When the Panic Becomes Obvious
On September 20, 2024, Microsoft announced something extraordinary: a $1.6 billion deal with Constellation Energy to restart Three Mile Island Unit 1, bringing 837 megawatts of nuclear power back online with a target date of 2028.
Let me be clear about what this means. Microsoft, the company that spent decades preaching about renewable energy, carbon neutrality, and environmental responsibility, just committed $1.6 billion to restart a nuclear reactor at Three Mile Island. The site of America's worst nuclear accident. The name synonymous with nuclear fear for an entire generation.
They're not doing this because they love nuclear power. They're doing it because they're bloody desperate.
And they're not alone. Within weeks, Google announced partnerships for Small Modular Reactors. Amazon committed to multiple nuclear investments totaling billions. Meta issued requests for proposals seeking 1 to 4 gigawatts of new nuclear capacity.
The tech giants who built empires on the promise that cloud computing would be more efficient, more sustainable, and cheaper than traditional infrastructure just admitted they can't keep the lights on without nuclear power.
And your business is paying for their panic through cloud bills that climb 10 to 14% annually while they scramble to secure power supplies that won't arrive until 2028 to 2035.
The Nuclear Commitments: Exact Numbers, Exact Timelines
Let's document precisely what's happening, because the scale is staggering.
Microsoft and Three Mile Island:
Amount: $1.6 billion deal with Constellation Energy
Capacity: 837 megawatts (Unit 1, renamed Crane Clean Energy Center)
Timeline: Target restart 2028
Duration: 20-year power purchase agreement
Context: Unit 1 operated safely 1974-2019, closed for financial reasons not safety
Additional: Stargate supercomputer network requiring 5 gigawatts (equivalent to five nuclear plants) across Wisconsin, California, Texas, Virginia, and Brazil
Unit 1 never had an accident. That was Unit 2 in 1979. But the name alone tells you how desperate Microsoft is. They're willing to restart a facility at America's most infamous nuclear site because renewable energy simply cannot deliver the reliable baseload power AI data centres require.
Google and Kairos Power:
Announcement: October 14, 2024
Capacity: 500 megawatts total across 6 to 7 Small Modular Reactors
Timeline: First SMR online by 2030, additional through 2035
Significance: First corporate agreement for SMR development
Technology: Advanced reactor design never commercially deployed at scale
Google's betting on unproven technology. SMRs are theoretically safer, cheaper, and faster to build than traditional nuclear plants. Theoretically. They've never been commercially deployed at scale in the United States. Google's essentially funding the research, development, and deployment of an entirely new nuclear technology because they're that desperate for power.
Amazon's Multi-Pronged Nuclear Strategy:
Talen Energy: $650 million deal for 480 to 960 megawatts from Susquehanna Nuclear Plant (Pennsylvania)
Dominion Energy: Exploring SMR near North Anna nuclear station (Virginia)
Energy Northwest: Supporting 4 SMR projects (320 megawatt Phase 1, potential 960 megawatt scale-up)
X-Energy: $500 million investment for SMR development (early 2030s target)
Amazon's not picking one approach. They're funding everything simultaneously because they need power so badly they can't afford to bet wrong.
Meta's Gigawatt-Scale Ambitions:
December 2024: Request for Proposals for 1 to 4 gigawatts new nuclear capacity
June 2025: 20-year agreement for roughly 1.1 gigawatts from Clinton Clean Energy Center (Illinois) starting June 2027
Impact: Preserves 1,100+ jobs, $13.5 million annual tax revenue
Meta's RFP for up to 4 gigawatts is equivalent to four large nuclear power plants. That's extraordinary. Facebook started in a dorm room 21 years ago. Now they're securing nuclear power equivalent to small countries.
OpenAI and Sam Altman's Oklo:
Altman chairs Oklo, developing liquid metal-cooled microreactors at Idaho National Laboratory
2024 White House pitch proposed multiple 5-gigawatt data centres each requiring equivalent of five nuclear plants
The CEO of the company that created ChatGPT is personally invested in nuclear microreactor development. Connect those dots.
These aren't announcements. These are distress signals.
Why Renewable Energy Can't Solve This
The tech giants spent the last decade telling everyone renewable energy was the future. Solar farms. Wind turbines. Massive battery storage. Carbon-neutral data centres powered by clean energy.
What happened?
Intermittency killed the dream: Solar panels don't generate electricity at night. Wind turbines don't spin without wind. Data centres require 24/7 power with zero interruptions. Any gap in supply means servers crash, data corrupts, and SLA agreements get violated.
Battery storage technology can't bridge the gap at data centre scale. The largest battery storage facilities in the world can power modest loads for hours, not gigawatt-scale data centres for days. And batteries degrade. They're expensive. They require rare earth minerals with problematic supply chains.
AI workloads are baseload-hungry: Traditional IT workloads have predictable patterns. Higher demand during business hours, lower at night. Renewable energy with storage can handle those patterns reasonably well.
AI training runs don't care about business hours. They run 24/7 for weeks or months continuously. GPUs burn maximum power constantly. There's no off-peak. There's no load balancing. It's full-throttle power consumption around the clock.
Only baseload power sources work: coal (environmental disaster), natural gas (fossil fuel with emissions), hydroelectric (geography-dependent and mostly tapped out), or nuclear (politically controversial but zero-emissions during operation).
Grid capacity is the bottleneck: Even if renewable energy could theoretically provide enough power, the electrical grid can't deliver it. Data centre projects are facing 5 to 15 year waits for grid connections. The queue has grown 10-fold over the last 5 years. In southern England, data centres are told to wait 5 to 7 years for additional power.
Building new renewable generation doesn't help if you can't connect it to data centres. Nuclear plants, especially SMRs, can be located near data centres, reducing transmission requirements.
The mathematics don't work: The International Energy Agency projects data centre electricity consumption will double by 2030 to 945 terawatt-hours, equivalent to Japan's total electricity consumption. That's with optimistic assumptions about efficiency improvements.
AI adoption is accelerating faster than IEA models assumed. If current trends continue, data centres could consume 3 to 5% of global electricity by 2030 instead of the projected 3%. Building enough renewable capacity to meet that demand while maintaining grid stability is mathematically implausible with current technology.
Nuclear isn't a choice. It's the only option that delivers the reliable, high-density, zero-emission baseload power AI requires at the scale tech giants need.
The 2028-2035 Gap Nobody's Talking About
Here's the problem that should terrify anyone paying attention: every nuclear commitment has delivery timelines of 2028 to 2035.
Microsoft's Three Mile Island restart: target 2028. Google's first SMR: 2030, with additional units through 2035. Amazon's projects: early 2030s. Meta's nuclear capacity: similar timelines.
AI's energy demands are happening right now. Today. This quarter. ChatGPT launched November 2022. Within three years, it drove demand that forced Microsoft to commit $1.6 billion to nuclear power.
What happens between now and 2028? Between now and 2035?
Scenario 1: Constrained growth. AI companies limit deployment based on available power. Model training slows. Inference capacity caps out. The AI revolution stalls waiting for nuclear plants to come online.
This seems unlikely given competitive pressures. No tech giant wants to cede AI leadership because they couldn't secure power.
Scenario 2: Fossil fuel bridge. Data centres run on coal and natural gas during the transition period. All the carbon-neutral promises evaporate. Emissions spike. Climate commitments get abandoned because AI is too valuable to throttle.
This is probably what's happening quietly. Nobody's advertising it, but data centre natural gas consumption is climbing.
Scenario 3: Cost explosion. Limited power capacity drives prices up. Cloud providers pass costs to customers through higher pricing. SaaS vendors increase subscription fees. Small businesses get priced out of AI tools they need to remain competitive.
This is definitely happening. Your cloud bills prove it.
Scenario 4: Grid instability. Data centres consume so much power that electrical grids become unstable. Brownouts in data centre regions. Priority power allocation conflicts. Consumer electricity costs spike as residential users subsidize industrial AI consumption.
Early warning signs visible in Virginia, Illinois, and parts of the UK where data centre concentration is highest.
The five to ten year gap between AI's current energy appetite and nuclear power delivery creates a crisis nobody's solved. And you're paying for it through rising cloud costs while tech giants frantically build nuclear plants.
What This Means for UK Small Businesses
You might think, "That's America's problem. Nuclear politics, data centre locations, grid capacity. I'm in Birmingham running a 23-person business. How does this affect me?"
Direct cost pass-through: Microsoft Azure, Amazon AWS, and Google Cloud serve UK businesses from global data centre networks. Energy costs in one region affect pricing globally. When Virginia's electricity costs spike, Azure pricing increases everywhere.
Your cloud bill reflects American data centre energy crises, Chinese grid capacity issues, and European renewable energy shortfalls. You're subsidizing global infrastructure problems through subscription fees.
UK-specific energy pressures: UK data centres currently consume 2.5% of UK's electricity, expected to quadruple by 2030. UK data centre electricity costs are four times more expensive than the United States and the most expensive compared to other European markets.
The UK faces the worst of both worlds: high energy costs driving cloud pricing up, and insufficient domestic data centre capacity forcing reliance on international providers with their own cost pressures.
SaaS vendor amplification: Every SaaS application you use runs in someone's data centre. When their infrastructure costs increase 10 to 14%, they pass it through. Salesforce, Slack, Zendesk, HubSpot, Monday.com: every vendor facing rising cloud bills adjusts pricing.
SaaS inflation at $9,100 per employee in 2025 (up 15% over two years) partly reflects this energy cost cascade. You're not just paying for your direct cloud usage. You're paying energy premiums embedded in every software subscription.
Competitive disadvantage: Larger enterprises negotiate multi-year contracts with price protections. They have procurement teams analyzing total cost of ownership. They can absorb 10% annual increases.
Small businesses renew monthly or annually at current market rates. You have no negotiating leverage. You eat every price increase immediately. This creates a growing cost disadvantage versus larger competitors who locked in better terms.
The AI productivity trap: Microsoft Copilot wants £30 per user per month as an add-on. Salesforce Einstein charges $500 per month per seat. Every vendor is adding "AI-powered features" with 20 to 30% premiums.
The productivity tools that could help you compete are increasingly unaffordable specifically because they consume so much energy. You need AI to remain competitive, but AI costs are rising faster than the productivity gains justify. That's a trap with no easy escape.
The Nuclear Solution That Might Not Arrive in Time
Even if every nuclear commitment delivers on schedule (which history suggests is optimistic), the timelines are daunting.
Three Mile Island Unit 1 restart: Target 2028. That's regulatory approvals, infrastructure repairs, safety inspections, staffing, testing, and commissioning. Any delays push to 2029 or 2030. Nuclear projects are notorious for schedule overruns.
Small Modular Reactors: Google's betting on SMRs delivering by 2030. But SMRs have never been commercially deployed at scale in the United States. The first-of-a-kind deployment always encounters unexpected challenges. First commercial SMR might realistically arrive 2032 to 2035.
New construction: Traditional nuclear plants take 10 to 15 years from planning to operation. Even with streamlined approvals and massive funding, you're looking at 2035+ for new large-scale nuclear facilities.
Meanwhile, AI capabilities are doubling roughly every 18 months. GPT-5, GPT-6, and beyond will demand exponentially more compute. The gap between AI's energy appetite and nuclear power delivery is widening, not closing.
Renewables continue improving: Solar efficiency gains, better battery storage, and smart grid technology might close part of the gap. But baseload nuclear remains essential for 24/7 data centre operations at gigawatt scale.
The most likely outcome: a messy transition period of 5 to 10 years where AI growth outpaces power capacity, fossil fuels bridge the gap despite carbon commitments, costs continue climbing, and small businesses bear the burden through higher subscription fees and cloud bills.
That's not a future anyone planned. That's the future we're getting because tech giants scaled AI faster than energy infrastructure could support.
The Uncomfortable Questions Nobody's Asking
Question 1: If AI requires nuclear power plants to run, is it actually sustainable?
The tech industry sells AI as clean, efficient, and environmentally friendly. But if the only way to power AI at scale is restarting nuclear plants and building new reactors, what does that say about sustainability claims?
Nuclear produces zero emissions during operation. But it generates radioactive waste requiring thousands of years of storage. It creates catastrophic risks if safety fails. It requires mining uranium with environmental impacts. It's better than coal, but calling it "clean energy" stretches credibility.
Question 2: Who decided AI was worth nuclear power plants?
Nobody held a public referendum asking, "Should we restart Three Mile Island to power ChatGPT?" Nobody consulted small business owners about whether AI features justify nuclear risks and multi-billion-dollar infrastructure investments.
Tech giants made unilateral decisions that AI was valuable enough to warrant nuclear power. Then they passed the costs downstream to customers through higher pricing. You're funding nuclear plants whether you want AI or not.
Question 3: What happens if nuclear timelines slip?
Microsoft's 2028 target for Three Mile Island restart is aggressive. Regulatory approvals, equipment repairs, staffing, safety inspections: any delay pushes to 2029, 2030, or beyond.
If nuclear power doesn't arrive on schedule, what's Plan B? More fossil fuels? Curtailed AI development? Continued cost increases for businesses? Nobody's publicly discussing fallback plans.
Question 4: Are we prioritizing AI over human needs?
Electricity diverted to data centres can't power homes, hospitals, schools, or essential services. When data centres consume 3 to 5% of global electricity, that's capacity unavailable for human welfare.
Are ChatGPT responses worth potential brownouts in residential areas? Is AI-generated content worth higher electricity bills for families? Tech companies say yes. Have they earned the right to make that decision for everyone?
Question 5: What if biological computing or neuromorphic chips solve this first?
Microsoft's betting $1.6 billion on 2028 nuclear delivery. But what if FinalSpark's biological processors achieve commercial viability by 2030? Or Intel's neuromorphic chips deliver sufficient efficiency gains that nuclear becomes unnecessary?
Nuclear investments create sunk costs and path dependency. If better alternatives emerge before nuclear plants come online, tech giants will continue using nuclear because they've already committed billions. That's not optimal. That's organizational momentum overriding better options.
Nobody's asking these questions publicly because the answers are uncomfortable. But small business owners paying the bills deserve answers.
What You Should Actually Do About This
Stop paying for AI features you don't use. Seriously. Microsoft Copilot, Salesforce Einstein, every "AI-powered" add-on: audit them monthly. If you can't demonstrate concrete productivity gains exceeding the subscription cost, cancel them. You're not just paying for software. You're paying a premium to subsidize nuclear power plants.
Negotiate multi-year contracts where possible. Lock in current pricing before the next round of increases. Cloud providers and SaaS vendors offer discounts for multi-year commitments. Take them. You're hedging against energy cost inflation.
Diversify cloud providers. Don't put everything in Azure or AWS. Split workloads across providers. When one raises prices, you have negotiating leverage: "Move our workloads or we leave." Provider lock-in eliminates your bargaining power.
Watch for energy-efficient alternatives. Neuromorphic computing, edge computing, and hybrid architectures might offer better cost-performance as energy prices climb. Early movers identifying efficient platforms gain cost advantages.
Challenge your vendors. When SaaS subscriptions increase, demand specifics: What are we getting for the extra money? Can we opt out of AI features? Are there alternative pricing tiers without energy-intensive capabilities?
Vendors test how much they can push costs downstream. Push back. Enough customers refusing price increases forces vendors to justify their costs or lose business.
The Bottom Line: You're Funding Nuclear Plants Whether You Want Them or Not
Microsoft spent $1.6 billion to restart Three Mile Island. Google's betting on unproven SMR technology. Amazon's funding multiple nuclear projects simultaneously. Meta wants 4 gigawatts of new capacity.
These aren't investments in your business. These are panic responses to an energy crisis tech giants created by scaling AI faster than infrastructure could support.
And you're paying for their panic through cloud bills that climb 10 to 14% annually, SaaS subscriptions inflating five times faster than consumer prices, and AI feature premiums that cost more than the productivity they deliver.
The nuclear plants won't arrive until 2028 to 2035. Your costs are rising right now. That gap represents pure profit extraction: charging you for energy problems they haven't solved yet, funded by your budget while they scramble for solutions.
Tomorrow, we're examining exactly how these energy costs translate to your specific cloud bills and what UK businesses can do to protect themselves. Because understanding the problem is just the start. You need practical strategies to survive the next five to ten years while tech giants build their nuclear power plants on your dime.
Stay curious. Stay sceptical. And stay bloody furious about who's paying for this mess.
| Source | Article |
| Microsoft News Centre | Microsoft and Constellation to Power the Future of AI with Three Mile Island Nuclear Energy |
| Google Official Blog | Partnering with Kairos Power for First-of-a-Kind Clean Energy |
| Amazon Press Release | Amazon Nuclear Energy Investments 2024 |
| Meta Newsroom | Meta Nuclear Power RFP and Agreements |
| International Energy Agency | Energy and AI: Executive Summary |
| CNBC | Utilities Grapple with AI Data Centre Power Demand |
| Bloomberg | AI Data Centres Are Sending Power Bills Soaring |
| Vertice | SaaS Inflation Index 2025 |
| House of Commons Library | Data Centres: Planning Policy, Sustainability, and Resilience |
| Akita | Microsoft Increase Pricing By 9% As Cloud Costs Rise |