OpenAI's revenue growth slowdown in the second quarter masks a deeper shift in the AI market: demand is moving away from cheap, commodity models toward specialized, high-performance AI tools that solve real business problems. This realignment suggests that the months-long price war triggered by Chinese AI competitors may be ending, and U.S. frontier AI models are regaining competitive advantage.

Why Is OpenAI's Growth Slowing While Its Market Position Strengthens?

On the surface, OpenAI's second-quarter revenue growth of 18% quarter-over-quarter looks weak compared to Anthropic's more than doubling in the same period. However, analysts at Hyundai Motor Securities argue this headline number obscures what's actually happening in the market. The real story lies in where revenue is coming from and where demand is heading.

OpenAI's growth is increasingly concentrated in Codex, its developer-focused AI service. Tracked annual recurring revenue (ARR) for Codex, a metric that estimates ongoing subscription value, jumped approximately 50 percent between early July and mid-August, rising from $5.88 billion to $8.83 billion. Over the same period, Anthropic's Claude Code, a competing developer tool, grew only 6 percent, from $14.26 billion to $15.12 billion.

"The slowdown in OpenAI's second-quarter revenue growth should not be taken as a weakening of its AI model competitiveness," explained Kim Jae-seung, an analyst at Hyundai Motor Securities. "Because second-quarter revenue is a lagging indicator, the recent trend should be examined together."

Kim Jae-seung, Analyst at Hyundai Motor Securities

This shift matters because it reveals where corporate customers are actually spending money. Companies are moving away from simply choosing the cheapest AI option and instead investing in tools that deliver measurable business value. For developers and enterprises, that means paying more for AI that writes better code, reasons through complex problems, and powers autonomous agents.

What's Driving the Shift Away From Cheap AI Models?

Since May, demand had been flowing toward low-cost, open-source AI models from Chinese companies. These models offered significant savings on a per-token basis, the standard unit for measuring AI usage. However, that trend is reversing. Starting in August, token demand for U.S. AI models rebounded after stalling in July, signaling that some of that diverted demand is returning to American frontier models.

The key insight is that corporate buyers are shifting their decision-making criteria. Instead of asking "What's the cheapest AI per token?" they're now asking "What's the total cost to complete my task?" A high-performance model that solves a problem in one attempt may be cheaper overall than a budget model that requires multiple tries or produces lower-quality results.

This is especially true in three high-value areas where performance matters most:

• Coding and Software Development: Developers need AI that generates reliable, production-ready code, not just cheap inference.
• Reasoning and Complex Problem-Solving: Tasks requiring multi-step logic and analysis benefit from more capable models.
• Agentic AI: Autonomous AI agents that take actions on behalf of users demand high accuracy and reliability.

GPT-5.6 Sol, which OpenAI unveiled in June, received favorable reviews for its coding performance and competitive pricing, further supporting this trend. The model represents OpenAI's bet that developers will pay for quality.

How to Evaluate AI Models Beyond Price Alone

• Task Completion Cost: Calculate the total expense of completing a task end-to-end, not just the per-token price. A more expensive model that succeeds on the first try may cost less than a cheaper model requiring multiple attempts.
• Performance on Specialized Tasks: Assess how well an AI model performs on your specific use case, whether that's code generation, reasoning, or autonomous decision-making, rather than relying on generic benchmarks.
• Reliability and Consistency: Consider the consistency of results and the cost of errors. In production environments, reliability often outweighs raw cost savings.
• Infrastructure Requirements: Evaluate the computing power needed to run the model. High-performance inference demands more powerful hardware, which affects total cost of ownership.

Evidence of this shift appears in GPU pricing data. High-performance graphics processing units (GPUs) used for AI inference remain expensive in the United States, suggesting sustained demand for the computing power needed to run sophisticated models. If demand were truly moving toward cheaper, simpler AI, GPU prices would be falling. Instead, they're holding steady, indicating that companies are still investing in high-performance AI infrastructure.

The broader implication is that the AI market's race to the bottom may be ending. Chinese competitors' low-price strategy has limits. Continuously lowering prices while improving model performance is economically unsustainable. As the criteria for selecting AI models shift from simple per-token expense to the total cost of actually completing tasks, U.S. frontier models are expected to regain competitive advantage.

Hyundai Motor Securities projects that starting in the third quarter of 2026, as demand continues to shift toward high-performance U.S. frontier AI models, concerns about AI revenue monetization that have plagued the industry will gradually diminish. This suggests that OpenAI and other U.S. AI leaders may be entering a more stable, profitable phase of the market, even if headline growth rates appear modest.