A practice the tech world dubbed "tokenmaxxing" appears to be over, and a hardware shortage is the reason, according to coverage circulating this week.
The Sydney Morning Herald (SMH.com.au) frames the shift bluntly in its headline: "'Tokenmaxxing' is dead: How the hardware shortage is blowing up AI budgets." The same theme is drawing attention on Hacker News, where an article titled "Tokenmaxxing is dead, long live tokenmaxxing" reached the front page with 123 points and 147 comments at the time of writing.
The core claim across both sources is the same: a shortage of computing hardware is reshaping the economics of building and running AI systems, and it is pushing budgets higher. The phrase "tokenmaxxing" refers to optimizing around tokens, the units of text that AI models process and that companies are billed for. The reporting suggests that whatever strategy that term described is no longer viable under current hardware constraints.
It's worth noting what the sources do not spell out. Neither item provided here details which specific chips are in short supply, by how much budgets are rising, or which companies are most affected. The available material establishes the trend and its framing rather than the underlying numbers.
The Hacker News title's twist, "long live tokenmaxxing," hints that the underlying pressure to squeeze value out of AI spending isn't disappearing, even if one particular approach to it has run its course.
Why it matters: hardware scarcity is quietly becoming one of the biggest cost drivers in AI, which means the price of the AI tools businesses and consumers increasingly rely on may rise alongside it.