AI Token Prices Fall Below $1, Pressuring Frontier Labs

Token pricing hits historic threshold

Artificial intelligence token prices have crossed a symbolic barrier that threatens the revenue models of leading AI companies. The LLM Token Expenditure Index, tracked by Silicon Data and listed on Bloomberg as SDLLMTK, fell to 97 cents per million tokens on Monday—the first time the benchmark has dropped below $1 since its launch late last year, according to CNBC.

The index now sits at less than half its summer peak and declined 8.6% in the seven days leading up to Monday's close. Silicon Data's measure tracks the usage-weighted effective price across a defined set of large language models, combining provider pricing with actual consumption volumes observed through multi-provider routing gateways.

Multiple forces driving prices down

Several factors are converging to push token costs lower. The emergence of lower-cost open-source models from Chinese developers—including Moonshot's Kimi K3—has created downward pricing pressure across the market. OpenAI contributed to the decline by cutting prices on two GPT-5.6 models in late July.

Other frontier laboratories have introduced dynamic pricing structures that allow rates to fluctuate with demand, adding further pressure. Meanwhile, the underlying costs to produce tokens have also fallen, compounding the deflationary trend.

Why it matters

The pricing collapse arrives at a particularly sensitive moment for frontier AI companies. Both OpenAI and Anthropic filed confidential IPO paperwork with regulators this summer, and sustained token deflation could undermine their valuations. When consumers become accustomed to cheaper access, providers lose pricing power—a dynamic that makes it harder to justify the massive compute investments these companies have made.

"Foundation model labs are the most directly exposed," Charles-Henry Monchau, investing chief at Syz Group, wrote in a Tuesday analysis. "Token deflation compresses the revenue line while compute commitments stay fixed."

The competitive landscape is already shifting in response. Companies are pivoting away from competing solely on model performance—where open-weight alternatives have narrowed the gap to just months—and instead emphasizing advantages in distribution, memory capabilities, and context handling.

Market implications

Steve Hou, Silicon Data's head of research, suggested the price decline may indicate that existing supply across frontier and lower-cost models is already adequate for most use cases. That interpretation would signal a maturing market where differentiation becomes harder to monetize.

Broader technology equities felt pressure on Tuesday, with the Nasdaq Composite losing nearly 1% and the S&P 500 declining 0.4%.

These details were first reported by CNBC.