AI Token Prices Hit Record Lows as Competition Drives Down the Cost of Intelligence

The cost of using artificial intelligence is falling rapidly, potentially changing the economics of building AI into everyday business applications.

A closely watched measure of AI token prices fell below $1 per million tokens for the first time this week. Silicon Data’s LLM Token Expenditure Index reached $0.97 on Monday, its lowest level since the index was launched late last year and less than half its early-summer peak.

Tokens are the units of text that large language models process and generate, and they form the basis of pricing for many commercial AI services. Falling token prices therefore mean that businesses can potentially process more information and run more AI interactions for the same budget.

For companies building AI products, this could significantly reduce one of the barriers to scaling.

Competition Is Pushing AI Prices Down

Several factors are contributing to the decline.

Competition between AI providers continues to intensify, particularly as lower-cost and open-weight models improve. Chinese AI developers have also increased pricing pressure by releasing capable models at significantly lower costs than some frontier alternatives.

At the same time, major AI companies have been reducing their own prices. OpenAI cut prices for two GPT-5.6 models in July, while other providers have introduced different pricing mechanisms to make model usage more competitive.

The underlying cost of producing tokens is also falling as models and infrastructure become more efficient.

For AI users, the result is straightforward: access to increasingly capable models is becoming cheaper.

For the companies developing those models, however, the economics are more complicated.

Lower prices reduce revenue per unit of AI usage, while infrastructure commitments remain expensive. This could put pressure on companies such as OpenAI and Anthropic as investors increasingly examine whether enormous spending on chips, data centers, and computing capacity can generate sustainable returns.

From Expensive Experiments to Everyday Infrastructure

For businesses outside the foundation-model industry, falling token prices create a different opportunity.

Until recently, the cost of AI could limit how extensively companies deployed it. A small experiment involving a few thousand documents or customer conversations is one thing. Running AI continuously across millions of products, searches, translations, or customer interactions is another.

As inference becomes cheaper, the threshold changes.

E-commerce companies can potentially use AI more extensively for product classification, content generation, translation, search, recommendations, customer support, and catalog management.

Instead of reserving advanced models for a small number of high-value tasks, businesses can apply them across larger datasets and more routine processes.

For e-commerce in particular, this matters because scale is fundamental. Large retailers and marketplaces may manage millions of products, while product information must be continuously updated, translated, categorized, verified, and distributed.

Lower AI costs make automation across those catalogs more economically realistic.

Cheaper AI Makes Data More Important

There is another consequence of falling token prices.

If access to powerful models becomes inexpensive and widely available, simply having access to an advanced LLM becomes less of a competitive advantage.

The differentiator increasingly becomes what businesses give those models to work with.

AI systems still need accurate information, context, and domain-specific data. A model asked to classify a product needs reliable product characteristics. A shopping assistant recommending a laptop needs specifications it can interpret. An AI system generating product descriptions needs a trustworthy source from which to ground those descriptions.

Structured product information provides AI systems with context about products, including specifications, identifiers, categories, descriptions, images, and other attributes. AI can then help transform, translate, classify, enrich, or retrieve that information more efficiently.

Lower token costs could make these workflows practical at a much larger scale.

Instead of asking whether it is economical to apply AI to an entire product catalog, companies may increasingly ask how they can use AI effectively across that catalog while maintaining accuracy and consistency.

AI Economics Are Entering a New Phase

Falling prices are good news for companies consuming AI, but they also raise questions about the business models of the companies producing it.

The AI industry has committed enormous amounts of capital to computing infrastructure. If the price customers pay for model usage continues falling, providers will need increasing volumes, greater efficiency, or additional sources of revenue to justify that investment.

Competition may also move beyond raw model performance. When several models can perform similar tasks at increasingly affordable prices, distribution, integrations, proprietary data, user context, and specialized applications become more important.

The cost of intelligence is falling. That makes it easier to add AI to more workflows, process larger catalogs, and experiment with applications that previously may have been too expensive to operate at scale.

But cheaper AI does not automatically produce better results. As access to models becomes increasingly commoditized, the quality of the information surrounding them may become an even more important source of differentiation.