Why Tech Giants Are Buying the World's Energy And Who Is Actually Paying For It
Why Tech Giants Are
Buying the World's Energy
— Water Crisis, Power Shortages
& Who Is Actually Paying
$700 billion in infrastructure deals. AI data centers draining water equal to Japan's electricity consumption. Electricity bills up 36%. Grid failures already forming. Secret NDA contracts. US Congress demanding answers. This is the complete investigation — with every data point sourced.
Let me start with three numbers that, taken together, should alarm every person on this planet who uses electricity, drinks water, or earns a living in the digital economy.
$700 billion. The combined capital expenditure that Microsoft, Google, Amazon, and Meta have committed to technology infrastructure in 2026 — much of it flowing into energy deals, nuclear contracts, and power plants.
36%. The rise in US residential electricity prices since 2020. In some areas near data centers, electricity costs 267% more than five years ago.
3 bottles of water. The amount consumed to generate 100 words using GPT-4. Not metaphorically. Physically — evaporated in cooling towers that keep the servers from overheating.
These three numbers are connected. They are all products of the same decision made by the same companies: to build AI infrastructure at a scale and speed that the world's energy and water systems were never designed to support.
This is the full investigation. What is driving the energy race. What the data centers are really doing to water supplies. What global challenges are coming. And who is actually paying for all of it.
01 Why Tech Companies Need This Much Energy — The Physics of AI
Before the deals, the dark side, or the politics — you need to understand the physical reality of what AI actually demands. Not in marketing language. In watts and liters.
The Energy Math Nobody Shows You
- Training one large AI model consumes more electricity than 100 US homes use in an entire year
- One ChatGPT text query uses nearly 10 times more electricity than a Google Search
- Generating one AI image uses thousands of times more energy than a text search — equivalent to charging a smartphone
- ChatGPT alone used over 500,000 kilowatt-hours of electricity daily as of 2024
- AI agents — autonomous multi-step AI workflows — are far more energy intensive than simple chatbot queries
- Global data center electricity is set to approach 1,000 TWh by 2026 — if data centers were a country, they would be the world's 5th largest energy consumer, between Japan and Russia
- US data center demand is projected to nearly double from 80 to 150 gigawatts between 2025 and 2028 — equivalent to adding a Spain-sized energy consumer in three years
The problem is not just scale. It is type of power required. AI data centers need constant, uninterrupted, 24/7 baseload electricity. Solar produces electricity only when the sun shines. Wind only when wind blows. Neither guarantees the millisecond-level reliability that a real-time AI inference system demands.
This is why every major tech company — after years of sustainability reports celebrating solar and wind — has pivoted to nuclear energy.
Microsoft: 20-year $16B deal to restart Three Mile Island (837MW, 2028). Total clean power contracted: 34.7 GW — largest corporate commitment in history. Google: First-ever US corporate SMR fleet — 500MW from Kairos Power across 6-7 reactors (2030-2035). Acquired Intersect Power for $4.75B to own generation directly. Amazon: $500M in X-energy SMR developer, targeting 5GW by 2039. 17-year, 1.92GW deal with Talen Energy's Susquehanna nuclear plant. Meta: Up to 6.6GW nuclear portfolio spanning Constellation Energy, TeraPower, Oklo, and Vistra Energy. Oracle: Building permits secured for three SMRs to power a gigawatt-scale data center.
| Company | Partner / Deal | Scale | Timeline |
|---|---|---|---|
| Microsoft | Three Mile Island — Constellation Energy | 837 MW | 20yr, online 2028 |
| Microsoft | Total clean power contracted | 34.7 GW | Ongoing |
| Kairos Power SMR fleet | 500 MW (6-7 reactors) | First 2030, full 2035 | |
| Intersect Power acquisition | $4.75B | Early 2026 | |
| Amazon | X-energy SMR investment | $500M / 5GW target | By 2039 |
| Amazon | Susquehanna nuclear — Talen Energy | 1.92 GW | 17-year PPA |
| Meta | Multi-company nuclear portfolio | Up to 6.6 GW | 2027-2035 |
| Oracle | Three SMRs — gigawatt data center | ~1 GW | Permits secured |
02 The Hidden Crisis — AI Is Running Out of Water
Energy gets most of the attention. But there is a second resource crisis unfolding alongside it — one that is less visible, less discussed, and in many ways more immediately dangerous for the communities hosting these data centers.
AI is consuming water at a scale that most people have never heard about.
- 100 words from GPT-4 — consumes up to 3 bottles of water (for server cooling)
- Global AI water footprint 2025 — 312 to 764 billion liters (equivalent to world's entire annual bottled water consumption)
- AI carbon + water footprint 2025 — equivalent in scale to New York City's annual carbon emissions, per ScienceDirect research
- Texas data centers 2025 — using 49 billion gallons of water; projected to reach 399 billion gallons by 2030 (equivalent to draining Lake Mead by 16 feet annually)
- Two-thirds of data centers built since 2022 are located in areas already experiencing water stress, per Bloomberg investigation
- Evaporative cooling systems — used by majority of AI-specialized data centers — permanently remove water from local supplies through evaporation; it does not return as usable water
- Approximately two-thirds of total data center water use is linked to electricity generation; one-quarter to direct cooling
Where the Water Goes — and Why It Does Not Come Back
Server racks generate enormous heat. The cheapest way to cool them is evaporative cooling — essentially, industrial-scale swamp coolers that blow air over water, causing it to evaporate and carry heat away. It works. It is cheap. And it permanently removes water from the local water cycle.
Unlike water used for drinking or agriculture that eventually returns to the ground or the atmosphere locally, evaporated cooling water disperses into the broader atmosphere, often far from where it was extracted. For communities in water-stressed regions, this is not a theoretical concern.
Phoenix draws 40% of its water from the Colorado River system. Lake Powell and Lake Mead — the system's two major reservoirs — have fallen from roughly 90% capacity in 2000 to around 30% today. The same region is now one of the fastest-growing data center markets in the US. Hyperscale data centers requiring millions of gallons of cooling water are being built in a desert city already running out of water from a 25-year megadrought. Microsoft is rolling out closed-loop water recycling systems for its Arizona facilities starting 2026 — but the majority of existing and planned facilities still use evaporative cooling.
The Industry Response — Real Solutions vs. Greenwashing
✅ Genuine Progress
- Microsoft closed-loop water recycling — same water circulates continuously, no freshwater top-ups needed
- Applied Digital closed-loop direct-to-chip liquid cooling — near-zero water consumption
- AirJoule atmospheric water generation — extracts water from air for cooling
- Net Zero Innovation Hub (Google, Microsoft, Schneider Electric) — technology acceleration program for sustainable data centers
- Crusoe Energy closed-loop systems at OpenAI Stargate Abilene site
❌ Still The Reality
- Majority of AI-specialized data centers still use evaporation-based cooling 24/7 or on hot days
- More data centers expected to use water evaporation across the industry by 2028 (Lawrence Berkeley National Lab)
- Closed-loop systems are more energy-hungry — tradeoff between water and electricity use
- Water is "the last consideration" in data center siting decisions because it is cheap compared to real estate and power
- Water stress impact assessments not yet required in most US states for data center approvals
03 The Rise of Energy Companies — Who Is Getting Rich
Nuclear energy stocks surged 40% year-to-date through 2025 as tech giants signed multibillion-dollar power agreements. This was not a coincidence. Tech companies have become the primary financial anchors of an entire nuclear renaissance — financing reactor designs, underwriting construction, and providing the long-term contracts that make previously unfundable projects economically viable.
The New Energy Power Class
- Constellation Energy — operator of America's largest nuclear fleet; became one of Wall Street's hottest stocks after Microsoft's Three Mile Island deal
- Kairos Power — startup that went from obscurity to receiving Google's first-ever US corporate SMR fleet order
- X-energy — received $500M Amazon investment for gas-cooled SMR development
- TeraPower & Oklo — beneficiaries of Meta's diversified 6.6GW nuclear portfolio
- Talen Energy — 17-year PPA with Amazon worth over $20B in Pennsylvania investment
- Intersect Power — acquired by Google for $4.75B in early 2026 — first time a tech giant bought a power generation company outright
- Uranium producers — seeing strongest growth cycle in decades as nuclear demand surges
04 The Dark Side — What the Press Releases Do Not Mention
The Electricity Bill Nobody Authorized
- US residential electricity rose from 12.76 to 17.44 cents per kWh between 2020 and February 2026 — a 36.7% increase faster than inflation
- Near data center clusters — electricity costs up to 267% more than five years ago (Bloomberg)
- New Jersey — average electric bills surged over 20% in 2025 alone
- Baltimore — average bills jumped $17/month after power auction record; set to rise again by $4 in mid-2026
- Record rate hike requests — utilities filed $31 billion in rate hikes in 2025, more than double 2024
- Federal Reserve Bank of Dallas — estimates wholesale power prices could rise another 50% as data center demand doubles in 5 years
- 38% of Americans believe data centers are "mostly bad" for home energy expenses (Pew Research, 2026)
- Political flashpoint — Democratic candidates won Virginia and New Jersey governor races partly on anti-data-center platform; nearly half of Americans expect this to be a 2026 midterm issue
The Secret Contracts — Congress Steps In
- Bipartisan Senate inquiry led by Senators Markey and Capito demands transparency from utilities over NDA-protected contracts with Amazon, Google, Microsoft, Meta
- Communities affected by these deals have no legal right to see contract terms that directly shape their electricity prices
- Federal vs. state jurisdiction — FERC oversees interstate transmission but data center cost allocation falls under fragmented state-level regulation
- 11 states are considering legislation to temporarily ban new data centers pending impact assessments
- Ratepayer Protection Pledge (March 2026) — voluntary, no enforcement, no penalties, no independent oversight; described by Public Citizen as "few specifics or teeth"
The Environmental Gap — Clean Narrative vs. Gas Reality
- SMRs don't exist commercially yet — first Google/Kairos reactor: 2030. Full fleet: 2035. Amazon X-energy: 2039
- The fossil fuel bridge is being built right now — Georgia Power proposing billions in new gas capacity for data centers
- Behind-the-meter gas generation — data centers installing on-site gas plants and converting jet engines to generate electricity
- Rural fracking risk — North Carolina communities face potential fracking to fuel nearby data centers
- AI carbon footprint 2025 — between 32.6 and 79.7 million tons of CO2 emissions — equivalent to New York City's annual output
- The gap between the "carbon-free future" narrative and the gas-powered present is at least 10 years long
05 Global Challenges — What Is Coming For The Entire World
Most of the energy crisis coverage focuses on the United States. But the consequences of the tech-energy collision are global — and they fall most heavily on countries and communities that had no part in creating the AI boom.
Specific Global Challenges Ahead
- Grid capacity bottlenecks — regions like Oregon, Virginia, and Ireland may experience Power Stress Index values exceeding 0.25, indicating serious local grid vulnerability (arxiv.org research, 2026)
- Electricity demand growth mismatch — data center electricity growing at 12% compound annually since 2017, more than four times faster than total global electricity growth
- Interconnection backlogs — Dominion Energy Virginia received 40.2 GW of power connection requests in February 2025, up from 21.4 GW just seven months earlier
- Planning system overload — IEA April 2026 report: "Planning and regulatory systems are being stretched by the wave of project applications for data centres, amid a broader trend of rapid load growth"
- Stranded asset risk — if AI demand growth slows, $700B in committed infrastructure could become stranded; utilities and communities would still carry the cost
- Developing world disadvantage — countries in South Asia, Africa, and Southeast Asia face higher energy import costs while losing outsourced work to AI displacement simultaneously
- Nuclear waste — SMR waste management frameworks are not yet fully established; first-generation SMR waste will need solutions that don't yet exist at scale
- Cybersecurity — concentration of critical AI and energy infrastructure creates high-value targets; a successful attack on interconnected AI data centers and their dedicated power plants could constitute a national security event
Pakistan, India & the Developing World — The Double Hit
For countries like Pakistan, India, Nigeria, and the Philippines — where millions of people depend on outsourced digital work for income — the tech-energy convergence creates a uniquely damaging double compression:
- Energy cost spillover — global electricity price increases driven by US and European AI demand translate into higher energy import costs for countries that import fuel
- Equipment scarcity — transformers, generators, and grid components needed for grid modernization are now in short supply globally because tech companies are buying them first
- AI displacement of outsourced work — the same data centers driving up energy costs are training models that displace writing, coding, customer service, and data work that developing-country workers provide
- No seat at the table — the nuclear deals, the congressional inquiries, and the Ratepayer Protection Pledge are entirely US-centric; no mechanism exists for developing countries to influence these decisions despite bearing their consequences
- Brain drain acceleration — the global race for AI talent concentrates skilled workers in tech hubs, depleting human capital from countries that can least afford to lose it
06 The Human Impact — Consumers, Communities & The Grid
Ordinary Households
- Average US family pays $17-25 more per month near major data center clusters
- Those on fixed incomes — retirees, low-income households — have no offsetting benefit from the AI economy
- Energy and piped gas became the two largest drivers of US inflation in 2025 — rising 7% and 11% respectively
- Projected to keep increasing through 2026 and beyond per US Energy Information Administration
- Expected price: 19.01 cents per kilowatt-hour by September 2027
Small Businesses
- Cannot negotiate the long-term power purchase agreements that shield tech giants from price spikes
- Commercial electricity rates rising alongside residential without offsetting revenue from AI
- Energy becoming an unpredictable and uncontrollable cost category for the first time in decades
- Data centers in the same region compete for grid capacity that small businesses rely on for reliable service
Grid Reliability Risk for Everyone
- ERCOT (Texas) projects data center demand could more than double by 2030
- Dominion Energy (Virginia) warned it may not keep up with connection requests
- PJM — nation's largest grid operator covering 13 states — warning of reliability stress
- Harvard/Boston University research (2025) warns of potential "power grid crisis" if AI energy demand outpaces infrastructure development
- AI data centers have "rapid and large swings in demand" that stretch technical capabilities of even on-site gas plants (IEA, April 2026)
07 The Freelancer Problem — The Cost Nobody Is Counting
If you earn money online — as a writer, developer, designer, translator, data analyst, virtual assistant, or any other digital professional — the tech-energy convergence is hitting you from two directions simultaneously. Almost no mainstream analysis discusses both impacts together.
Hit 1: Your Electricity Bill Is Their Operating Cost
- Every hour you work from home, you consume electricity for your computer, monitors, router, lighting, and cooling
- In high-impact areas near data centers, this electricity costs up to 267% more than five years ago
- This represents a direct, invisible reduction in freelancer net income — one that does not appear in any invoice or tax calculation
- You did not vote for these deals. The contracts were made behind NDAs. Your utility commission had limited power to object
- The Ratepayer Protection Pledge is voluntary and unenforceable — you have no legal protection
- For a freelancer working full-time from home, the cumulative electricity cost increase since 2020 represents hundreds of dollars annually that directly reduce take-home income
Hit 2: The AI Running on That Energy Is Competing For Your Work
- Writing and content — ChatGPT, Claude, Gemini already displace significant entry-to-mid-level writing work
- Code generation — GitHub Copilot, Claude Code, and other tools reducing demand for routine coding tasks
- Design — Midjourney, DALL-E, Adobe Firefly displacing stock illustration, basic graphic design, social media imagery
- Translation — DeepL and AI-powered translation reducing demand for routine language work
- Data entry and analysis — AI tools automating data processing tasks at scale
- Platform evidence — Upwork and Fiverr have reported declining average project values in AI-susceptible categories over the past two years
- The double compression — rising electricity costs + falling AI-disrupted revenue = a financial squeeze that most freelancer income advice has not yet caught up to
What Freelancers Can Actually Do
- Shift toward AI-resistant skills — complex strategic judgment, original investigation, client relationship management, culturally specific content, privacy-sensitive tasks
- Use AI tools actively — rather than competing against AI, use it to multiply your own productivity; the productivity gains are real even as the market contracts in some categories
- Track your utility rate cases — state public utility commissions hold public comment periods on rate hike proposals; your voice has legal standing in that process
- Solar panels where viable — EnergySage estimates 6-9 year payback periods; meaningful for full-time home workers in high-electricity-cost areas
- Advocate for large load tariffs — policies requiring data centers to pay their proportional grid costs protect residential ratepayers including home-based workers
- Diversify income geography — if your current market is heavily AI-disrupted, the global demand for human judgment, cultural fluency, and trusted relationships is growing, not shrinking
08 The Verdict — Three Realities Operating Simultaneously
I have spent weeks going through congressional letters, IEA reports, utility filings, academic research, and corporate press releases to write this investigation. My conclusion is not pessimistic — but it is honest.
The tech-energy convergence is real, necessary, and in some ways genuinely promising. A world powered by clean nuclear energy and made more efficient by AI is a better world. The technology is not the problem.
The problem is the process. Secret contracts. Voluntary pledges with no teeth. Communities bearing costs they never authorized, from deals they never saw, for companies whose shareholders they are not. A fossil-fuel bridge that will last a decade while the nuclear narrative is marketed today. Water crises unfolding in real-time while the industry calls itself sustainable.
And for those of us in the developing world — in Pakistan, India, Nigeria, the Philippines — we face a uniquely cruel version of this story. The same AI infrastructure that raises our energy costs and displaces our work was built with capital we don't have, by companies we don't control, governed by regulations we have no influence over.
The nuclear future, if it arrives on schedule, may genuinely transform global energy. But the gap between that future and today is a decade of gas bridges, rising bills, secret deals, and dried-up water tables. That gap deserves far more scrutiny than the press releases acknowledge — and far more representation for the communities paying the cost while others capture the benefit.
Frequently Asked Questions
15 most-searched questions — answered with data
AI data centers require constant 24/7 baseload power that solar and wind cannot reliably provide. Nuclear is the only carbon-free source that runs continuously. Training one large AI model consumes more electricity than 100 US homes use in a year. Microsoft signed a 20-year $16B deal to restart Three Mile Island. Google ordered 500MW of Small Modular Reactors from Kairos Power. Amazon invested $500M in X-energy. The nuclear pivot is driven entirely by the physical demands of AI infrastructure, not ideological commitment to clean energy.
Generating 100 words with GPT-4 consumes up to 3 bottles of water for server cooling. Global AI water footprint could reach 312-764 billion liters in 2025 alone — equivalent to the world's entire annual bottled water consumption. Texas data centers will use 49 billion gallons of water in 2025, rising to 399 billion gallons by 2030 — equivalent to drawing down Lake Mead by 16 feet annually. About two-thirds of data centers built since 2022 are in water-stressed areas. Evaporative cooling permanently removes water from local water cycles through evaporation.
Yes, in several regions. Phoenix draws 40% of its water from the Colorado River system whose reservoirs have fallen from 90% capacity in 2000 to around 30% today — while simultaneously becoming a major data center market. Texas, Arizona, and Virginia data centers are competing directly with agriculture and residential users for shrinking water supplies. Ireland and Singapore have restricted new data center approvals partly due to resource constraints. The Lawrence Berkeley National Laboratory confirms the majority of AI-specialized data centers use evaporation-based cooling that permanently removes water from local supplies.
Serious shortages are already forming. Ireland's data centers consumed 21% of total national electricity in 2023, heading to 30% by the early 2030s. Northern Virginia faces multi-year grid connection delays. Singapore has restricted new data center approvals. 50% of global data center projects in 2026 face delays due to power limitations. The IEA projects data center electricity consumption will double by 2030. Countries with fragile grids face the greatest risk as energy equipment and investment is diverted to wealthy-country AI infrastructure.
US residential electricity prices rose 36% between 2020 and February 2026 — from 12.76 to 17.44 cents per kilowatt-hour. In areas near major data center clusters, electricity costs up to 267% more than five years ago. New Jersey bills surged over 20% in 2025 alone. The Federal Reserve Bank of Dallas estimates wholesale prices could rise another 50% as data center demand doubles. Utilities requested a record $31 billion in rate hikes in 2025 — more than double the 2024 figure. Expected price by September 2027: 19.01 cents per kWh.
Announced by President Trump during his February 2026 State of the Union address, the Ratepayer Protection Pledge was signed by Amazon, Google, Meta, Microsoft, OpenAI, and others. They committed to build their own power and pay for grid infrastructure upgrades. However, the pledge is entirely voluntary with no enforcement mechanism, no penalties for non-compliance, and no independent oversight. Critics including Public Citizen have called it "a PR document with few specifics or teeth" — a voluntary commitment that companies can abandon without consequence.
Freelancers face a double compression: rising electricity costs directly increase home office operating costs (up to 267% more in high-impact areas), while simultaneously the AI tools powered by these energy-hungry data centers compete with freelancers for writing, coding, design, and translation work. Upwork and Fiverr have reported declining average project values in AI-susceptible categories. The result is rising operating costs and falling revenue simultaneously — a financial squeeze that most freelancer income advice has not yet fully addressed.
Microsoft has contracted 34.7 GW of clean power — the largest corporate clean energy commitment in history, surpassing Amazon. Key deals include Three Mile Island (837MW, $16B, 20-year) and $15B in Nordic infrastructure. Google made the most strategically significant structural move by acquiring Intersect Power for $4.75B — buying a generation company outright rather than just purchasing electricity. Meta has the most diversified nuclear portfolio at up to 6.6 GW. All four hyperscalers are collectively guiding to approximately $700 billion in 2026 capex.
A Small Modular Reactor produces 80-300 MW — much smaller and faster to build than conventional 1,000+ MW reactors. Google's Kairos Power fleet targets first reactor online in 2030, full fleet by 2035. Amazon X-energy targets 5GW by 2039. SMRs do not yet commercially exist in the US. In the meantime, utilities are building natural gas plants as a fossil-fuel bridge — meaning the clean energy narrative is 5-15 years ahead of the current reality. Nuclear waste management frameworks for SMR waste types are also not yet fully established.
A single ChatGPT text query uses nearly 10 times as much electricity as a Google Search. Generating an AI image uses thousands of times more energy than a text search — equivalent to charging a smartphone. ChatGPT used over 500,000 kilowatt-hours of electricity daily as of 2024. Generating 100 words with GPT-4 consumes up to 3 bottles of water for cooling. AI agents — autonomous multi-step AI workflows — are dramatically more energy and water intensive than simple chatbot queries, representing the fastest-growing demand category.
Currently legal in many US states — which is exactly why a bipartisan group of US senators led by Markey and Capito demanded transparency. These NDA-protected contracts between utilities and tech giants like Amazon, Google, Microsoft, and Meta shape electricity prices for communities that have no legal right to see the contract terms. FERC oversees interstate transmission, but data center cost allocation falls under fragmented state-level regulation that varies widely in consumer protection. 11 states are now considering legislation to impose moratoriums on new data center approvals pending full impact assessments.
Significant and mostly negative in the near term. AI infrastructure is 90%+ concentrated in North America, Western Europe, and Asia-Pacific. Developing countries compete for scarce transformers, generators, and grid equipment now prioritized by wealthy tech companies. Rising global energy prices increase import costs for fuel-dependent nations. Simultaneously, AI tools displace the outsourced writing, coding, customer service, and data work that millions of people in Pakistan, India, Nigeria, and the Philippines depend on for income. No seat at the table for affected developing-country communities in any of these decisions.
Theoretically possible over the long term. Nuclear SMRs, once operational, could provide cheaper stable power. AI efficiency per task is improving at an unprecedented rate (IEA). But the operative word is eventually — SMR fleets arrive 2030-2039. Gas bridges are being built now. Bills are rising now. Water is being drained now. Communities paying the development cost of this infrastructure are unlikely to see lower prices for a decade or more, while companies that built it will capture the financial returns starting immediately. The timeline asymmetry is the core ethical problem.
The carbon footprint of AI systems alone is estimated at 32.6 to 79.7 million tons of CO2 emissions in 2025 — equivalent to the annual carbon output of New York City, according to ScienceDirect research. The water footprint is similarly substantial: 312-764 billion liters — equal to global annual bottled water consumption. These figures are growing rapidly as AI adoption scales. The IEA projects electricity consumption from AI-focused data centers will triple by 2030. Despite tech companies' clean energy narratives, the fossil fuel bridge between now and SMR deployment means current AI is substantially powered by gas.
Practical steps: (1) Monitor your state public utility commission's rate cases and file public comments when data center-driven rate hikes are proposed — your voice has legal standing. (2) Join consumer energy advocacy organizations. (3) Install solar panels where economically viable — EnergySage estimates 6-9 year payback. (4) Advocate for "large load tariffs" requiring data centers to pay proportional grid costs. (5) For freelancers: diversify into AI-resistant skill areas (strategic judgment, cultural fluency, trusted relationships). (6) Track your electricity usage with smart meters to reduce controllable costs while advocating for systemic change.
