Corporate buyers push back on larger AI contracts as renewal deadlines arrive

Corporate customers are pushing back against demands for larger annual spending commitments when renewing their contracts with major AI providers Anthropic and OpenAI.

Reporting from The Information reveals that enterprise buyers are moving away from automatic contract expansions, demanding clear evidence of commercial returns before agreeing to higher spending levels. Both AI providers have been encouraging customers to sign multi-year agreements with high minimum spending commitments, backed by heavy infrastructure investments. However, buyers are resisting those long-term commitments as model performance changes quickly and prices fall across the sector.

Shifting workloads to smaller models

A central factor in negotiations is the growing usability of smaller, cheaper systems. Corporate technology teams are routing everyday workloads away from flagship systems toward lightweight alternatives, including OpenAI models like Terra and Luna, or Anthropic models like Haiku. These smaller options cost a fraction of top-tier systems while handling standard automation tasks with similar accuracy.

Flexibility has also become a priority over single-vendor loyalty. Companies increasingly rely on model routers and intermediary tools such as OpenRouter, which allow them to redirect traffic between providers when prices change or when a rival service launches an update. For example, coding platform Replit shifted its core model integration from Anthropic to OpenAI in just a few hours.

A recent survey by Deloitte found that only 44 percent of businesses track their actual cost per unit of AI processing, even while overall workplace usage increases. That tracking gap makes procurement leaders wary of signing large, open-ended minimum spend agreements that could leave them paying for unused compute.

What this means for Australian businesses

Small and medium Australian businesses often face pressure from software sales teams to commit to expensive, multi-year tiers. The pushback among large enterprise buyers offers clear guidance on how local firms should manage their own supplier discussions.

• Avoid multi-year lock-in agreements for model access. Frontier models drop in price and improve in performance every few months, so paying on usage or sticking to monthly software tiers keeps options open.
• Match the model size to the task. Most administrative automation, email drafting, customer service replies and document sorting run effectively on small models like Claude Haiku or GPT mini tiers, which can cut running costs by more than 80 percent. Reserve top-tier reasoning models for complex programming or detailed analytical work.
• Measure the return before expanding seats. Rather than buying workplace licences across an entire team, track the exact hours saved or tasks completed in a small pilot before adding more users.
• Build your workflows with portable connections. Connecting tools through open standards, direct webhooks or standard APIs rather than proprietary walled gardens makes it easy to switch providers if prices increase or service terms change.