AI Is Experiencing Its First Deflationary Shock — The Token Economy Is Feeling The Pain

• Instead of encouraging unrestricted AI adoption, companies are beginning to manage AI spending like any other operating expense.
• The Verge reported last month that Microsoft began cancelling most of its Claude Code licenses to cut costs.
• CEO Satya Nadella also questioned whether a future dominated by a handful of frontier-model providers is inevitable, arguing instead for a more competitive AI ecosystem.

Since ChatGPT ignited the generative AI boom in late 2022, the industry's playbook has been straightforward: build bigger models and spend more on computing power. But nearly four years later, that equation is beginning to break down.

As open-source models narrow the gap with frontier AI systems and inference costs fall, AI is entering what could be its first deflationary phase.

Shifting Paradigm

The change is already rippling through the industry's emerging token economy. As more AI vendors charge customers based on token usage rather than flat subscriptions, every prompt, completion and API call has become a measurable operating expense.

Instead of encouraging unrestricted AI adoption, companies are beginning to manage AI spending like any other IT budget, tracking token consumption, imposing usage limits and routing workloads to lower-cost models whenever possible.

Last month, The Verge reported that Microsoft Corp. (MSFT) began cancelling most of its Claude Code licenses to cut costs.

Earlier this week, CEO Satya Nadella also questioned whether a future dominated by a handful of frontier-model providers is viable, arguing instead for a more competitive AI ecosystem where customers can choose among multiple models rather than relying on a few premium vendors.

The Challenge For OpenAI And Anthropic

Those developments underscore a broader challenge facing frontier AI companies such as OpenAI and Anthropic. Their competitive advantage is no longer defined solely by building the most capable model. They must also preserve pricing power in a market where high-performing open models are proliferating, enterprises are becoming more cost-conscious, and AI intelligence itself is becoming increasingly commoditized.

Both OpenAI and Anthropic have confidentially filed their IPOs with the U.S. Securities and Exchange Commission (SEC). This could draw investor focus not only on their technological lead but also on whether they can sustain premium pricing in an increasingly competitive market.

Growing Competition

Enterprise cost controls are only one side of the equation. The other is a surge in capable open-weight models that are steadily reducing the premium once commanded by frontier AI systems.

Many of the most competitive open-weight models are emerging from China, rapidly narrowing the gap with proprietary offerings while undercutting them on price.

The latest example is Z.ai's GLM-5.2, released this month and praised by Silicon Valley developers for its coding and agentic capabilities.

The model ranks among the top performers on public AI benchmarks and delivers comparable coding performance to leading proprietary models from OpenAI and Anthropic at roughly one-sixth of the cost, according to a Reuters report citing the company and benchmark trackers.

Other Chinese companies, including Z.ai, DeepSeek and Alibaba, have increasingly embraced open-weight releases that allow enterprises to self-host and customize models rather than rely exclusively on premium APIs.

US Restrictions May Have Accelerated China’s Open AI Ecosystem

The rapid rise of cheaper open AI models may have an unexpected catalyst: U.S. export controls. Researchers from the University of Chicago and Chapman University argue that restrictions on advanced AI chips unintentionally encouraged China to invest in open AI ecosystems and compute-efficient innovation instead of relying on proprietary frontier models.

The researchers found that after major U.S. export-control measures, Chinese developers increased engagement with open-source LLM repositories far more than their U.S. counterparts while shifting research toward inference optimization, model compression and parameter-efficient fine-tuning.

Whether this deflationary trend ultimately hurts AI leaders remains an open question. OpenAI and Anthropic still hold advantages in frontier research, enterprise relationships and ecosystem integration.

But as open-weight models improve and enterprises prioritize cost over marginal performance gains, the AI race may increasingly be decided by economics rather than benchmarks.

The Global X Artificial Intelligence & Technology ETF (AIQ) is up 47% over the past 12 months, while the iShares U.S. Technology ETF (IYW) is up 44%.

Also See: IBM Just Packed 100B Transistors Into A Fingernail-Sized Sub-1nm Chip — Is It Rewriting Moore's Law?

For updates and corrections, email newsroom[at]stocktwits[dot]com.