For years, AI companies have been selling investors on the idea that scaling up operations is the key to success — the bigger the AI model, the more powerful it becomes. But in practice, that premise is leading to enormous business problems. As The Atlantic points out, LLMs are suffering from severely diminishing returns: the cost of inference, or the act of using a trained AI model to process new data, is
[rising exponentially](https://epoch.ai/publications/inference-economics-of-language-models), making the tech
[far less lucrative than it was](https://futurism.com/future-society/palo-alto-ceo-ai-arora-automation-labor)even half a year ago.
Put simply, it’s effectively the opposite of what investors would conventionally want to see, as The Atlantic argues, which would be a continuous drop in cost per user instead of the reverse.
That’s bad news for an industry pouring billions of dollars into the construction of enormous data centers across the country, despite having no clear path to profitability in the foreseeable future. While it’s impossible to predict when exactly fears of an AI bubble will hit a breaking point, analysts warn it’s a matter of when, not if. The consequences could be disastrous if the industry were to collapse in on itself, bringing down entire economies — which have vastly over-indexed on AI tech — with it.
Nonetheless, AI companies are steadfast in their belief that chatbots, like OpenAI’s ChatGPT and Anthropic’s Claude, are the future. They’ve become practically inescapable in our day-to-day lives, from integrations in operating systems to replacing humans on customer service calls — despite a major public backlash that represents another headwind for the industry.
Meeting AI’s rapidly growing demands will also require revolutionary leaps in hardware tech that are becoming increasingly harder to come by, with experts warning of the imminent end of Moore’s Law, the decades-old truism that more transistors steadily improved computing power.
In short, as behemoth AI models are quickly becoming unaffordable to those who’ve grown to rely on them, they’re also becoming far less efficient — a concerning trajectory that could nudge the entire economy closer to the brink.
For a while now, Yann LeCun, one of the so-called godfathers of AI, has been warning that LLMs are a “dead end” and won’t bring us to the much-touted point where AI intelligence matches or exceeds that of human intelligence. But whether tech leaders who are already deeply invested in scaling up the current crop of AI models will heed those warnings feels increasingly unlikely.
More on AI: AI Bubble Fears Are Starting to Spill Over