Enterprise AI expenditure management is accelerating. In a report published on 23 June, the Swiss Syndicate Analyst Karl Keirstead team stated that approximately 60 per cent of businesses had begun to limit AI expenditure in some way, with a focus on the use of the fence for Token. The report suggests that this change may slow AI income growth in the short term, but will not change the direction of long-term demand expansion.
The high-price model is compressed
The enterprise did not stop using AI, but began to redistribute the budget. With the spread of AI Agent and the code tool, token consumption has shifted from a piecemeal trial to a running cost, and CFO has started to focus directly on the billing. Swiss Bank research shows that some companies have reduced the number of internal AI tools or imposed more severe monthly spending on individual users.
The model route has become the main instrument.
The most common practice for enterprises is not simple limits, but model paths. In other words, different tasks are assigned to different models, and only complex reasoning, key codes and context analysis can call for more expensive models. According to the Swiss Bank, this moves the high-end model from a default option to a more cautious one, with the firm more concerned about how much of the effective results could be obtained for each dollar.
China Open Source Model entered the procurement list
The report mentions that Chinese open source models such as Ali Qwen, DeepSeek, Mini Max, and GLM, are entering procurement and deployment options for more enterprises. A large global bank has deployed locally to balance the use of high-end models such as Claude. AWS Bedrock and Azure AI Foundation have also included multiple related models in optional menus.
The software company is in two positions.
According to the Bank, the impact on cloud manufacturers and hardware layers is relatively limited, as the need for reasoning remains on the clouds. Software companies, on the one hand, face a tight client budget and, on the other, have the opportunity to make themselves a “Token Optimization Platform”. Companies such as Palantir have been promoting model scheduling and cost optimization tools in an attempt to help clients reduce AI usage costs.