[REVISION HISTORY]

AI pricing, token economy, and intensifying safety debates

Updated 21 times since CLSTR started tracking revisions of this situation.

What changed

2026-09-12 16:26 UTC → 2026-09-12 20:41 UTC · added ~~removed~~

The AI industry is experiencing significant pricing volatility and a phenomenon described by Andreessen Horowitz as ‘LLMflation,’ characterized by a roughly 10-fold annual price decline for models of fixed capability. While input token costs are falling, output tokens remain significantly more expensive. This deflationary trend is driven by intense market competition, improved architectural efficiency, and the rise of cheaper open-source and Chinese models. While lower costs benefit end-users, analysts note this trend threatens the profit margins of foundation model developers, compressing revenue while compute commitments remain fixed. By Q2 2026, a divergence emerged between infrastructure growth and operational costs. Global semiconductor equipment billings rose 23% year-over-year to $40.53 billion, marking a second consecutive record quarter. Conversely, token prices for frontier models have hit record lows. Despite these falling unit prices, enterprise AI budgets are frequently exceeding projections. This is attributed to complex factors such as reasoning models generating hidden thinking tokens, agentic workflows triggering cascading calls, and a lack of unified visibility across vendors. To mitigate this, companies like Salesforce are exploring more predictable, outcome-based pricing models. Furthermore, the industry is undergoing a fundamental shift from a focus on raw computing power (FLOPS) toward a ‘Token economy,’ where tokens serve as the primary unit of value. In China, this is manifesting through the development of ‘Token factories’ in cities like Beijing, Tianjin, and Wuxi, as enterprise-level MaaS token calls are projected to reach 400 trillion by 2026. However, structural risks are mounting; approximately 95% of organizations have yet to see measurable financial returns from generative AI, creating a risk that capital expenditure could outpace revenue, potentially leading to a market correction. In September 2026, the release of OpenAI’s GPT-6 Astra model triggered volatility in the software sector. ~~In mid-September 2026,~~ Simultaneously, safety and regulatory concerns have intensified.

Versions

Only revisions since CLSTR began indexing content versions appear here. Select a version to see what changed compared to the one before it.