AI token prices have reached historic lows, driven by increased competition and falling production costs. The rise of Chinese open-source AI models and strategic price cuts by major providers have created a deflationary trend in the market. While this benefits end users by reducing costs, it poses a significant challenge to companies like OpenAI and Anthropic, which are preparing for initial public offerings (IPOs) and must now demonstrate profitability. The promise of massive profits from AI is diminishing, and investors are closely watching how this shift could impact returns on investments in global digital infrastructure. In the field of large language models (LLMs), a "token" is a unit of text processed by the AI—this can be a word, part of a word, a punctuation mark, or even a space. On average, one token corresponds to about four characters in English, but in languages like French, where accents and conjugations are more complex, the efficiency is lower. Just a few years ago, the cost of processing a million tokens could reach tens of dollars. Today, even top-tier models can be used for less than a dollar per million tokens. This dramatic decline is due to architectural improvements, increased competition, and advancements in AI hardware. According to the LLM Token Expenditure Index, which tracks daily AI token prices, the cost of a million tokens dropped to 97 cents on September 1, 2026—the lowest since the index's creation in late 2025. This represents a more than 50% drop from the peak recorded in the summer of 2026. The index combines pricing from various AI models with the actual usage data observed on routing gateways, making it a key indicator of market trends. The drop has been driven by the rapid spread of low-cost Chinese open-source models, such as Moonshot AI’s Kimi K3, and price cuts by OpenAI on two of its GPT-5.6 models. Additionally, the overall decrease in production costs across the industry has contributed to this downward trend. This deflationary pressure has raised concerns for AI startups like OpenAI and Anthropic, which are under pressure to prove their business models’ viability. As consumers become accustomed to lower prices, model providers risk losing their pricing power permanently. Charles-Henry Monchau of Syz Group notes that companies must shift their competitive advantage from raw model capacity to areas like distribution and memory. Meanwhile, financial markets have reacted to the trend, with major stock indices like the Nasdaq Composite and S&P 500 showing declines following the release of the token price data. This uncertainty has also affected the valuations of key players like SpaceX, whose stock has fallen below its IPO price. The financial stakes are high for companies like Oracle, which has massive debt and lease commitments tied to its $300 billion partnership with OpenAI. Analysts are questioning whether Oracle can justify these investments without clear evidence of AI monetization. The broader AI industry faces a critical juncture as the 2027-2028 deadline approaches, when many AI contracts are set to come into effect. If these commitments fail to materialize, it could trigger a prolonged investment crisis. With the U.S. AI sector facing growing competition from Chinese open-source models, the pressure is mounting on companies to deliver on the promises that have fueled their valuations.