Experts say that as the AI frontier shifts toward more powerful yet lower-cost models, major labs may gain the financial flexibility needed to intentionally slow the pace of capability development. Finding that balance matters because the leading companies must make slowing development financially viable or the economic incentives to keep racing will remain overwhelming.
Why this matters
Top U.S. startups including OpenAI and Anthropic, along with other AI players, are currently burning far more cash than they generate and are expected to do so for the foreseeable future. Executives from these firms have expressed a desire to slow model development following a series of troubling incidents in which rogue AI agents hacked outside companies.
To implement a meaningful slowdown, companies need a value proposition for customers that does not rely solely on achieving the absolute frontier of intelligence. Staying at that frontier requires continuous, often enormous spending — particularly on the compute resources needed to train successive models — effectively creating a treadmill of escalating costs.
How cheaper models help
Reducing the per-model cost could lower the pressure to keep increasing spending and improve returns for the labs, making a deliberate slowdown more affordable. Economists and AI professionals note this would not necessarily reduce investment in computing hardware, data centers or infrastructure upgrades such as the electrical grid, because demand for AI services is still rising rapidly and will require continued capacity.
Market signals this week
A clear signal from the market is that newly released models are increasingly cheaper without sacrificing top-level performance. OpenAI, Anthropic and SpaceX all emphasized the relatively low costs of their new models and their ability to outperform some of their own leading systems from only a few months earlier.
Chinese models have shown the same pattern. DeepSeek’s V4.1 Flash, released earlier this month, rose to the top of the OpenRouter leaderboard this week with a 172% spike in usage.
Economic dynamics and a paradox
The push to cut prices is not driven by safety concerns alone. Citadel Securities reported to clients that falling per-token costs are prompting enough additional usage to increase overall AI spending, meaning labs can see higher revenues even while reducing unit costs. This paradox explains why AI labs have often reduced model prices over time: lower prices expand the customer base and can boost profits.
Bottom line
The competition to lower AI costs may be the mechanism that lets labs financially support the slowdown they and some safety experts seek. Cheaper but capable models could reconcile the industry’s safety priorities with the economic incentives that have driven rapid development.
