Photo Credit: Depositphotos.com
Microsoft’s latest AI pullback is raising an uncomfortable question for the tech industry: What if using artificial intelligence at scale ends up costing more than the labor it’s supposed to streamline?
Microsoft is canceling most of its direct Claude Code licenses just months after encouraging employees to embrace the tool.
Fortune, citing The Verge, said that Microsoft steered engineers away from Anthropic’s Claude Code and over to GitHub Copilot CLI, even though access to Claude Code was opened only about six months ago. Thousands of developers, designers, project managers, and other employees had reportedly been urged to try it, and the tool seems to have spread quickly.
The change does not alter Microsoft’s broader Foundry arrangement with Anthropic, which involves a multibillion-dollar commitment and customer access to Claude models. However, it does suggest that internal use may have become hard to justify at the scale employees were using it.
Microsoft is not alone. Fortune, citing The Information, reported that Uber CTO Praveen Neppalli Naga said in April the company had used up its 2026 budget for AI coding tools in only four months. That came after internal incentives pushed teams to compete on AI usage.
Many companies have pitched AI as an efficiency booster that saves time and money, but these reports suggest the math may be more complicated. If such tools are expensive to run at scale, employers may limit access, shift expectations, or make cuts elsewhere to cover the cost.
If businesses are spending big on AI infrastructure and software, those costs can show up in the price of digital services, enterprise tools, and even hiring decisions. It also complicates the argument that AI will replace large amounts of humans because, in some cases, the computing bill may be higher than the payroll savings.
There’s also a direct connection to the energy grid. AI can help utilities forecast demand, manage transmission, and make it easier to integrate solar and wind power. But AI systems also require enormous amounts of electricity and water, driven by power-hungry data centers. With high demand, communities face grid strain, rising utility costs, and pressure on local resources; there are concerns about misuse, security, and unintended social effects, too.
Some companies appear to be responding by tightening controls rather than abandoning AI. Microsoft is steering workers toward an option that is closer to home, and other firms may follow with usage caps, narrower approvals, or more targeted rollouts focused on tasks that actually save time.
Fortune also cited Goldman Sachs’ projection that agentic AI could lift token consumption 24-fold by 2030, reaching 120 quadrillion tokens per month. The outlet summarized research firm Gartner’s view that the cost of running highly advanced models may fall sharply but not enough to guarantee lower enterprise bills because these systems use far more tokens per task.
“For my team, the cost of compute is far beyond the costs of the employees,” Nvidia’s Bryan Catanzaro said, per Fortune.
Gartner senior director analyst Will Sommer offered a similar warning, saying, “Chief product officers should not confuse the deflation of commodity tokens with the democratization of frontier reasoning.”
Get TCD’s free newsletters for easy tips, smart advice, and a chance to earn $5,000 toward home upgrades. To see more stories like this one, change your Google preferences here.
