Macro strategist Andreas Steno Larsen is sounding the alarm on a worrying trend: the Silicon Data LLM Token Spending Index is heading south. He argues that the price of LLM tokens is the single most important signal for the entire AI infrastructure cycle. If token pricing keeps dropping, the current investment wave — from memory chips to data centers — could be over.
https://twitter.com/AndreasSteno/status/2064043229863903516
So what is this index? It's a spending-weighted metric that tracks the average price per million tokens across the market. Since most major AI providers charge based on token consumption, the index effectively links software usage to the underlying hardware: GPU compute, DRAM memory, and data center construction. It's become a key barometer for AI capex cycles.
The data shows the index surged in early 2026, then peaked at the end of May and started falling. And that decline is closely tied to a shift in corporate attitudes toward AI spending. Early on, cost controls were virtually nonexistent — some tech giants even saw internal "Tokenmaxxing" contests, where employees burned through tokens to boost their productivity rankings. But then the bills arrived, and companies realized they weren't getting proportional returns. Now, Microsoft and Amazon are tightening internal AI tool budgets and even killing some projects. The pivot from "throw everything at it" to "show me the money" has directly stalled and softened the token spending index.