Enterprise AI cost controls arrive as Walmart, Uber, and Microsoft rein in usage

Walmart has placed usage caps on Code Puppy, its internal AI coding assistant, after strong employee uptake pushed costs significantly higher than expected. The retailer implemented a token-based system to tie consumption to business value, while keeping access to other platforms, including Claude and ChatGPT, available to staff. The move, reported by Bloomberg and cited by The Economic Times, is one of the clearest signs yet that corporate AI strategy in 2026 is no longer about getting everyone on the tools.

It is about figuring out what actually pays off.

A pattern forming across large tech users

Walmart's decision is not an isolated one. Uber rolled out Claude Code to around 5,000 engineers earlier this year and, according to The Economic Times, reportedly burned through its entire annual AI budget within months. The speed of that spending forced a harder look at how engineers were using the tool and whether the returns justified the pace.

Microsoft took a different approach. The company asked thousands of its own engineers to migrate off Claude Code and onto an internally developed alternative before the end of June 2026. The decision was widely interpreted as a cost-management measure, a signal that even one of the largest AI investors in the world is applying tighter scrutiny to tool selection and licensing spend.

GitHub, meanwhile, moved Copilot to token-based pricing, directly linking what customers pay to what they consume. The change makes AI costs more visible and gives procurement teams a clearer picture of where money is going.

From experimentation to accountability

For most of the past two years, enterprise AI strategy was driven by urgency. Companies wanted employees using AI tools quickly, and the priority was access over oversight. That posture made sense during a period when the technology was new and the goal was to build fluency across organizations.

What is emerging now is a second phase. Governance frameworks, usage controls, and cost-optimization are moving up the agenda. The question driving boardroom conversations has shifted from 'are we using AI?' to 'where is AI actually generating returns?'

This is not a pullback. AI remains central to enterprise technology investment across sectors. But the unlimited-access model that defined early adoption is giving way to something more deliberate. Organizations are building the internal infrastructure to measure impact, allocate budgets more precisely, and concentrate AI deployment on the use cases that move the needle.

What the shift means for vendors and buyers

The move toward consumption-based controls has real implications for AI vendors. Token-based pricing models, already in place at GitHub, put more pricing power in the hands of enterprise buyers and create pressure on providers to demonstrate clear per-unit value. Flat-rate or seat-based licensing becomes harder to justify when finance teams are asking for line-item accountability.

For enterprise buyers, the shift creates both an opportunity and an obligation. Companies that build rigorous frameworks for evaluating AI ROI now will be better positioned to scale the tools that work and cut the ones that don't. That discipline is likely to become a competitive differentiator as the field matures.

The organizations setting the pace in this new phase are not necessarily the ones with the most AI tools deployed. They are the ones that can show what each tool is worth. Walmart's token-based system, however early-stage, is one model for how that accountability gets built into daily operations.

What comes next

The second half of 2026 will likely see more enterprises introducing similar controls, particularly as annual budget cycles force a reckoning with AI line items that grew faster than planned. Procurement, finance, and IT leaders are increasingly at the table for decisions that used to sit entirely with engineering or innovation teams.

Microsoft's internal deadline of end-of-June 2026 for its engineer migration has already passed, making it one of the first concrete data points on how a major tech company manages an internal AI tool transition at scale. How productivity holds up in its aftermath may shape how other large organizations approach similar decisions in the months ahead.