Companies Invest in AI Without Controlling Costs

AI Investments Outpace Cost Control Capabilities

In a context where artificial intelligence is becoming a strategic pillar for many companies, a recent survey conducted among 107 companies reveals a concerning trend: investments in AI infrastructure are accelerating at a pace that exceeds companies' ability to measure their actual costs. While most organizations rely on hyperscalers and model provider APIs to execute their AI projects, they are increasingly turning to specialized computing. However, few of them are currently utilizing these advanced technologies. A majority are considering changing or adding suppliers in the coming months, highlighting a purchasing dynamic focused on integration and total cost of ownership rather than token prices.

A Concerning Calculation Gap

The study highlights a significant calculation gap: companies are investing heavily in AI infrastructures without having clear visibility on their unit economics. Only about 21% of the surveyed companies have reached a large-scale production stage for their AI projects, yet spending intentions far exceed this level of maturity. The field of specialized AI clouds is particularly sought after, with 45% of companies planning to evaluate these solutions over the next year, although few are currently using them. Meanwhile, 83% of companies report GPU usage at 50% or less, and less than half (44%) can accurately track the cost of their AI computing, underscoring an underutilization of existing resources.

A Willingness to Change Suppliers

Companies are not only concerned about optimizing their current resources, but they also show a strong intention to reassess their partnerships with infrastructure providers. A clear majority (64%) plan to change or add a supplier in the next twelve months, and 38% are considering doing so in the upcoming quarter. This intention to change is notably high for such a critical sector. When choosing new suppliers, companies prioritize integration with their existing technology stack (41%) and total cost of ownership (35%), while the cost per million tokens is a decisive factor for only 8% of companies. Furthermore, the transition from GPU computing to memory bandwidth, as inference develops, remains largely underestimated, with about one in five companies still unaware of this issue.

Survey Methodology

The survey, conducted by VentureBeat as part of its Pulse research series, focused on companies' AI infrastructure, computing, and inference economics. Responses were collected from organizations with more than 100 employees (n=107), excluding smaller companies. This study, conducted in Q2 2026, provides a cross-sectional view rather than a monthly trend analysis. Participants primarily came from the mid-market: 101–250 employees (36%), 251–1,000 (27%), and larger enterprises. Respondents' roles included managers (38%), individual contributors (28%), and end decision-makers (45%) for AI solutions.

Conclusion

The survey results highlight a growing ambition among companies regarding AI that exceeds their current production capabilities. While infrastructure decisions are heavily influenced by integration and total cost of ownership, specialized AI clouds are emerging as a priority for the future. However, without better visibility on actual costs and optimized use of existing resources, these investments could lead to costly inefficiencies.