AI泡沫迹象：用户支出削减，价格指数腰斩

As tech companies' AI-related spending has ballooned into the hundreds of billions of dollars, one of, if not the big questions, has been about revenue. In other words, when should we expect these investments to finally start paying off? But, while a couple of AI companies have indeed started turning a profit, including perhaps most notably Anthropic, industry-wide data now suggests that users are actually cutting back on spending.

So, in this video, we're going to take a look at the data and explain why it really does not bode well for the industry.

[music]

This spring, we visited Greenland to discover the country and find out why it's been such a major geopolitical flash point. While we were there, we camped on a glacier, trekked through [music] the wild to find abandoned military installations, and explored the melting oceans, all to answer the question, "What's the big deal with Greenland?" The documentary is now streaming exclusively to TDR party members. You can sign up or watch the full trailer by clicking the link in the description.

As we see it, the main reason that AI looks extra bubbly at the moment is this graph, which is something called the LLM token expenditure index. To understand what it's measuring, you need to understand what a token is. Basically, when you talk to an AI, it breaks down your instructions into little chunks of text known as tokens, each usually a couple of characters long. Tokens are essentially the basic unit of text that an AI reads and processes.

So, they're the best way of measuring how much AI you're actually using. The LLM token expenditure index basically measures how much the average user is willing to pay for 1 million tokens worth of AI. Anyway, having essentially doubled between its inception in December and its peak in late May, this index has since fallen by nearly 50%. To be clear, that does not mean that demand for AI and AI-related tools is falling.

On the contrary, basically all available evidence suggests that demand is still rising and pretty quickly. Rather, it means the average price that people are willing to pay for AI is falling. In practice, this basically means that users are becoming more price sensitive and moving towards cheaper models, which are usually a bit less, quote unquote, intelligent than the advanced models, but cost less per token. This would be consistent with headlines we've seen recently about various big companies telling their employees to pair back their AI use.

And reports OpenAI is reportedly considering a price cut, apparently in anticipation of a price war with Anthropic. It would also be consistent with other data suggesting that Chinese models, which are generally cheaper, account for an ever larger share of overall token use. More generally, if people are shunning more advanced models in favor of cheaper, more basic models, this implies that a growing fraction of users see AI as essentially a time-saving convenience, rather than some world-altering tech