This is the cleanest chart I've seen for understanding the AI infrastructure trade Memory prices UP, token prices DOWN If cheaper AI meant less demand for hardware, these lines would move together. They're doing the exact opposite. Here's why: A token getting cheaper doesn't mean the task got smaller. Answering a question, running an agent, writing code... it takes roughly the same number of tokens as it did back in May when the price was double Same tokens = same compute and memory burned per task If anything it goes the other way. When tokens cost $2 per million, engineers optimized every prompt. At 97 cents, nobody bothers. Cheap tokens make people more lazy with tokens (I've watched it happen) So the per-task hardware bill doesn't fall when token prices fall, what changes is the NUMBER of tasks Every leg down in token prices unlocks use cases that were unprofitable before. More agents running, more workflows automated, more code generated. Total token consumption keeps climbing even as the unit price collapses And all of those new tokens need memory and compute underneath them. DDR5 spot prices are up over 50% this year and analysts don't see supply relief before late 2027 (Intel's CEO says the industry told him 2028) The market keeps reading falling token prices as bearish for AI. This chart is telling you that the cheaper intelligence gets, the more of it we consume, and the more hardware we need to serve it Cheaper output, scarcer input. You want to own the scarce input That's a big part of why I hold $MU in my Milk Road PRO portfolio. You can track 5 analysts' real time portfolios and research with live trade notifications inside and it's just $1 to try it out (insane price just to check it out). Learn more here: Don't forget to give me a follow @kylereidhead for more insights on AI and markets