Subject: AI industry numbers For: #ai and @stux Image: SMBC comic about AI startups Short link: https://dansu.org/ai-industry-numbers stux said: It’s all just big tech buying datacenters and hardware from each other while investing in said loop. It’s a disastrous circus 🤡 It’s a bit worse than disastrous. Apocalyptic is a better term for the likely outcome. stux posted an image that listed the following 2025 #Anthropic figures: $4B in revenue, $40B in annual loss, $500B in obligations, and $2T valuation. To begin with, let’s revise some of the numbers. The list stated above is misleading and/or incorrect. The 2025 $4B revenue figure for Anthropic is expected to be 6 times larger for 2026. Not all of it will be from the financial loops. Many corporations are rushing to deploy #Claude Code and other tools so as not to be “left behind”. Additionally, Anthropic’s valuation as of the end of 2025 was far less than $2T. The valuation is still less than $1T today. [Though not by much.] Finally, the $500B “obligations” figure is too high for Anthropic and too low for the industry as a whole. For the 5 largest hyperscalers by themselves, total obligations are believed to be well over $1T right now. That is the good news for Anthropic. The bad news for the industry as a whole is as follows. If you’re a fan of disaster movies, this is the moment to get out the popcorn. 🍿 Add up Capital, Operational, and Debt Repayment expenses for the industry. Free Cash Flow needed to cover the total is expected to be about $750B per year short before 2030. Large piles of money will be blazing up to the sky. In short, AI investors are hoping for 💰 🌊 but they’re going to get 💰🔥. The AI industry will use accounting tricks to report profits that aren’t real for as long as possible. However, in the end, it’ll be the landscape depicted in “Mad Max 2: The Road Warrior” (1981). Crazed mutant VC people running here and there looking for the last drops of their funds. It’s incomprehensible that the business level of an entire country could be so foolish. And that is before we get to the matter of a President who wants to rebrand “AI” as “SI” so as to promote himself as the savior of the country because a new and more wonderful name will be used. I fall between the perspectives on AI that I see in the Fediverse [on average, rage] and at LinkedIn [on average, positive with pockets of skepticism]. My position is simple. In one sentence, #LLM and #GenAI can be useful tools, but the industry operates in a destructive, impractical, and illegal manner. There is a related point that that I don’t see discussed enough. In exchanges that you have with excited AI people, point out the following: The LLM approach to AI, in particular, was selected based on, not what would work well, but on working backwards from the goal of the easiest path to more money for billionaires. 💵 💰 The approach in question breaks down into, simply stated, steal the entire Web and every book ever published 🏴‍☠️🦜 dump everything into a blender, grind it into a smoothie of 4 TB in size, and wait for intelligence to show up magically as an emergent property of language. It’s the other way around! 🙄 AI is Fake, For Goodness Sake. See my essay of that title at LinkedIn: https://dansu.org/ai I won’t apologize for saying that #LLM and #GenAI can be useful or for using them in small ways. But I’ll note that the practical, ethical, and legal issues are collectively intractable. The industry does need to be torn down and rebuilt. 👷🚧 I think that that is likely to happen as a natural consequence starting in at most 5 years. The first steps towards the collapse could start as early as 2029. To see a larger version of the image, in most browsers, right-click and select “Open Image in New Tab”. @mastodonmigration @stux