The **Tokenomics Foundation**, a new organization focused on the financial and operational aspects of artificial intelligence, has released its first community survey, shedding light on how businesses are adapting to **FinOps** — a set of practices for managing financial operations in cloud computing and AI environments. The survey gathered nearly 500 responses from organizations across a variety of economic sectors, with the median revenue of the companies represented close to $2 billion. One key finding was the preference for proprietary AI models over open-source ones. More than half of the respondents (51%) strongly favored closed models, rating them 8, 9, or 10 on a scale from 0 to 10. However, the outlook for the next year is more open: around 70% of the organizations are planning to use open models at least as much as closed ones. This shift suggests a growing interest in open-source AI, which can be more cost-effective and flexible. Most of the organizations surveyed are already using AI models from major providers such as **Anthropic**, **OpenAI**, **Google**, and **DeepSeek**. Nearly all (96%) are using at least one of these services, and a large majority (87%) are using token-based inference services like **Amazon Bedrock**, **Vertex AI**, and **Azure Foundry**. These services allow companies to run AI models without maintaining their own infrastructure. The survey also found that 64% of the organizations are using embedded AI, which refers to AI tools integrated into specific applications or platforms, such as **Cursor**, **Windsurf**, **Databricks Genie**, and **Snowflake CoCo**. However, fewer organizations (31%) are using local or edge AI, which runs on devices closer to the data source. Similarly, 31% rent GPU computing power from cloud providers, while 29% use their own data centers or colocation facilities. The survey also highlighted challenges in managing the financial aspects of AI use. Many respondents expressed limited confidence in tracking costs and measuring return on investment (ROI). Those who felt most confident in these areas had the ability to attribute costs to specific workloads and teams, and to track concrete business metrics like customer support tickets, product releases, or revenue. Governance of AI use was mostly handled by IT departments (35%), though some organizations shared responsibility across functions (26%). Fewer mentioned involving management (9%), finance (5%), or data and AI functions (6%). The **Tokenomics Foundation** also noted that while AI routing gateways — tools that manage how AI models are used — are gaining attention, the market remains fragmented, with options ranging from established platforms like **OpenRouter** and **LiteLLM** to in-house tools and cloud-native solutions. The survey also showed that 38% of respondents are considering energy consumption in their AI operations. This is more common among those who own or rent hardware and use open-source models or train their own AI systems. Finally, the **Tokenomics Foundation** noted that 7% of respondents mentioned **FOCUS**, a FinOps specification that is expected to be updated in December. The update will include features like standardized model identification, workload attribution, and price catalogs for better cost prediction. The foundation has already released a guide on managing AI model caches and proposed a new version of the **Big-O notation** to better understand how AI systems consume computational resources.