Businesses Shifted Toward Open-Weight AI Models
Companies are increasingly migrating enterprise tasks to open-weight infrastructure to cut costs and gain control.
Updated on Sept. 28, 2026 in Artificial Intelligence
Live Poll
Do you believe businesses should prioritize cutting technology costs over using premium proprietary software tools?
Mentions of open-weight models in earnings calls and investor conferences increased sixfold in August and September 2026 compared to the prior year. This transition reflects a broader trend of firms seeking to reduce expenditures on proprietary artificial intelligence services.
Why it matters
Enterprises are adopting open-weight models to bypass the high costs of proprietary AI, aiming for greater flexibility in their infrastructure. This shift highlights a growing corporate strategy to optimize AI spending while maintaining control over data and deployment processes.
Tinder scaled its annual AI spending from $1 million in January 2026 to $10 million in July 2026, shifting some non-technical user queries to open-weight models to manage this growth. These models allow firms to host software on their own hardware rather than relying solely on external cloud-hosted APIs.
The players
Tinder
A social discovery platform utilizing AI to manage user queries at a scale that reached $10 million in annual spending by mid-2026.
PNC Financial Services
A diversified financial services organization that has publicly discussed the integration of open-weight artificial intelligence models.
CH Robinson
A global logistics provider exploring the implementation of open-weight models within its operational technology stack.
Siemens
An industrial technology conglomerate incorporating open-weight artificial intelligence models into its digital infrastructure.
Nvidia
A leading provider of specialized graphics processing units and compute hardware critical for training and running high-performance artificial intelligence models.
The details
Companies utilize open-weight models—AI systems where the internal parameters are accessible for deployment but the underlying training data may not be—by running them locally on internal hardware stacks. By routing specific tasks to these models, organizations can reduce dependency on proprietary, cloud-hosted providers. This transition provides companies with direct control over their infrastructure, allowing them to iterate faster than if they were limited to the constraints of external proprietary platforms.
Timeline
January 2026: Tinder's annual AI spending rate was $1 million.
July 2026: Tinder's annual AI spending rate increased to $10 million.
August 2026: Mentions of open-weight models in earnings calls began a sixfold increase.
September 2026: The surge in mentions of open-weight models continued through the month.
September 27, 2026: Financial Times reported on these emerging usage trends.
The Tech Race
This move toward open-weight models represents a tactical departure from the dominance of black-box proprietary APIs. It marks a shift where enterprise utility and cost management are challenging the early lead of consolidated frontier model providers.
Enterprise software users may notice faster deployment cycles and reduced latency as companies migrate workloads to internal hardware. For developers, this signals a transition toward building on open architectures that prioritize infrastructure ownership over exclusive reliance on commercial cloud APIs.
The takeaway
The pivot to open-weight models is a strategic reaction to the rising costs of scaling proprietary AI within enterprise workflows. Watch upcoming quarterly earnings reports for data on whether this trend successfully decouples company spending from the rapid growth in commercial AI API fees.
Further reading
For broader context on the industry's shift toward accessible infrastructure, explore our coverage of Artificial Intelligence.
Source note: This article includes information reported by PYMNTS.
Live Poll
Do you believe businesses should prioritize cutting technology costs over using premium proprietary software tools?
