What’s going on in: 🤖 AI & Compute
A 5-minute crash briefing on compute, carbon, and the coming power crunch
👋 Welcome to Climate Drift: your cheat-sheet to climate. Each edition breaks down real solutions, hard numbers, and career moves for operators, founders, and investors who want impact. For more: Community | Accelerator | Open Climate Firesides | Deep Dives
Hey there! 👋
Skander here.
The Drop & NY Climate Week are around the corner. Got your talking points ready?
If not: don’t worry. Here’s a 5-minute crash briefing to make sure you can hold your own when someone brings up “AI is eating the grid.”
Each sector briefing hits:
⚡ TLDR
📊 Key Numbers
📜 Last Year’s Narratives
🏆 Winners & 🫠 Losers
🔮 Next Year’s Narratives
👀 Who to Watch
🌊 Let’s dive in
Want to discuss these topics with Drifties and some of the world’s best investors?
Join our presessions next week:
🤖 AI & Compute (Monday)
⚡ Grid & Energy (Tuesday)
🤝 Scaling & Collaboration (Tuesday)
🌱 Food & Nature (Wednesday)
🛡️ Adaptation & Resilience (Wednesday)
💸 Investment & Impact (Thursday)
🌍 Geopolitics & Global Markets (Thursday)
♻️ Circularity & Materials (Friday)
What’s going on in: 🤖 AI & Compute
⚡ TLDR
AI is chewing through the world’s electricity like Pac-Man on steroids.
Data center demand will triple by 2030, straining grids already creaking under renewables build-out delays. But here’s the paradox: the same AI could help us save up to 5 billion tons of CO₂ a year. Spend carbon to save carbon.
📊 Key Numbers
415 TWh → 945 TWh: Global data center demand, 2024 → 2030. That’s Japan’s entire grid.
9%: Share of US electricity AI/data centers could swallow by 2035.
30B → 338B liters: AI data centers’ water use, 2023 → 2030.
$200B+: Global AI infra spend by 2028.
$6.7T: Goldman Sachs estimate for total spend to keep up with compute demand by 2030.
166 Mt CO₂: Data center emissions by 2030 (up from 29 Mt in 2023).
3x: Carbon efficiency gains between TPU v4i → TPU v6e.
📜 Last Year’s Narrative
2024 was the year AI jumped from “cool demo” to “civilization-scale bet.”
The story was techno-optimism on steroids: AI will solve climate, cure disease, and make work easier, while efficiency gains keep the power bill manageable. Smart grids, autonomous cars, infinite PPAs, even fusion.
Reality looked different. Google’s emissions rose 48% since 2019. Microsoft’s footprint surged 155%. Sustainability reports quietly added caveats: “Reducing emissions may be challenging as AI scales.”
The vibe was clear: scale the model, wow the world, raise the money, and let someone else figure out the power.
And yet the promise lingered. AI-curated supply chains, satellites catching methane leaks, new materials discovery.
🏆 Winners & 🫠 Losers
🏆 The Winners
Turbine makers like GE Vernova, Siemens, and Mitsubishi, sold out into the next decade and raising prices at will.
Google and Emerald AI, proving data centers can actually flex demand by 25–40%. Nuclear stocks like Constellation and Vistra, surging on the hope of small modular reactors still years away.
Last but not least: The bridge power brigade: diesel and gas units rebranded as “AI-ready solutions.”
🫠 The Losers
Teams without a power plan, stuck in decade-long interconnection queues.
Anyone betting on cheap memory arriving next quarter.
Utilities insisting on 24/7 firm load, overbuilding stranded peakers.
And climate startups, pushed to the margins as AI vacuumed up all the venture capital oxygen.
🔮 Next Year’s Narratives
The next 12 months will be a reality check:
💸 The Revenue Reckoning
The $600 billion question becomes the $1 trillion question. OpenAI burns through another $10B. Someone asks if anyone's actually making money from AI. Stock market pretends not to hear.
⚡ With great power comes great responsibility
Siting near gigawatt nodes, locking PPAs, and reusing waste heat moves from comms fluff to core strategy.
🔨 Regulatory Hammer Time
Dublin and Amsterdam froze new builds. Singapore lifted its moratorium only with strict efficiency rules. The EU now forces big centers to report energy use, cut cooling losses, and reuse waste heat. Caps on power, efficiency mandates, and siting restrictions make “build anywhere” a thing of the past.
⚔️ The Great Power Struggle
By 2035, data centers could eat 8.6% of US electricity. Utilities will be forced to choose between AI clusters, households and industry.
✨ The Flexibility Gold Rush
Everyone suddenly discovers data centers don't actually need 99.999% uptime. Utilities offer "FastTrack Flex" programs: 4-year interconnection instead of 8 if you promise to turn off during the Super Bowl.
🌀 The Turbine Wars
With 4-year lead times, securing turbines becomes the new securing chips. Smaller players resort to diesel generators marketed as "AI-ready transitional power solutions." Someone definitely tries to blockchain the turbine supply chain.
☢️ The Nuclear Reality Check
First SMR delays announced (shocking nobody). Three Mile Island restart pushed to 2029. Google quietly extends its fossil fuel contracts while maintaining nuclear is "the future."
💡 The Efficiency Awakening (Maybe)
After DeepSeek's moment, the industry split: Half doubles down on brute force scaling, half starts actual efficiency work. Jon Koomey becomes either prophet or pariah depending on Q3 earnings. Someone rediscovers that inference doesn't need frontier model performance (yes I am feeling the pain of using GPT 5 too).
👀 Who to Watch
Varun Sivaram | Emerald AI
Ex–chief technology officer at ReNew Power and former advisor to the U.S. State Department, now Founder & CEO of Emerald AI. He’s proving data centers can flex loads by 25–40% without breaking service. Oh and he also wrote one of my favorite books on solar.
LinkedIn →
Caie Kelley | Lowercarbon Capital
Partner at Lowercarbon Capital, where she drives growth-stage bets across batteries, materials, and climate software. Known for sharp analysis and focus on scaling and traction (the thing that matters both in AI & climate), she’s one of the people deciding which moonshots get rocket fuel.
LinkedIn →
Ben Eidelson | Stepchange
Founder of Stepchange, a climate-focused VC fund backing early-stage software and AI startups. Ex-Google & Stripe, he’s betting that smarter code & product can speed up everything from grid integration to carbon accounting.
LinkedIn →
Join our community discussions, we’d love to hear your take.
If you’re passionate about climate solutions, share this briefing with your friends and family.
And don’t forget to join our sessions next week:
🤖 AI & Compute (Monday)
⚡ Grid & Energy (Tuesday)
🤝 Scaling & Collaboration (Tuesday)
🌱 Food & Nature (Wednesday)
🛡️ Adaptation & Resilience (Wednesday)
💸 Investment & Impact (Thursday)
🌍 Geopolitics & Global Markets (Thursday)
♻️ Circularity & Materials (Friday)