Highlights from "The Hater's Guide To The AI Bubble"

This article by Edward Zitron takes a hard look at the AI industry. It offered me a different perspective from the usual tech-focused discussions I read, as it shows the business challenges of AI. I think it’s worth a read if you’re interested in understanding the broader context of this industry that has an impact on everyone and/or if you need a healthy dose of skepticism among all the hype.
It’s a lengthy piece, worth a read, but here’s a quick rundown to get you started should you want to read the whole thing:

• NVIDIA is the bubble’s weakpoint: the article identifies NVIDIA as a critical weak point in the AI bubble. The industry’s heavy reliance on NVIDIA’s GPUs means that any slowdown in NVIDIA’s growth could have significant consequences. Zitron highlights that “35% of the US stock market is held up by five or six companies buying GPUs”
• The balance between revenue vs. capital expenditures is broken for all the big players: major tech companies are pouring billions into AI infrastructure, but the returns are not justifying these massive expenditures. The article points out that “Meta, Amazon, Microsoft, Google and Tesla will have spent over $560 billion in capital expenditures on AI in the last two years, all to make around$35 billion,” illustrating the imbalance between spending and revenue.
• Companies built on top of LLMs don’t make money: many companies built on LLMs are struggling financially. The article notes that “outside of OpenAI, Anthropic and a few others, there are no LLM companies making more than $500 million in annualized revenue.” Despite the hype, AI adoption remains low, and these companies often tweak financial reports to appear more successful.
• OpenAI and Anthropic are their customers’ weak point: companies relying on OpenAI and Anthropic are vulnerable to changes in pricing or terms of service. The article explains that “these are now the new terms of doing business with these companies: a shakedown, where you pay up for priority access or face indeterminate delays or rate limits.” This dependency is very likely unsustainable and poses a risk to businesses built on these models.
• LLMs cannot do everything: while LLMs have certain capabilities, their limitations are often understated. Zitron criticizes the use of the term “agent,” stating that it is “one of the most egregious acts of fraud,” as it suggests a level of autonomy that LLMs do not possess. The article argues that the practical utility of AI is limited, and the term “agent” is misleading.