Every business, small or big has to address one critical aspect – ROI or return on investment. For a good part of the last two years big tech companies and beyond chased AI like Aladdin’s genie – like it would magically turn everything around with efficiency, cost cuts and cutting edge products. This bubble may be starting to burst, as several big tech firms such as Uber and Microsoft start raising important questions around whether the outcomes are justifying the cost. Microsoft, Amazon, Alphabet and Meta are collectively expected to have spent roughly $320 billion on AI infrastructure and related capital expenditure in 2025, according to estimates compiled from company guidance and earnings commentary.
Recently, Uber Chief Operating Officer (COO) Andrew Macdonald admitted the company is struggling to clearly justify some of its AI spending. Internally, executives reportedly started referring to the issue as “tokenmaxxing” which is essentially developers consuming huge amounts of AI compute tokens without a reliable way to determine whether all that spending is resulting in noticeably better products.
Just last month, Uber CTO Praveen Neppalli Naga had revealed that the company had already exhausted its entire allocated 2026 budget for Claude Code in just four months. Macdonald described the internal reaction as “head-exploding”.
What makes the situation interesting is that adoption itself does not appear to be the problem. Uber employees have embraced the tools aggressively. More than 92% of the company’s 5,000-plus developers reportedly use AI assistants every month. AI also now reportedly writes around 10% of Uber’s code.
But executives appear far less convinced that more AI-generated code automatically translates into a significantly better experience for users. “You talk to your senior engineering leaders, and you’re saying, ‘Okay, how many projects that were on the cutting room floor got moved above the line because of the productivity gains…?’ That link is not there yet,” Macdonald said.
The Uber COO highlighted that “If you’re not actually able to draw a direct line to how much useful features and functionality you’re shipping to your users, that trade becomes harder to justify.”
What’s really interesting in the backdrop of this is that just earlier this month during an earnings call, Uber CEO Dara Khosrowshahi announced a slowdown in hiring as the company shifted more resources toward AI investments. However, whether the economics of it all makes sense is now clearly under scrutiny.
Uber is not the only company that’s now wanting to tread with caution. Microsoft has reportedly started reducing some external AI coding subscriptions, including Anthropics’s Claude Code, as more and more companies come under growing pressure to control costs tied to large-scale AI deployments.
Language-learning platform Duolingo recently reversed an internal policy linking employee performance reviews to AI usage metrics after staff reportedly complained that they felt pushed to use AI tools even when those tools did not genuinely improve their work.
AI educator and creator Ansh Mehra believes many companies are making a more basic mistake altogether. According to Mehra, employees often rely on AI systems before properly structuring their own thinking, leading to output that still needs heavy human correction, review and debugging afterwards. He argues this can quietly create massive token costs without necessarily producing proportional gains in efficiency.
“The correct way to adopt AI is to audit specific steps in low stake/mundane work in departments and then have the most experienced people from those teams test LLMs and validate their outputs. Document what’s working and what is not working in isolation, before rolling out licenses to everyone,” Mehra said.
Source link
[ad_3]
