This earnings season delivered a message that nobody building an AI IPO prospectus wants to hear: the cost of intelligence is collapsing, and the companies that were supposed to own the premium tier are watching their pricing power evaporate in real time.

CNBC published a devastating analysis on Wednesday that puts hard numbers on the problem. Artificial Analysis, an AI benchmarking firm, ran every major model through the same 10 evaluations and tracked the total cost. The results: Anthropic's Claude came in at $4,811. OpenAI's ChatGPT: $3,357. DeepSeek: $1,071. Kimi: $948. Zhipu's GLM: $544. That's Claude costing nearly nine times more than the cheapest Chinese alternative for the same workload.

And the cheap alternatives aren't junk. DeepSeek's next-generation model matches or nearly matches the latest from OpenAI, Anthropic, and Google on coding, agentic, and knowledge benchmarks. This isn't a quality gap anymore. It's a pricing gap. And the pricing gap is getting wider.

The Enterprise Migration Is Already Happening

On OpenRouter, a marketplace where developers access hundreds of AI models through a single interface, Chinese models went from about 1% of usage in 2024 to more than 60% in May 2026. That's not a trend. That's a regime change.

Databricks CEO Ali Ghodsi has a front-row seat. His company's AI gateway sits between thousands of enterprise customers and the models they use, and he says the dominant pattern is now the "advisor model": a cheap open-source model handles the bulk of work, and when it hits something it can't solve, it calls out to a frontier model from OpenAI or Anthropic for help. The expensive model becomes the backup, not the default.

"You can curb costs really well this way," Ghodsi said. Enterprises are hearing that message. CloudZero found that 45% of companies spent more than $100,000 per month on AI in 2025, up from 20% the year before. Those budgets are growing, but the question of where they flow is changing fast.

Even Google Is Selling the 'AI Is Too Expensive' Narrative

At Google I/O this week, Sundar Pichai told developers that "many companies are already blowing through their annual token budgets, and it's only May." He pitched Gemini 3.5 Flash as the solution: if the largest Google Cloud customers shifted 80% of workloads from frontier models to Flash, they'd save more than $1 billion a year. Google is actively telling the market that frontier AI is overpriced. When your biggest cloud competitor is selling the cost reduction story, your premium moat has a credibility problem.

Meanwhile, a wave of Western challengers is building alongside the Chinese labs. NVIDIA, Cohere, Reflection, and Mistral are all building cheaper, more efficient alternatives specifically for enterprises that won't touch Chinese models for security reasons. The American monopoly isn't being replaced by a Chinese monopoly. It's being replaced by abundant choice at every price point.

The IPO Timing Problem

Here's what makes this so uncomfortable for OpenAI and Anthropic: both are racing toward IPOs valued north of $800 billion. Those valuations assume they'll hold their market share and pricing power. They assume competitors can't catch up. They assume enterprise customers will keep paying premium prices because there's no real alternative.

Every one of those assumptions is being challenged right now. The bill is already showing up in earnings: Meta, Shopify, Spotify, and Pinterest all flagged rising AI and inference costs as margin drags this quarter. Shopify specifically said economies of scale were "partially offset by increased LLM costs." Those are exactly the enterprise customers that OpenAI and Anthropic need to keep paying full price.

The counterargument is that these companies aren't selling a commodity. They're selling the platform, the brand, the safety research, the enterprise support. OpenAI has $55 billion in annualized revenue. Anthropic has $15 billion. Those numbers are real. But revenue at a premium price and revenue at a compressed price produce very different margin profiles, and margin profiles are what public market investors actually care about.

By the time these prospectuses become public, the central premise of both valuations may already be outdated. The frontier model premium is eroding from below (Chinese labs), from the side (Google), and from within (the advisor model pattern that turns frontier AI into a rarely used tool call). OpenAI and Anthropic aren't racing to IPO because the timing is perfect. They're racing because the timing might never be this good again.