Of every $100 spent running an AI data center, $60 covers GPUs losing value. Electricity is $7. The real bill isn't power, it's depreciation.

# Of every $100 spent running an AI data center, $60 covers GPUs losing value. Electricity is $7. The real bill isn't power, it's depreciation. **Channel:** Ill_Vegetable169 Source: [https://www.youtube.com/watch?v=qcg_8Li2XKA](https://www.youtube.com/watch?v=qcg_8Li2XKA) ## Description Where does Big Tech's AI money actually go? Meta, Google, Microsoft, Amazon and Oracle will spend almost $800 billion on AI this year. We follow the money into Nvidia GPUs, AI data centers, power and cooling, and ask whether the AI boom can pay off. In 2022, ChatGPT reached 100 million users in two months and Google declared a "code red". Since then, every Big Tech company has been racing to build the computers that train and run large language models (LLMs). Mark Zuckerberg said wasting "a couple of hundred billion dollars" would be very unfortunate. Then he kept spending. This animated explainer uses one simple picture, an automated cake factory, to show how the AI industry really works: electricity is the heat, Nvidia chips are the ovens, data centers are the factories, AI models from OpenAI and Anthropic are the chefs, and apps like Cursor and Lovable are the decorators. We break down: • Why ChatGPT made Google, Microsoft, Amazon and Meta afraid for their core business • AI training vs inference, and why every question costs money • What one Nvidia chip and one AI rack really cost • Why a gigawatt AI data center costs $50 to $60 billion • Where $100 of AI data center spending goes: chips, network, power, cooling, building, land • DeepSeek, Nvidia's $600 billion day, and the Jevons paradox • Why GPU depreciation, not electricity, is the real bill • Debt, circular deals and the $1.5 trillion question • Open models like Qwen, on-device AI at Apple, and OpenAI's $1.4 trillion of orders • What the 2000 fiber-optic bubble teaches us about the AI bubble Chapters 0:00 Zuckerberg's "very unfortunate" bet 0:25 ChatGPT and Google's code red 1:08 What each company is afraid to lose 1:28 Everyone runs to Nvidia 1:41 The AI cake factory: chips, data centers, models, apps 3:15 Inside one Nvidia chip and one AI rack 3:57 A gigawatt AI data center 4:57 Where $100 of AI spending goes 5:39 DeepSeek and Nvidia's $600 billion day 6:21 The Jevons paradox and the price war 7:13 Pichai: "elements of irrationality" 7:27 Two clocks: power and better chips 8:29 Almost $800 billion this year, and the debt 9:12 Circular money and the $1.5 trillion question 9:26 Three things that could break the bet 10:17 The fiber bubble of 2000 10:48 Bigger every wave? 11:08 So, was it unfortunate? The real question is not whether AI is the future. It is whether people will keep asking AI for more, whether someone will pay enough for each answer, and whether the chips earn back what they cost before a better one arrives. Is this an AI bubble, or was it right not to wait? Tell me in the comments. #AI #LLM #Nvidia #ChatGPT #OpenAI #DataCenters #AIBubble #BigTech