# Is Energy the New Gold of AI? > Google just canceled a data center for the first time ever, under citizen pressure. AI's energy crisis is real, and smart money is already moving. - Canonical: https://siliconvalleyconfidential.com/en/dossier/is-energy-the-new-gold-of-ai/ - Site: Silicon Valley Confidential (https://siliconvalleyconfidential.com) — weekly executive intelligence on Silicon Valley and global tech - Author: Jose Luis Cases (https://es.linkedin.com/in/jose-luis-cases-lozano) - Language: en - Published: 2025-10-18 (original LinkedIn edition: https://www.linkedin.com/pulse/es-la-energ%C3%ADa-el-nuevo-oro-de-ia-jose-luis-cases-j6upf/) --- THE BOMBSHELL OF THE WEEK --------------------- Google CANCELED a data center last Tuesday for the first time in its history: the AI energy crisis is real On October 14, Google withdrew its proposal for a data center in Franklin, Indiana. Not for lack of capital. Not because of a strategy shift, but because of citizen opposition worried about water and electricity consumption. It's the first time in Big Tech history that a data center has been canceled due to local pressure over energy resources. The numbers are scary: * US data centers: 4.4% of total electricity (2024) → 6.7%-12% projected (2028) * OpenAI signed a deal with Nvidia for 10 GW of capacity — that's all the electricity NYC consumes in summer * Big Tech (Amazon, Google, Microsoft, Meta) adds up to $320B in AI infrastructure investment (2025) * NPR report (Oct 14): utilities face a "multibillion-dollar" question — how much AI demand is REAL? Utility companies are building capacity without knowing whether the demand will materialize. It's the biggest energy bet since the oil rush. Google lost in Franklin because citizens understood their electricity bills would go up to subsidize AI. The dangerous precedent: if citizens can block Google, they can block anyone. Amazon, Microsoft and Meta have 40+ data centers slated for 2025-2026. Every single one will face similar opposition. CNBC reported (Oct 17) that utilities are publicly admitting they don't know whether they can meet AI demand. It's the first time they've said it out loud. We'll see 3-5 more data center cancellations in Q4 2025 driven by local opposition. Big Tech will have to: (a) pay enormous premiums for electricity, (b) build in remote locations without infrastructure, or (c) look for alternative decentralized solutions. And this is where Framework Ventures comes in... * * * MONEY TALKS --------------- ### Daylight Energy: $75M — Framework bets that energy is AI's real bottleneck (Oct 16) The timing: * Oct 14: Google cancels data center (local opposition) * Oct 16: Framework invests $75M in Daylight, 48 hours later What Daylight does: it installs solar panels + batteries in 1,200 homes. When data centers need electricity, they buy it from this network of houses instead of building giant centralized plants. Think of it as the Airbnb of electricity: homeowners rent out their solar capacity to whoever pays the most (AI data centers, utilities during demand peaks). Why Framework is investing: Google lost in Franklin because a giant data center = easy to block politically. 1,200 solar-powered homes are impossible to block. Are you going to ban solar panels on voters' houses? I don't think so. The thesis: Vance Spencer (Framework): "AI accelerates power demand. Daylight connects capital to renewable infrastructure." Everyone is investing in AI chips. Framework is betting on the real bottleneck: electricity nobody can block. Framework normally invests in speculative crypto/blockchain projects. Daylight is different: real physical infrastructure (solar panels) generating real revenue (electricity sales to data centers). * * * ### Woz: $6M for the anti-vibe-coding "AI app factory" The round: * $6M seed * Cervin Ventures (lead) * Y Combinator, Burst Capital, Untapped Ventures, MGV, Lacob Family (Warriors owners) Founders: Ben Collins and Brad Eckert, both ex-MIT While everyone else does AI coding with total freedom (Cursor, Copilot, etc.), Woz is betting on "anti-vibe coding": AI with strict guardrails and mandatory human oversight. Every app goes through review by a human engineer. It's the first seed-stage startup to openly say "autonomous AI doesn't work for enterprise production." They're selling exactly the opposite of the dominant narrative. And they have traction — they're hiring to scale. They're not selling speed. They're selling trust. For enterprise CTOs who can't afford broken AI-generated apps, this could work. * * * NUMBERS vs REALITY ------------------- ### Oracle & AMD: the chip deal nobody celebrated The official announcement: "Oracle will use AMD chips in its cloud services for AI workloads" The real numbers that matter: * AMD is Nvidia's "closest rival" in AI chips * Oracle had been 100% Nvidia until now * This is AMD's first significant win in hyperscaler cloud AI Oracle is diversifying suppliers because Nvidia has a 9-12 month backlog on H100/H200s. It's not a technical bet on AMD — it's supply chain risk management. AMD is probably offering 30-40% discounts vs Nvidia to win share. Oracle wouldn't do it for "superior performance" (because there isn't any). They're doing it for availability + price. * * * VALLEY DRAMA — culture and conflicts ------------------------------------------ ### JPL (NASA): the exodus toward defense tech startups JPL (Jet Propulsion Lab in Pasadena) is on its third round of layoffs in 6 months due to NASA budget cuts. The employees who are NOT being laid off are also leaving voluntarily. Where to? Defense tech startups (Anduril, Shield AI, etc.) paying 2-3x + equity. Defense tech startups have raised $38B in H1 2025 alone (JPMorgan data). They're hiring aggressively and offering packages NASA can't match. The result: JPL loses talent through two channels: 1. Forced layoffs → available talent that defense tech absorbs 2. Voluntary resignations → people choosing to multiply their salary And on to the last topic.... MY TAKE --------------- Decentralized energy is gaining momentum again. There was a lot of talk about this back when bitcoin mining depended heavily on electricity costs and there were projects to mine BTC next to hydroelectric plants to buy cheap energy. In Norway/Iceland, miners used surplus hydroelectric and geothermal power. In China, specifically in Sichuan, miners took advantage of the rainy season to buy cheap energy pre-2021. But with that volatility, China's 2021 mining ban, and the impossibility of signing 10-year PPA contracts around something so volatile, it never took off. Now I see it differently... I like the Daylight Energy idea, and a possible new life for owners of solar PV plants, even in Spain. If AI needs energy, we'll sell it to them — but at a premium :) I think Framework will chase 3-4 more decentralized-energy deals before year end. "Energy infrastructure for AI" will be a top 3 VC category in 2026. As for Woz, what can I say... a few days ago I wrote a post about AI and the 80% problem, which you can read here (later — first finish this one and share it :)) [https://www.linkedin.com/feed/update/urn:li:activity:7381727969473531904/](https://www.linkedin.com/feed/update/urn:li:activity:7381727969473531904/?trk=article-ssr-frontend-pulse_little-text-block) Andrej Karpathy has said the true AGI coding agent is still a decade away. The fight for the 20% is coming. Companies like [Maisa](https://www.linkedin.com/company/maisa-agentic-process-automation?trk=article-ssr-frontend-pulse_little-mention), from Valencia's own [David Villalon](https://www.linkedin.com/in/davidvillalonpardo?trk=article-ssr-frontend-pulse_little-mention) and his co-founder — to whom I wish the best of luck — are trying to solve it. Delivering production-grade products, whether with humans in the loop or however your imagination designs it, but finished and "sellable." If they succeed, we'll see massive pivots in 2026 from AI coding startups toward "human-in-the-loop" as a feature, not a bug. And finally, AMD... for me this is an investment story. It's up 93% on the year, but if you think about it, as a counterweight to NVIDIA it might even be cheap. I think we'll see 2-3 similar announcements in Q4 from other hyperscalers "diversifying" toward AMD, since Nvidia can't deliver fast enough. And that's it for this week. This one was harder to write — lots of data, lots of verification and cross-checking — but by making it for you, I learn. I hope you do too. See you next week. PS: subscriptions aren't growing as much as I'd like. If you enjoyed this, please comment, share it, drop it in your geek friends' WhatsApp group. All help is welcome :)