AI Vendors Shift Infrastructure Costs, Driving Up Software Budgets
Forrester warns of price hikes and usage charges as vendors pass compute expenses to clients
An extensive survey by research firm Forrester warns that enterprise software budgets will surge as vendors pass down mounting AI infrastructure costs to customers. Facing multi-billion dollar capital expenditure bills for data center buildouts, major tech companies are shifting away from flat-rate subscriptions toward unpredictable usage-based pricing models. This strategy marks a significant turning point in who bears the financial burden of scaling generative AI applications.
Key details
According to a Forrester survey of over 2,600 business and technology decision-makers, 80 percent expect their data and software budgets to increase next year. The rise is directly attributed to software and AI vendors raising baseline prices and introducing new usage-based fees. The report highlights that prominent vendors—including Anthropic, OpenAI, GitHub, and Microsoft—have already begun shifting premium services toward usage-based billing or introducing more expensive high-tier packages.
This pricing shift comes as vendors face astronomical bills for AI training and inference. Consultants Bain & Company recently estimated that the cumulative build cost for AI data centers could reach $2 trillion by 2030. To recoup these infrastructure investments, software companies are opting for token-based, usage-driven runtime pricing. For example, Microsoft's premium E7 license bolts generative AI assistant tools onto enterprise contracts for an additional premium, while other providers are strictly billing based on API consumption. This has created significant forecasting issues, with consultancy KPMG reporting that nearly one-third of corporate leaders struggle to understand or control operating costs when implementing AI at scale.
Why this matters
This development demonstrates that the underlying resource costs of AI—compute time, specialized server hardware, and massive grid electricity—cannot be sustained indefinitely through venture funding or subsidized flat-rate subscriptions. As these physical resource limits translate into financial line items, enterprise customers are being forced to adapt. Runaway token spend could severely impact the adoption rate of generative AI, transforming it from an easily accessible tool into a highly optimized, metered utility.
Context
Over the past year, AI hardware manufacturers like Nvidia have raised component prices significantly, with Blackwell-generation GPUs seeing price hikes of up to 55 percent. At the same time, regional utility grids are experiencing unprecedented strain from data center operations, leading to higher electricity rates. Because software vendors must build and maintain these capital-intensive facilities, they have found themselves caught in a margin squeeze. By adjusting their business models to pass operating expenses directly to the end user, they are protecting their own margins at the expense of their clients' IT budgets.
Risks and open questions
The primary risk is the high unpredictability of usage-based billing, which conflicts with traditional enterprise budgeting cycles. Unmanaged, automated AI agents running in production could trigger runaway query loops, resulting in catastrophic billing surprises. Furthermore, there is a risk that companies will cut back on essential IT staffing to cover ballooning software bills; though Forrester notes that data-specific hiring remains strong for now, overall IT departments are facing a structural margin squeeze.
What happens next
To mitigate these escalating software expenses, enterprises are expected to rapidly expand their FinOps (financial operations) practices to cover AI token consumption. This will drive a wave of "token-trimming" practices, model routing, and semantic caching, where companies deploy local middleware to intercept and reduce redundant calls to external LLMs. Additionally, organizations will likely set hard usage guardrails and budgets, potentially slowing down experimental deployments in favor of strictly scrutinized, high-ROI use cases.
Source: The Register Published on AI Usage Global, author: AUG Bot
