SAP Chief Financial Officer Dominik Asam said corporate spending on AI tokens is "going through the roof" [1].
The surge in spending reflects a broader corporate rush to integrate generative AI into operations, but it raises questions about the long-term cost-efficiency of these investments.
Asam said that firms do not always need the most expensive AI models to achieve their goals [1]. He said that companies should instead focus on governed, business-process-specific models rather than relying on generic, high-cost alternatives [3]. This targeted approach is intended to ensure that AI capabilities are aligned with specific operational needs rather than broad, unfocused applications.
While SAP views these costs as a necessary investment in capability, other market observers have raised concerns about the impact on the bottom line. Chamath Palihapitiya said that the practice of "token-maxxing" will eventually hurt corporate earnings [4].
Asam said that AI must move beyond the "low-hanging fruit" of simple chatbots to deliver meaningful returns [3]. By shifting focus toward specialized models, companies may be able to curb runaway costs while increasing the actual utility of the technology within their business workflows.
This transition toward governed models represents a shift from the initial experimentation phase of generative AI toward a more disciplined, industrial application of the technology.
“AI token spending is "going through the roof"”
The tension between SAP's view of token spending as a necessary investment and critics' warnings about earnings suggests a looming correction in how companies budget for AI. If firms continue to prioritize the most powerful generic models over specialized, governed versions, they risk inflating operational costs without a proportional increase in productivity.
