A new industry report highlights that the primary challenge for enterprise AI is no longer just deployment, but managing continuous model updates. For investors, this shift indicates that IT services and enterprises must now navigate volatile token-based costs and complex technical migrations to protect margins and ensure long-term value.
The phase of simply experimenting with artificial intelligence is ending, and the era of active lifecycle management is beginning. A new report by Straive highlights that the main hurdle for businesses today is not just putting AI into production, but managing the constant updates and behavioral shifts in foundation models. This evolution has direct implications for how companies—especially those in the IT services and technology sectors—will measure the return on their AI investments.
The Complexity of Model Migration
For many years, software updates were predictable. Developers would push a new version, and the system would generally behave as expected. However, the report points out that foundation models are fundamentally different. When a provider releases an updated model—such as a new version of a Large Language Model—it often brings significant changes to reasoning capabilities, accuracy, and even how the AI interacts with data. This means that migrating from one model generation to another is not a standard update. It requires a total re-evaluation of safety protocols, evaluation datasets, and business workflows to ensure the AI remains reliable.
The Risk of Fluid Cost Structures
One of the most critical financial risks for enterprises is the unpredictability of consumption-based pricing. Unlike traditional software with fixed licensing fees, AI models often charge based on usage, measured in tokens. This creates a cost structure that is highly sensitive to every interaction. The report notes that small adjustments to settings or reasoning depth can lead to significant cost variances—sometimes up to four times higher—without any change in the end application's goals.
Using recent examples, the price jump between model generations can be substantial. For instance, the move from Gemini 2.0 Flash to Gemini 2.5 Flash saw input token costs rise from $0.10 to $0.30 per million, with output costs jumping from $0.40 to $2.50 per million. These pricing swings make it difficult for companies to forecast AI budgets accurately. If IT service providers and enterprises cannot effectively manage these consumption-based costs, it could put pressure on project profitability and overall margins.
Investor Angle: What to Track
For investors following the technology and IT services sector, the focus is shifting from "how many AI projects a company has" to "how efficiently companies can manage AI at scale." Companies that prioritize cost-efficient system design and robust AI lifecycle management processes are likely to see better returns on their technology investments.
Moving forward, the key monitorable will be how enterprise leadership teams and IT services firms discuss their strategy for AI maintenance and cost control in their earnings calls. Investors may look for management commentary on how these firms are navigating model migration risks and whether they have established frameworks to optimize token usage, rather than simply consuming the latest models without regard for the financial impact.