What happens next, after the decline of tokenmaxxing?

2025-06-07 · Source: Marcus on AI · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Fundamental Awareness, quick

Summary

Converging evidence suggests a decline in "tokenmaxxing," a trend characterized by the rapid expansion of large language models. This decline is supported by several indicators: the rental price for an Nvidia H200 GPU collapsed by 40% from \$7/hr to \$4/hr in three weeks, and a Financial Times analysis projected that only one hyperscaler's AI investment would yield positive returns between 2025 and 2030. Additionally, Amazon reportedly scrapped an internal AI leaderboard to discourage workers from solely pursuing usage scores, and Fortune's Jeremy Kahn also observed this shift. Following this apparent decline, the article presents two contrasting predictions for the future: the author's more cautious outlook and AI researcher Lisan al Gaib's significantly more optimistic forecast, which they plan to re-evaluate in 12 months.

Key takeaway

For Directors of AI/ML evaluating infrastructure costs, the 40% drop in Nvidia H200 rental prices signals a potential market correction or increased efficiency. You should reassess your current cloud GPU expenditure and consider optimizing model sizes or inference strategies. If you are an investor, scrutinize AI project ROI projections, especially given the Financial Times' finding that most hyperscaler AI investments may not clear positive returns by 2030. Align internal metrics to avoid unintended "usage score" chasing, as Amazon experienced.

Key insights

The AI industry faces a potential shift from "tokenmaxxing" due to declining GPU costs and questionable ROI, prompting diverse future predictions.

Principles

• Declining GPU costs can signal market saturation or efficiency gains.
• ROI analysis is crucial for long-term AI investment sustainability.
• Internal metrics can inadvertently drive undesirable behaviors.

In practice

• Monitor GPU rental prices as a market health indicator.
• Re-evaluate AI project ROI assumptions regularly.
• Design internal metrics to align with strategic goals.

Topics

• AI Market Trends
• GPU Pricing
• AI Investment ROI
• Hyperscaler AI
• Tokenmaxxing
• AI Strategy

Best for: VP of Engineering/Data, Executive, Entrepreneur, Director of AI/ML, CTO, Investor

Related on AIssential

• The bubble is slowly popping, investment isn't able to keep up — The AI market faces a critical debate regarding its investment sustainability and the long-term viability of current pricing models.
• Breaking: bad news for three of the biggest IPOs in history — Excessive GenAI usage without clear ROI is unsustainable, leading to significant financial losses for major tech companies.
• Nvidia pumps over 40 billion dollars into AI partners so far in 2026 — Nvidia's $40B+ AI investments in 2025 position it as a key industry backer.
• Sam Altman and the day Nvidia’s meteoric rise came to an end — The "scaling = AGI" myth's collapse led to a tech stock plateau, revealing LLMs' commodity nature.
• AI boom: Big Tech capital expenditures now seen topping $1 trillion in 2027 - CNBC — Hyperscalers are dramatically increasing AI capital expenditures, driving market growth despite investor skepticism over immediate ROI.
• Tokenomics Is Dead. The Lie Of Affordable Inference — Current "cheap inference" is a VC and hyperscaler-subsidized illusion, masking significant AI industry losses.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Marcus on AI.