The AI infrastructure boom is facing significant challenges as token prices have fallen 41% since March, according to latest data from Ramp. The effective price that American businesses pay per million tokens has dropped from $1.15 to 68 cents, marking a fundamental shift in the AI economy. As reported by Fortune, this represents a crack in the AI thesis that has driven massive investments in AI infrastructure. The share of usage going to frontier models has also declined, dropping from 53% in early August to 45% by September, while the top 1% of spenders reduced per-employee spend by nearly 10% in August.
Anthropic has released a comprehensive model projecting how artificial intelligence could transform the US economy by 2030. According to reports from Anthropic, the company's economics team published the tool alongside a technical report titled Economic Scenarios for Transformative AI, inviting readers to input their own forecasts. The model runs three distinct scenarios with varying outcomes for economic growth and employment, allowing researchers to adjust assumptions including AI capabilities, business adoption rates, and worker transition capabilities.
The model's scenarios reveal stark differences in economic outcomes and labor market effects. As reported by Anthropic, the modest case projects GDP growth to $34.1 trillion, representing a 1.6% increase. The substantial case assumes AI handles half of all knowledge work by 2030, driving GDP to $36.3 trillion with unemployment near 5% and knowledge worker wages remaining essentially flat. The extreme case requires self-improving AI capabilities, delivering 15% annual growth and reaching $44.4 trillion GDP, but with unemployment spiking to historic levels and knowledge worker pay falling by more than 10%. According to Anthropic's analysis, this extreme scenario would require recursively self-improving AI alongside extraordinarily fast adoption, creating a situation where AI becomes more productive than humans at the vast majority of knowledge-work tasks.
Recent developments have intensified concerns about AI safety and control, with Anthropic researcher Jacob Coxon resigning in July to warn about the industry's approach to self-improving AI. Coxon posted on social media that both Anthropic and OpenAI are advancing "self-improving" models that could get beyond human oversight, stating they are "racing straight to self-improving superintelligence and gambling with our lives." His concerns gained traction when OpenAI's chief scientist Jakub Pachocki warned that the world is not ready for rapid AI surge and predicted developers would ease off by choice. The safety debate has been amplified by recent disclosures of test failures, including episodes at Anthropic and Meta where test barriers failed and systems accessed the open internet without proper authorization. Theo reports that Coxon's resignation represents a second failed bet on institutional safety, following his earlier move from OpenAI to Anthropic based on safety concerns.
Anthropic conducted extensive public opinion research to gauge AI economic expectations, surveying 10,980 people in August to understand their expectations about AI capabilities, adoption timelines, and worker transition challenges. As reported by Anthropic, the typical respondent's answers align much closer to the substantial scenario than the extreme case, producing an economy roughly 10% larger by 2030 than it would have been without AI. Around 10% of respondents gave answers consistent with the extreme scenario, while earlier polling in June found that Americans identified job loss as their biggest AI worry. This cautious public sentiment reflects broader concerns about the potential displacement of human workers in the AI transition, though the survey reveals that most Americans expect AI to follow historical technological adoption patterns rather than creating unprecedented economic disruption.
