Renowned macro strategist Andreas Steno Larsen recently dropped a serious warning on social platform X (formerly Twitter), urging global investors to keep a close eye on a critical indicator—the 'Silicon Data LLM Token Expenditure Index.' After several months of strong growth, this index showed a significant downturn for the first time from late May to early June 2026.

This is the chart that everyone should be watching. If the Token Pricing rolls over, everything from the memory trade to the broader hardware and data center trade is over for this cycle, in my humble opinion. The whole setup depends on this.. pic.twitter.com/qB0mRj0xJr

— Andreas Steno Larsen (@AndreasSteno) June 8, 2026

Larsen bluntly stated: "This is the chart that everyone should keep an eye on right now. If the Token pricing starts to roll over, in my view, the transition from memory trading to broader hardware and data center trading during this cycle could very well signal the end."

Warning signs of a downward index

"Token" is the basic unit of measurement for processing text and computing power in large language models (LLMs). The "Token Spending Index" referenced by Larsen essentially reflects the "actual total expenditure" of global companies and developers on AI software applications (i.e., Token consumption × Token pricing).

When the index curve shows a clear downward trend at the far right, the reasons can only be attributed to "weakening demand" and "software price cuts." Weakening demand's negative impact on the industry is self-evident, and while "software price cuts on Token pricing" can theoretically stimulate greater usage demand, in the current environment, the demand stimulated by price cuts is often low-tier demand with diminishing marginal returns (such as entertainment generation, repetitive scraping), which have poor monetization capabilities and contribute minimally to the profitability of model companies. Therefore, if this curve continues to decline, it is undoubtedly a warning for the entire industry.

Demand for hardware equipment may slow down

Another key point to watch is that the marginal cost of software (Token) is extremely low, allowing vendors to bleed market share; however, the underlying hardware (high-bandwidth memory, advanced process chips, cooling systems, electrical grids, and data center construction) is a tangible physical entity, and its production and operating cost reductions are very limited. A decline in the software Token Spending Index may indicate that the demand for hardware from AI model companies will also begin to slow down.

Currently, the extremely high valuations and capacity expansions of hardware giants like Micron, Samsung, and NVIDIA are entirely based on the assumption that "the software application side will be able to generate massive profits and continue to buy hardware." Once model companies realize that "the new revenue stimulated by price cuts cannot even cover the electricity bills and chip installment payments for new data centers," the software layer will be forced to halt expansion and cut capital expenditures. This will lead to a slowdown in growth for semiconductor and chip manufacturers.

Are we facing a periodic cooldown?

Larsen's warning does not deny the long-term future of AI, but rather points out that "the short-term cyclical bubble has already peaked." When the market views hardware revenue (a lagging indicator) as bullish, the actual consumption amount of upstream software computing power (a leading indicator) has already hit the brakes.

If Token pricing continues to roll over and the software layer fails to achieve a commercial closed loop, this winter will quickly reverse transmit to memory trading and the data center supply chain. In simple terms, Larsen's tweet aims to warn global investors to stay vigilant; this AI party driven by hardware may soon face a valuation correction and a cooling period in demand.

Source