Trending Thursday #21

AI runs out of electricity + Hollywood sues everyone with a GPU

Author

Dev Chandra
October 23, 2025 • Estimated Reading Time: 14 minutes

This week proved that AI's biggest bottleneck isn't compute or data; it's finding somewhere to plug in the damn things.

Between Meta selling 80% of its data center to Wall Street and utilities begging for gas turbines like it's Black Friday, we're watching the entire energy grid reorganize itself around ChatGPT's electricity appetite.

As we continue to build at Context Ventures, we've launched Startup Intros with a simple mission: to help early-stage founders find the right investors, faster and smarter.

🎯 BLUF: Bottom Line Upfront

  • The $27B Desperation Play: Meta handed over 80% of its Hyperion data center to Blue Owl Capital because as it can't afford AI infrastructure alone

  • The 7-Year Turbine Wait: Gas turbine lead times exploded to 2030 as data centers force utilities to burn fossil fuels over clean energy

  • 52 Lawsuits and Counting: The AI copyright war hit critical mass with Warner Bros, Disney, and Universal deploying their legal tools

  • Reddit's Search Result Shakedown: Reddit is suing companies for scraping Google search results containing Reddit content

  • The 409 TWh Reality Check: US data centers will consume 9% of all electricity by 2030, doubling from today

⚡ The $100B Infrastructure Crisis: When AI Runs Out of Outlets

🔥 America's Fossil Fuel U-Turn for ChatGPT

The AI boom just triggered the most embarrassing energy policy reversal in history, as utilities scramble for gas turbines to power data centers that literally can't find enough electricity:

  • 7-year Wait Times: Gas turbine delivery pushed to 2030 from the normal 2 years

  • 19 GW Annually: New gas-fired capacity being built, doubling since 2021

  • 120 GW Retiring: Coal plants shutting down as replacements can't come fast enough

  • 75% Market Control: GE Vernova, Siemens & Mitsubishi hold all the cards

The Irony Olympics: The same companies pledging net-zero by 2030 are frantically building fossil fuel plants because their AI models need more juice than CA, TX, and NY combined. So every clean energy commitment just gets disregarded.

📊 The Brutal Math of AI's Energy Appetite

Bain & Company confirmed the nightmare scenario: US data centers will devour 409 TWh by 2030, doubling today's consumption and taking 9% of America's entire electrical output:

  • 409 TWh By 2030: More electricity than most countries use in a year

  • 150 TWh Above Baseline: The entire AI premium on the grid

  • 5-year Utility Delays: Just to get connected, if you're lucky

  • 8-24 Month Equipment Waits: Making construction planning impossible

The Reality: Bain calls the next phase "more disciplined," meaning that "everyone realized GPUs are expensive paperweights without gigawatts." Thus, the infrastructure arms race is causing companies to build on-site gas generators, so they can actually turn the GPUs on.

🌍 AI's Global Grid Meltdown

The infrastructure crisis has gone international, with 60% of the world's largest data centers now outside the US, exporting Silicon Valley's electricity problems to every continent:

  • 32% of Ireland's Grid: What data centers will consume by 2026 (21% today)

  • 79% in Dublin: The capital is basically a server farm with a parliament attached

  • 500 TWh Globally: Current consumption, heading to 1,500 TWh (India's entire usage)

  • 35-50% for AI: Share of data center power going to AI workloads by 2030

The Colonial Compute Model: Virginia already uses 26% of its electricity for data centers. China and Europe, respectively, burn 25% and 15% of global data center power. Every country racing to become an "AI hub" is finding that frontier models fry power grids.

The Bottom Line: We've turned AI infrastructure into the new extractive industry. Rich countries build the models, poor countries provide the power, and local communities get rolling blackouts so ChatGPT can help students cheat on their homework.

💸 Quick Hits: The Infrastructure Finance Revolution

🏦 Meta's $27B Infrastructure Surrender: Meta handed Blue Owl Capital 80% equity in its $27B Hyperion data center, keeping just 20% while becoming a tenant in its own 2GW facility. The company, worth $1.4T, more than Spain's GDP, took $3B cash upfront and signed a 4-year lease because even Zuckerberg can't afford AI infrastructure alone.

☁️ Google's Compute-as-Currency Play: Google is negotiating to give Anthropic "high tens of billions" in TPUs and cloud infrastructure (or proprietary chip access locking them into Google's ecosystem forever). Forget buying companies with cash; Google's buying them with electricity and silicon, making Anthropic dependent while also keeping Amazon happy on the side.

🏗️ CoreWeave's $9B Power Grab: CoreWeave refuses to budge on its $9B all-stock offer for Core Scientific's 1.3GW of data centers, despite shareholders revolting over getting less than 10% of the combined company. When a cloud company built on not owning infrastructure starts panic-buying concrete, you know they need those gigawatts more than happy investors.

⚖️ The Great AI Data Heist Trial: 52 Lawsuits and No Answers

Since October 8, the AI copyright war has hit 52 active lawsuits, but courts have just punted on the core question of whether training on copyrighted data is fair use until summer 2026 at the earliest.

🔍 Reddit's Insane Public Search Lawsuit

Reddit just sued Perplexity and three scrapers for the crime of... using Google search results that contain Reddit content:

  • The Logic Pretzel: "Yes it's public on Google, but scraping it is theft"

  • The Property Claim: Reddit owns its exhaust fumes' exhaust fumes

  • The Precedent: Making public data suddenly proprietary

  • The Real Game: Forcing everyone to pay for Reddit's API

Peak Absurdity: Reddit is literally claiming ownership over Google's index of Reddit. It's like suing someone for reading a newspaper article about your newspaper. We've reached the point where everything is simultaneously public and private, free and licensed, open and closed.

🎵 When Stream-Ripping Meets AI Training

Suno and Udio face fresh lawsuits from indie artists alleging they didn't just train on music, they allegedly pirated it:

  • YouTube Ripping: Using "well-known piracy" tools for training data

  • Lyrics Scraping: From Genius, AZLyrics, and Musixmatch

  • No Deletion Protocols: Keeping pirated libraries indefinitely

  • $150K per Song: Maximum statutory damages per track

  • The Defense: "Stream-ripping isn't technically illegal" (seriously)

The Admission: Their legal strategy is essentially "we didn't rob the bank, we just copied the money." When your defense is that piracy tools are legitimate data collection, you've lost the moral argument.

🦇 Hollywood Deploys the IP Arsenal

Warner Bros, Disney, and Universal are done playing nice, simultaneously suing Midjourney and Chinese firm MiniMax for turning Batman and Mickey Mouse into training data:

  • The Big Three United: Only Sony & Paramount are still watching from the sidelines

  • Two-front War: Domestic (Midjourney) and international (MiniMax) suits

  • Billions at stake: Some of entertainment's most valuable IP

  • Fair Use Defense: "We just learned by watching" (good luck with that)

The Nuclear Option: This isn't exploratory litigation, as Hollywood realized its content is being used to train its own replacements. Thus, the studios are seeking damages, injunctions, and precedent-setting rulings that could either legitimize or obliterate AI training overnight.

🤖 Product Launch Quick Hits: AI Colonizes Everything, Resistance Futile

🌐 OpenAI's Browser Takeover: Launched ChatGPT Atlas web browser on macOS with an "Ask ChatGPT" sidebar on every webpage, plus Agent Mode for paid users that autonomously handles web tasks. Because if you can't beat Chrome, just add ChatGPT and call it innovation.

🔋 Samsung's Two-Hour Reality: Launched the Galaxy XR headset at $1,800 with 4K micro-OLED displays, Snapdragon XR2+ Gen 2 chip, and an external battery pack that lasts just two hours per charge. Sucks that you need to recharge before you finish watching a movie.

⚛️ Google's Quantum Speed Run: Willow quantum chip using the Quantum Echoes algorithm ran computations 13,000x faster than supercomputers for drug and materials research. Finally fast enough to solve problems, we're still trying to figure out how to formulate.

👓 Amazon's Bionic Delivery Drivers: Testing AI-powered AR glasses for delivery drivers with hundreds of North American drivers trying early versions, targeting mass production by mid-2026 for hands-free package scanning, navigation, and proof of delivery.

👀 GM's Eyes-Off Gamble: Rolling out Google Gemini conversational AI in vehicles starting in 2026, then launching an eyes-off autonomous driving system by 2028 in the Cadillac Escalade IQ. Finally, a car that's officially more attentive than its driver.

🎢 Personnel Quick Hits: Musical Chairs of Corporate Chaos

🧠 Meta's Selective Superintelligence Cuts: Meta is slashing ~600 jobs from its Superintelligence Labs, including roles in the decade-old FAIR unit, while completely sparing the newly formed TBD Lab working on next-gen foundation models.

🔇 Sequoia's Free Speech Hierarchy: COO Sumaiya Balbale resigned after five years when partners refused to discipline Shaun Maguire for calling NYC mayoral candidate Zohran Mamdani "an Islamist" and saying his culture "lies about everything."

🌏 Nexperia's Geopolitical CEO Rebellion: Dutch-controlled Nexperia accused ousted CEO Zhang Xuezheng of spreading "falsehoods" about independent Chinese ops and taking "unauthorized actions" after his court-ordered suspension due to national security concerns.

💸 What Other Startups are Popping Off

Here's a roundup of this week's trending startup activity:

🤖 Artificial Intelligence

🛠️ Developer Tools

  • OpenStatus: Open-source status page and uptime monitoring system.

  • Voice Gecko: Voice dictation at your fingertips, type less, say more.

  • Rybbit: Open Source Google Analytics replacement.

  • TinyFast: Fast file size reducer tool for macOS.

  • Kibo UI Patterns: 1,000+ free shadcn/ui examples for use in your project.

💼 Business & Productivity

🎵 Music & Creative

🌟 Editor's Note

At Startup Intros, our mission is to bring the latest founder-investor news straight to your inbox, keeping you ahead in the fast-paced world of Silicon Valley.

💭 Parting Thoughts

This week crystallized AI's impossible trinity: you can have cutting-edge models, sustainable energy, or respect for intellectual property, pick two.

Meta's surrendering equity for electricity, utilities are burning fossil fuels for GPUs, and Hollywood's deploying every lawyer in California because someone trained a model on Batman.

Forward to a friend or hit reply to let me know what you're seeing in your world.

Till next time!

Dev Chandra
Founder & CEO @ Startup Intros
Associate @ Context VC
LinkedIn: /in/devchandra

P.S. Raised this week, and we missed you? Want to be featured? Have tips or funding questions? Reply or DM us as we’re here to help.

Startup Intros Trending Thursday: Your trusted source for founder-investor insights, delivered with clarity and focus.