Open-Source Models Soar, But Big AI Providers Still Collect Most Spend

Subheadline: Usage of lightweight, public models is climbing among developers, yet platform data shows a handful of premium providers continue to earn the bulk of dollars flowing through enterprise infrastructure.

Enterprise buyers and developers are increasingly routing production traffic to smaller, openly available models — but that shift hasn’t yet translated into a collapse in revenue for leading AI providers. Recent platform reports show the newest open-source entrants handling huge volumes of tokens, while a small number of proven commercial models still attract more than one-half of overall spending. That pattern supports a growing view among industry watchers: high-end models often serve as the proving ground for new applications, and cheaper alternatives take over once a workload is stable.

Traffic snapshots from a major deployment dashboard illustrate the split. One open model surged to process roughly one-third of token volume across the service in the past week, and another popular open-source family climbed into the top five. Yet when analysts drill into spending rather than sheer volume, the firm that markets several large commercial models remains the dominant revenue source on the platform, even after its recent price hikes nudged its share down slightly in the past month.

Broader usage data from a second marketplace paints a similar picture at larger scale. The top-performing open model there is moving trillions of tokens per week — roughly double the throughput of the most-used premium model on that service. But measured on cost per token, the commercial option charges about 23 times more than the cheapest open alternative, meaning the pricier model likely still commands most dollars. And a new entrant backed by major hardware relationships is positioned to contest both volume and revenue in coming quarters.

Several factors help explain why incumbent providers aren’t yet being displaced. One is simple: the set of problems companies ask AI to solve keeps expanding, so premium models can sustain demand by leading on novel or high-risk tasks. I spoke with a product chief at a financial services firm who described the workflow they use: they prototype on large, high-quality models to validate new, sensitive features, then switch to lighter models for scale once behavior is predictable. “We don’t want surprises in production,” she said, noting cost and latency concerns drive the move to open models over time.

Experts caution this two-tier arrangement could persist unless the performance gap narrows or pricing models change. An independent analyst offered that many enterprise applications still rely on capabilities — like complicated reasoning under strict safety constraints — that cheaper models don’t yet match. Others say commercial vendors have retained pricing power because they deliver consistent results in early-stage exploration, where customers are most willing to pay a premium.

For now, the market looks less like a zero-sum fight and more like a pipeline: premium models fuel discovery; open models pick up steady-state work. That could keep leading providers profitable even as open-source adoption widens — unless new architectures, more aggressive pricing, or tighter integrations from hardware players rapidly alter the economics. Observers say the next few quarters of token volumes and dollar flows will be telling.