(Bloomberg) — At a time when markets are increasingly anxious about whether the vast sums being poured into artificial intelligence will ever pay off, the prices the sector charges for each unit of use are falling.
Most read by Bloomberg
The Silicon Data LLM Token Expenditure Index, which tracks what users pay for AI tokens, has fallen nearly 20% since its peak in May, having nearly doubled since its launch in December. The benchmark is the cleanest value available when it comes to the $700 billion-plus investment boom that has done the sector’s heavy lifting.
For stock investors, this could be a warning that AI companies are losing their pricing power with increasingly cost-sensitive customers and that expectations of a potential AI bonanza may prove misplaced.
“Reports are increasing that users of AI solutions priced in tokens are having to limit unlimited use due to the high cost,” said veteran investor Louis Navellier. “Talk of OpenAI delaying its IPO until next year is seen as a sign that profitability remains an issue at this time.”
Just to be clear, a softer index does not mean AI will be cheaper. The meter combines prices and usage, meaning a decline could mean very different scenarios: either list prices fall or demand shifts toward cheaper models. It could also indicate a real slowdown in buyers’ willingness to buy.
Each of these options has different effects. Silicon Data, which created the index, has warned people not to read it as a price tag. The company calls it a proxy for marginal willingness to pay.
Let’s start with an innocuous read: While token prices have plunged more than 90% since 2023, total spending has roughly doubled since last year. Cheaper tokens have expanded the market. This means that an index pause is merely digestion, while demand is real and capital spending is money well spent. Here lies the bull case for Nvidia Corp., memory maker and data center name.
Now to the interpretation that’s keeping people awake: bears warn that continued weakness in the index could end the trade that has driven a strong rally for almost the entire AI cohort this cycle.
These are symbolic expenditures that justify the next investment order, and the bill already looks excessive. According to Allianz Research, there is a nearly 46% growth gap between AI investments and sales. That’s worse than the 32% divergence measured during the 2001 telecom crisis.
The story continues
Fortunately for the bulls, the downtrend has paused. After a weak week, it’s still too early to call a bottom, but it’s enough to keep the case for a recovery alive.
“During the training phase, the costs of AI infrastructure and token generation are exceptionally high, but in the current inference phase the economics are significantly better,” said David Miller, senior portfolio manager at Catalyst Funds. “The net use of AI brings companies a positive return on investment, at least in the long term.”
There is also a newer, demand-side reason why the bearish trend could continue. Washington has a newfound willingness to exert control over a key industry. The US government just this week lifted foreign access restrictions on Anthropic PBC’s Fable 5 model, days after regulators asked OpenAI to delay the launch of an upcoming version.
Meanwhile, the European Union’s AI law targets boundary models for mandatory assessments and strict transparency requirements. None of these measures directly cap prices, but they create a deployment and compliance burden for top platforms that do not support fewer but still useful systems. This consideration may provide corporate finance chiefs with a rational reason to shift workloads to cheaper models.
Of course, this is not a case of chip flooding. High-end GPUs and high-bandwidth memory are sold out until 2026, and there won’t be any real relief until 2028. The hardware explanation is more subtle in that it suggests a shift in the demand mix away from premium training GPUs and toward inference-optimized parts. This changes the composition of the winners, but it doesn’t make you feel bad.
Nevertheless, the “unbridled” market enthusiasm, increasing competition from China and price sensitivity mean that DWS strategists around Chief Investment Officer Vincenzo Vedda are cautious. “We are monitoring areas where valuations may appear stretched,” they said.
The conclusion is that the token chart cuts in both directions and one should hold both reads at the same time. If the late June flattening continues and the dip was just a mix shift digestion, cheaper tokens will continue to expand the markets, meaning capex remains justified and the bull case remains intact.
However, if this is the point at which customers’ willingness to pay peaks while regulatory headwinds drive down demand, then the most expensive part of the trade will also be the first to collapse. That’s because it’s a pricing power story, not a silicon story, funding the march toward $1 trillion in capital spending in 2027.
– With support from Amy Thomson.
Most read by Bloomberg Businessweek
©2026 Bloomberg LP
https://finance.yahoo.com/technology/ai/articles/ai-trade-losing-one-key-061343277.html
