Businesses are increasingly redirecting a portion of their conventional software spending toward AI-focused providers such as Anthropic and OpenAI, a trend that has been previously highlighted. New data from procurement platform Zip, detailed in its report "Where Is the AI Budget Actually Going?", offers further evidence of this shift.

Over the twelve months ending in August, AI service providers—including OpenAI, Anthropic, Cursor, and Sierra—accounted for 8% of total spending by Zip’s clients, a significant jump from just 1.4% in the prior twelve-month period. During the same timeframe, the median overall enterprise software budget rose by 13%. The study drew on data from dozens of customers, with an average workforce of 2,400 employees; these clients have collectively invested $18 billion in software, including AI tools, over the past four years. Zip’s customer base includes companies like Snowflake, OpenAI, Datadog, Cloudflare, and AMD.

This reallocation of spending is putting pressure on established software incumbents, compelling them to sharpen their competitive edge. In response, many enterprise software firms—such as Workday and HubSpot—along with cloud providers like Microsoft and Amazon, have introduced discounts or free trial offers for their AI products to retain clients.

However, even as these legacy players roll out complimentary incentives, they are simultaneously adopting more expensive pricing structures. Instead of bundling unlimited AI access into subscription packages, they are now charging customers based on actual AI usage. Once these free promotions expire, some businesses may conclude that the complications introduced by these tools outweigh their benefits.

Despite some discounts, the transition to consumption-based billing has resulted in steadily climbing invoices for certain clients. For instance, this summer, software vendor Pegasystems saw its monthly bill for Microsoft’s AI coding tool GitHub Copilot soar from $20,000 to $260,000, triggered by Microsoft’s shift to a metered billing model. David Vidoni, CIO of Pegasystems, expressed his frustration, stating, "What really troubles me is that AI vendors are offloading a significant burden onto companies like ours, forcing us to figure out how to use these tools most cost-effectively." Pegasystems, which provides fraud detection and customer billing automation for banks and telecom firms, spends between $10 million and $15 million annually on software, including AI.

After Pegasystems’ management pushed back against the steep charges, Microsoft offered the company a credit worth $10,000, equating to 1 million points, for a one-month trial of another tool called Copilot Cowork—an AI assistant designed to rival Claude Cowork. However, this report notes that the incentive pales in comparison to concessions Microsoft has granted other clients. Still, it has had some effect: Vidoni plans to expand testing of Copilot Cowork within the finance and marketing teams. Meanwhile, he is actively working to curb expenses related to GitHub Copilot and other AI tools by imposing usage limits on employees.