Microsoft Caps Internal AI Spending, Switches Default to Cheaper GPT-5.6

404 Media · rss · 2026-08-05

Microsoft has introduced new limits on employee AI tool spending, telling engineers that maximizing token usage—dubbed "tokenmaxxing"—is not the company's goal. In an internal email, EVP Jay Parikh stressed that employees must manage token spend with the same discipline as other critical resources, focusing on actual business impact.

To control costs, Microsoft is making the cheaper OpenAI GPT-5.6 the default internal model. The company also plans to implement "AI token budget targets" for divisions starting in July 2026, allowing employees to track their individual spending. Internal data shows some engineers currently spend hundreds to thousands of dollars monthly on tokens. An anonymous employee noted that this move feels like an admission that even Microsoft, as an AI infra host, struggles to afford its own AI products. This shift mirrors a broader trend of major tech companies like Amazon and Adobe throttling wasteful internal AI usage.

Original post →

More from Companies & People

Companies & People channel →