U.S. companies are talking far more about open-weight AI models as their bills for proprietary services climb, PYMNTS reported Sunday, citing the Financial Times. Mentions of open-weight or open-source models on earnings calls and at investor events rose about sixfold in August and September compared with the same months of 2025, according to data from the research platform AlphaSense.

What Happened

The companies that have discussed using open-weight models in recent weeks span well beyond the technology sector. They include PNC Financial Services, logistics group CH Robinson, industrial company Siemens and dating app Tinder, according to the report.

Tinder offered the most concrete numbers. Chief technology officer Vinay Kuruvila said the company’s AI spending was running at $1 million a year in January and had reached a $10 million annual rate by July, and that he did not want another tenfold increase. To contain costs, Tinder has begun routing queries from non-technical users to open-weight models rather than sending everything to proprietary systems.

RelatedNew York City Comptroller Urges Anthropic to Delay IPO Over Safety and Governance Concerns

Open-weight models are those whose trained parameters are published, so a company can run them on its own hardware or through a hosting provider of its choice instead of paying per request to the model’s developer. PYMNTS noted that the savings come with a trade-off: companies that adopt them take on more responsibility for the infrastructure that serves them.

Why It Matters

Earnings calls are where executives explain spending to investors, so a sharp rise in mentions indicates that AI inference costs have become large enough to discuss as a line item, not just an engineering choice. Tinder’s tenfold increase in six months shows how quickly usage-based pricing can compound once AI features reach a large consumer base.

Traffic data points the same way. Vercel said in its September AI Gateway index that open-weight models handled 56% of all tokens on its gateway in August, the first month they carried a majority of volume, up from 7% in December 2025. Average token costs on the gateway fell 23.2% in August alone, Vercel said, and it described customers moving workloads to cheaper models as capability gaps narrow.

Volume and revenue are diverging, however. On the same gateway, Anthropic still accounted for about 64% of all spending in August, with Claude Opus 5 alone taking 22.5% of spend. Customers appear to be routing high-volume, lower-stakes work to open models while paying for proprietary models on the tasks where they judge quality matters most.

Context

The approach Tinder describes, splitting traffic by task and sending simpler queries to cheaper models, depends on routing logic that decides which requests can go to an open model and which still need a proprietary one. Building and monitoring that layer, along with hosting the models, is part of the infrastructure burden PYMNTS referred to.

Technology companies remain the most obvious adopters, according to the report, but the names now raising the subject on investor calls include a bank, a freight broker and an industrial manufacturer. For those firms, the ability to run a model on their own hardware also bears on where customer data is processed, not only on price.

What to Watch

Third-quarter earnings season begins in October and will show whether the jump in open-weight mentions continues and whether companies start quantifying savings. On the supply side, the question is how proprietary labs respond on price as more high-volume work moves to open models, and whether Anthropic’s large share of spending holds as those shifts continue.