As the artificial intelligence (AI) industry grows and related stocks move sharply, investors' attention is shifting to more precise demand gauges.
Until now, indicators such as Nvidia's GPU sales and the scale of data center construction by global big tech companies have been used as key metrics. But now, alternative data that shows in real time how much AI is actually being used is emerging as a new investment guide.
A new leading indicator in the AI era, the "Silicon Data LLM Token Expenditure Index" (SDLLMTK), is drawing attention. The index estimates actual usage of large language models (LLMs) by their token consumption and calculates it as a metric.
SDLLMTK is designed around tokens because most Generative AI services on the market charge based on "tokens," the basic units of letters or words. Therefore, token usage is the most intuitive and direct gauge of the scale of AI service usage.
The index goes beyond simple usage to measure "token expenditure" by reflecting each model's price in token usage. Expressed as a formula, it is "Σ(token usage × price per token)."
Thanks to this calculation method, the index reflects not only usage but also which models are used more. For example, if the share of usage shifts toward newer premium AI models with relatively higher unit prices, the index rises more even if the total number of tokens consumed remains the same. It captures both quantity (Q) and unit price (P) of revenue.
◇ GPU sales and power consumption are lagging indicators… tokens are "front-line demand"
Researcher Kang Hyun-gi at DB Securities said SDLLMTK is ahead of existing AI-related indicators in terms of timeliness.
Looking at the AI industry's value chain, when AI users consume tokens while using services, AI models run GPUs to perform inference. In this process, AI data centers linked to the cloud operate and large amounts of power are used.
Indicators that the market has focused on—Nvidia's GPU sales, infrastructure investment by big tech (hyperscalers), and data centers' power consumption—are all closer to "lagging indicators" that reflect the back end of the AI industry. In contrast, token usage serves as a "leading indicator" that most quickly captures actual demand changes at the industry's front line.
For stock investors, movements in SDLLMTK can inform strategies to time and size positions in AI-related stocks.
Kang noted that a continued rise in the index means actual front-end demand for AI is increasing. In that case, it is advisable to expand exposure to AI-related stocks or step up investments. Conversely, if the index turns down or slows, investors should allow for the possibility of demand stagnation and move to manage risk.
Kang said, "It is risky to put blind faith in a single indicator, so existing metrics such as GPU sales and infrastructure investment should be examined together," but added, "Given that it can identify substantive changes in AI demand at an early stage, a strategic approach that assigns relatively higher weight to SDLLMTK as a leading indicator is needed in future investment decisions."