Uber CTO Praveen Neppalli Naga just confirmed what finance teams across the industry have been quietly dreading: the company burned through its entire 2026 AI budget in four months. The culprit? Claude Code adoption that jumped from 32% to 84% of Uber's 5,000-engineer organization between February and March. Individual engineer costs ranged from $500 to $2,000 per month. Naga himself spent $1,200 in a single two-hour demo session. This is not an edge case. This is what happens when usage-based pricing meets highly motivated users with a direct incentive to consume as much as possible. And if your company has engineers with access to AI coding tools, your budget is next.

The story matters because it exposes a structural cost problem that no enterprise budgeting process was designed to handle. AI coding tools like Claude Code, GitHub Copilot, and OpenAI's Codex do not bill on a per-seat basis the way traditional enterprise software does. They bill on token consumption—meaning the invoice is a function of how many tokens the model processes across all engineer sessions combined. A developer running a simple autocomplete suggestion consumes negligible tokens. A developer running Claude Code as an autonomous agent across a monorepo, instructing it to refactor an API layer and generate associated tests in parallel, can consume thousands of dollars in a single afternoon. Scale that across 5,000 engineers running multiple concurrent agent loops, and the cost curve goes vertical in ways that annual budget cycles cannot absorb.

The Adoption Curve That Broke Finance

Uber rolled out Claude Code access to its full engineering organization in December 2025. By February 2026, 32% of engineers were using it monthly. By March, that figure hit 63%. By April, 84% of engineers were classified as "agentic coding users"—not just querying for code suggestions, but running multi-step autonomous workflows that generate, test, and refactor code without human intervention for long stretches. The tool did not fail. It did not underdeliver. Engineers loved it, productivity metrics improved, and roughly 70% of committed code at Uber now comes from AI. About 11% of backend updates are written entirely by AI agents with zero human code contribution. From a product and engineering standpoint, the rollout was a massive success. From a finance standpoint, the budget was incinerated. Naga was direct: "I'm back to the drawing board because the budget I thought I would need is blown away already."

The comparison to cloud cost is the most useful frame. In 2010, enterprise software teams started provisioning AWS compute with the same mental model they had used for on-premise servers: a capital expenditure, planned in advance, predictable. AWS bills arrived and were triple what finance had modeled. The pattern repeated across every organization that adopted cloud at scale. A decade of FinOps tooling, reserved instance strategies, tagging frameworks, and cost anomaly alerts was built to correct for that initial miscalibration. The AI coding cost problem is structurally identical. A usage-based pricing model has been placed in front of a highly motivated user base with a direct incentive to consume as much as possible. The tooling to monitor, cap, and allocate that spend at the individual team or engineer level does not yet exist at the maturity of cloud cost tooling. Uber is not unusual for having this problem. It is the first large company to surface it publicly at this level of specificity, which makes Naga's disclosure more valuable than it appears.

Photo by Lukas on Pexels

Why Leaderboards Accelerated the Burn

The internal leaderboard detail in Uber's rollout is worth examining separately. The company tracked and ranked engineer usage of Claude Code on internal performance visibility dashboards. That is a management choice designed to drive adoption, and it worked precisely as intended. It also created a cultural dynamic where using more AI tooling was visibly rewarded, and using less was implicitly under-performing. In a token-based billing environment, that incentive structure directly translates into budget acceleration. Engineers competing on leaderboards for AI usage have no obvious reason to be conservative about consumption. The people who designed the leaderboard were almost certainly not the same people responsible for the AI services budget line. That organizational gap—between the teams driving adoption and the teams managing spend—is the root cause of the overrun more than any pricing quirk of Claude Code itself.

For CTOs and VPs of Engineering, this creates a policy dilemma. Do you throttle adoption to protect budget predictability, or do you accept that AI coding tools are now a core infrastructure cost that will scale with engineering headcount and activity? Throttling adoption risks competitive disadvantage. If your engineers are using AI tools at 30% adoption while a competitor's engineers are at 80%, you are effectively operating with a productivity handicap. But accepting uncapped AI spend means renegotiating budget expectations with finance teams who are already skeptical about AI ROI claims. The middle path—implementing per-engineer or per-team spending caps, monitoring dashboards, and approval workflows for high-cost agent sessions—requires tooling that most companies do not yet have. Uber is building it now. Your company probably is not.

The CFO Perspective: From Cloud 2.0 to a New Budget Line

For CFOs, this is cloud overspend 2.0, except it is happening faster and with less visibility. Cloud cost at least had the benefit of being tied to infrastructure capacity—you could see the EC2 instances, the S3 buckets, the data transfer volumes. AI coding cost is tied to developer activity, which is inherently variable, unpredictable, and driven by task complexity rather than headcount alone. A single high-complexity refactor session can cost more than a month of routine autocomplete usage. There is no equivalent of a reserved instance or a savings plan in the current token-based billing model. You pay for what you use, when you use it, at spot pricing determined by the vendor.

The responses large enterprises will take from this situation are already visible in Naga's comments. Uber intends to give engineers access to OpenAI's Codex in the future, suggesting a multi-vendor strategy rather than a single-provider lock-in. That choice is likely motivated partly by competitive pricing leverage and partly by risk diversification. Companies watching Uber's experience will also accelerate conversations with Anthropic, OpenAI, and other major providers about enterprise framework agreements that replace token-based billing with committed spend deals at negotiated rates. Microsoft has already done this for Copilot: a flat per-seat model that limits the upside for the vendor but gives enterprise finance teams the predictability they need to budget reliably. As Claude Code usage scales across the industry, Anthropic will face increasing pressure from enterprise procurement teams to offer similar structures, or watch finance departments cap usage and throttle the adoption that is driving their revenue growth.

The third option—building internal coding agents on top of open-weight models—is gaining credibility precisely because of stories like this one. A company running Qwen 3.6-27B locally on dedicated GPU hardware has predictable per-query costs that are a function of hardware depreciation and electricity, not per-token billing. The setup cost is higher. The ongoing cost is bounded. For organizations with 5,000-plus engineers generating multi-hundred-dollar monthly AI bills per head, the build-versus-buy calculation is no longer theoretical. Uber burned its budget in four months. The next company to do the same will have more options for what happens in month five.

What This Means for Enterprise AI Budgets in 2026

The Uber case is not an isolated incident. It is a signal of how enterprise AI budgets will behave under current vendor pricing models. Token-based billing was designed for API usage at scale—chatbots, customer service automation, document processing. It was not designed for the consumption patterns of 5,000 engineers running autonomous coding agents in parallel across multiple repositories for 8-10 hours per day. The current pricing model assumes sporadic, task-specific usage. The actual usage pattern is continuous, workflow-embedded, and productivity-multiplying. The delta between those two assumptions is where budgets break.

For companies that have not yet rolled out AI coding tools at scale, Uber's experience provides a blueprint for what not to do:

1. Do not incentivize unlimited consumption (no leaderboards tied to usage volume)
2. Do not assume annual budget models will hold (plan for monthly variance)
3. Do not let engineering drive adoption without finance visibility (joint governance from day one)
4. Do not ignore multi-vendor strategies (avoid single-provider lock-in on token pricing)
5. Do not skip the build-vs-buy calculation (open-weight models may be cheaper at scale)

For companies that have already rolled out AI coding tools, the immediate actions are clearer:

1. Implement per-engineer or per-team spending caps (with override workflows for legitimate high-cost work)
2. Build real-time cost visibility dashboards (engineer-level attribution, not just aggregate spend)
3. Negotiate committed spend agreements with vendors (flat-rate or volume discount structures)
4. Evaluate open-weight alternatives for high-volume use cases (Qwen, Llama, DeepSeek)
5. Treat AI coding cost as a permanent budget line (not a one-time project expense)

The Vendor Pricing Reckoning

The broader implication is that Anthropic, OpenAI, and other AI vendors will face increasing pressure to offer enterprise pricing structures that decouple value from token consumption. The SaaS playbook is per-seat pricing. The cloud playbook is reserved capacity with volume discounts. The AI playbook is still being written, and Uber's budget overrun is evidence that the current model—pay-per-token at spot pricing—does not work for enterprise-scale adoption of productivity tools that are designed to be used continuously.

Microsoft's Copilot pricing model ($30/user/month) is one answer. It caps the downside for customers and the upside for Microsoft, but it provides budget predictability. Anthropic and OpenAI will likely introduce similar structures in 2026 for enterprise customers, or they will watch finance teams throttle adoption and limit market penetration. The alternative—letting every large customer burn through their annual budget in four months—creates a market dynamic where AI tools become a finance risk rather than a productivity asset. That is not sustainable.

Continue Reading

Sources

1. The Information - Uber AI Budget Report (Uber CTO confirmation)
2. Startup Fortune - Uber AI Budget Analysis
3. Anthropic Claude Code Documentation

---

What's your take? Have you seen AI tool costs spiral beyond budget projections? Share your thoughts on LinkedIn, Twitter/X, or via the contact form.

— Rajesh Beri
Enterprise AI perspectives for technical and business leaders