systems and workload scheduling, often reducing energy use by 10–40% in some facilities.[1]
Click any [n] citation above to open the primary source in a new tab.
The industry is moving fast to develop more efficient cooling technologies and better water stewardship practices — even as AI infrastructure continues to expand rapidly.
US vs China: the global AI buildout
While the U.S. debates new data center capacity, China continues rapid expansion of its own AI infrastructure, grid, and chip supply.
Estimated AI-related data center capacity (TWh equivalent)
China added more new power generation capacity in 2023 than the U.S. has in the last decade.
China's hyperscale data center footprint is growing roughly twice as fast as the U.S.
Winning the AI race requires building the infrastructure — power, water, and compute — today.
Foreign influence & long-term consequences
Ongoing congressional investigations are examining whether China-linked funding has supported U.S. nonprofits opposing AI data centers. The strategic stakes extend well beyond local zoning fights.
House Ways and Means Committee Chairman Jason Smith has stated that Chinese money has been traced to U.S. nonprofits organizing protests against data centers, with the goal of slowing American AI development.[1]
What if China wins the AI race?
- China would likely control the dominant AI chips, models, and software standards for decades.[2]
- Future smartphones, computers, cars, medical equipment, and critical infrastructure worldwide could run on Chinese-controlled technology.[3]
- The U.S. would shift from setting global AI rules to operating within systems shaped by a strategic competitor.[4]
Projected cost & technology landscape if China leads
Many devices could become more affordable due to Chinese manufacturing scale — but at the cost of technological dependency.
| Device category | U.S. / Western price | China-equivalent price | Implications |
|---|---|---|---|
| iPhone 17 (base / Pro) [5] | $799–$1,099+ | ¥4,499–¥6,999 (~$622–$968) | Consumer electronics cheaper globally; reliance on Chinese supply chains and standards. |
| Premium Chinese smartphone (Xiaomi / Huawei flagship) [6] | $700–$1,000+ (imported) | $650–$850 equivalent | Price/performance edge expands; Chinese brands set global benchmarks. |
| Handheld ultrasound [7] | $2,000–$4,000 | $950–$2,500 | Medical devices 40–60% cheaper, improving access but increasing reliance. |
| Patient monitors (hospital-grade) [8] | $5,000–$15,000+ | $1,000–$5,000 equivalent | Production costs 20–60% lower; widespread hospital adoption. |
| Surgical robot (da Vinci-class) [9] | $1.8M–$2.5M + ~$100K–$190K/yr | $200K–$1M (Surgerii, Jingrong) | 60–90% lower upfront + reduced ongoing costs. |
| Automotive FSD / ADAS hardware [10] | Tesla HW4: ~$800–$1,500; 100–150 TOPS | Horizon J6 / XPeng: $300–$800; 560+ TOPS | Higher compute at lower cost; vehicles dependent on Chinese tech stack. |
See references [5]–[10] below for pricing and hardware sources.
The broader long-term picture
Everyday devices, medical equipment, and vehicles become more accessible.[11]
Loss of control over critical AI-driven systems in phones, cars, hospitals, and infrastructure.[12]
National security, data privacy, and innovation leadership shift toward China.[13]
Domestic opposition to infrastructure carries real strategic costs. America must address legitimate local concerns while protecting its ability to lead in AI for generations.
Strategic Risk citations
- [1]U.S. House Committee on Ways and Means — Chairman Jason Smith, public statements and committee hearings on foreign-linked funding of U.S. nonprofits opposing AI data centers (2024–2025).
- [2]Center for Strategic and International Studies (CSIS) — "Full Stack: The Evolution of the U.S.–China AI Rivalry" (2024); Stanford HAI AI Index Report 2024.
- [3]U.S.–China Economic and Security Review Commission — Annual Report to Congress (2023, 2024), sections on AI integration in consumer electronics, autos, and medical devices.
- [4]National Security Commission on Artificial Intelligence — Final Report (2021); Brookings Institution, "Global AI governance and the risk of bifurcation" (2024).
- [5]Apple Inc. — iPhone 17 official U.S. pricing (apple.com, 2026); Apple China store pricing (apple.com.cn, 2026).
- [6]Counterpoint Research — Global Smartphone Premium Market reports (2025); Canalys Smartphone Analysis (2025).
- [7]Grand View Research — Handheld Ultrasound Market Size Report (2024); Mindray, Chison, and SonoStar published price lists; FDA 510(k) device database.
- [8]GlobalData Medical Devices Intelligence — Patient Monitoring Market Report (2024); Mindray and Edan Instruments published product pricing.
- [9]Intuitive Surgical — 2024 Annual Report (10-K), da Vinci system pricing and service revenue; Surgerii Robotics and Jingrong Surgical public pricing disclosures (2025).
- [10]Horizon Robotics — Journey 6 product specifications (2025); Tesla AI Day disclosures on HW4 (2023); Munro & Associates / Nikkei Asia ADAS hardware teardowns (2024–2025).
- [11]McKinsey Global Institute — "The state of AI in 2024"; OECD, "AI, data, and competition" (2024).
- [12]U.S. Department of Defense — 2024 Annual Report on Military and Security Developments Involving the PRC; CISA, "Securing the ICT Supply Chain" (2024).
- [13]Office of the Director of National Intelligence — Annual Threat Assessment of the U.S. Intelligence Community (2024).
Methodology & references
All figures on this site are drawn from U.S. government agencies, national laboratories, and peer-reviewed analyses. Numbers are rounded for clarity.
- Lawrence Berkeley National Laboratory — 2024 U.S. Data Center Energy Usage Report
- U.S. Department of Energy — Data Centers and Energy Reports
- U.S. Geological Survey — Estimated Use of Water in the United States
- USDA NASS — Irrigation and Water Management Survey, 2023
- USDA Economic Research Service — Food Loss and Waste
- International Energy Agency — World Energy Outlook 2024
- U.S. Energy Information Administration — Annual Energy Outlook
- Synergy Research Group — Hyperscale Data Center Tracker