The Hidden Cost of Moving Fast: What Every Leader Needs to Know Before Spending on AI
The Deliberate AI Leader — A Series for Executives Who Want to Get This Right – Part 5
Summary:
In April 2026, Uber’s CTO admitted the company had burned through its entire annual AI budget in four months. This wasn’t a story about AI failing. It was a story about an organization that encouraged adoption faster than it could govern what adoption actually cost. That pattern is playing out everywhere — and it is not limited to companies with billion-dollar R&D budgets. This post explains what went wrong, why the financial model most leaders are using for AI is built for a different era, and what deliberate AI investment actually looks like in practice.
A Budget Story That Should Be Required Reading
In April 2026, Uber’s Chief Technology Officer Praveen Neppalli Naga delivered an unusually candid admission: the company had already spent its entire annual AI budget, and it was only Q1.
The cause wasn’t an infrastructure overrun or a failed vendor contract. It was a coding assistant. Specifically, Anthropic’s Claude Code — which Uber had rolled out to its engineering teams in late 2025 and actively encouraged through internal leaderboards that ranked teams by usage volume. By March 2026, 84% of Uber’s developers were classified as active agentic coding users. Token usage had nearly doubled in three months. The annual budget was gone.
Naga’s quote is worth reading directly: “I’m back to the drawing board because the budget I thought I would need is blown away already.”
Source: Benzinga — Uber’s Anthropic AI Push Hits a Wall
Here’s what makes this story instructive rather than just alarming: the tools worked. Engineers were more productive. Code was shipping faster. Uber isn’t walking back its AI investment — it’s expanding it. This wasn’t a case of AI failing to deliver. It was a case of an organization that didn’t have the financial framework to handle what happens when AI actually delivers at scale.
Source: Humai — Uber Burned Its Entire 2026 AI Budget in Four Months
That distinction matters enormously for how leaders read this story — and what they do next.
A brief disclosure worth making: WHIM works with Anthropic’s tools, including Claude, and we have a direct view into how these cost dynamics play out for organizations of various sizes. That perspective informs everything in this post.
The Problem Wasn’t Adoption. It Was the Incentive Structure.
The most important detail in the Uber story isn’t the dollar amount. It’s the leaderboard.
Uber built internal rankings that tracked how many tokens each engineering team consumed. Teams competed. Usage climbed. The organization successfully engineered the behavior it was trying to drive — and then discovered it had no financial infrastructure to support the behavior it had engineered.
This pattern has a name now. It’s called tokenmaxxing — the practice of maximizing AI token consumption, sometimes for genuine productivity and sometimes simply because volume has become a proxy for status, performance, and job security. The New York Times reported on the trend in March 2026, finding that internal leaderboards at companies including Meta and OpenAI were tracking token consumption the way sales teams track quota.
At Meta, one employee created an internal leaderboard called “Claudeonomics” that awarded titles like “Model Connoisseur” and “Cache Wizard” to high consumers. It was taken down two days later.
Ali Ghodsi, CEO of Databricks, described the failure mode plainly: “If your goal is to just burn a lot of money, there are easy ways to do that. Resubmit the query to ten places. Put up a loop that just does it again and again. It’s going to cost a lot of money and not lead to anything.”
Source: CNBC — AI Demand Is Inflated, and Only Anthropic Is Being Realistic
The lesson isn’t that incentivizing AI adoption is wrong. It’s that measuring adoption by volume — rather than by outcome — produces exactly the behavior you’d expect: maximum consumption, unmeasured return.
Why Your Mental Model for AI Costs Is Probably Wrong
Most leaders who have approved AI spending in the last two years did so with a mental model built for a different era of software purchasing.
That model looks like this: a per-seat license, a predictable monthly cost, a straightforward calculation of cost per user. It’s how organizations budget for Salesforce, for Microsoft 365, for Slack. It’s intuitive, auditable, and relatively easy to govern.
AI doesn’t work that way.
As Deloitte documented in their January 2026 analysis of AI spend dynamics, traditional cost frameworks — total cost of ownership, per-user licensing, static pricing — were designed for predictable workloads. AI workloads are neither predictable nor linear. Costs are measured in tokens — the fundamental unit of AI computation — and token consumption scales in ways that bear no resemblance to a subscription model.
Source: Deloitte — AI Tokens: How to Navigate AI’s New Spend Dynamics
Here’s what that means in practice. A chatbot interaction might consume a few thousand tokens. An AI agent autonomously reading a codebase, planning changes across dozens of files, running tests, and opening pull requests can consume millions of tokens in a single session. The engineer who uses Claude Code for occasional suggestions and the engineer who runs parallel agentic workflows overnight are both “using the same tool” — at costs that differ by orders of magnitude.
Deloitte found that AI has become the single fastest-growing expense in corporate technology budgets, consuming between a quarter and half of IT spend at some firms. Nearly half of business leaders expect it will take up to three years to see ROI from basic AI automation. Cloud computing bills rose 19% in 2025, with generative AI as the primary driver.
Source: Deloitte — The State of AI in the Enterprise 2026
For organizations actively deploying agentic AI tools, the cost reality can be jarring. Analysis of the Uber case found that enterprise AI tool costs often run $150–$250 per developer per month on average, with power users reaching $500–2,000 per month — far from the $20/month subscription pricing that appears on most marketing pages.
Source: ByteIota — Uber AI Budget Blown: Claude Code Costs in 2026
The financial reality of AI is not what most leaders planned for. The question is whether that surprises you reactively, the way it surprised Uber, or whether you get ahead of it deliberately.
The Two Traps Leaders Fall Into (And the Way Through)
When organizations face a technology wave this significant — moving fast, poorly understood, carrying real competitive implications — leadership tends to fall into one of two broken modes. Neither one is actually strategic.
Trap 1: Fear. This looks like waiting for certainty before acting. Holding off on AI investments until the technology “matures,” until a competitor proves the model works, until someone on the team really understands it. Fear-based inaction has a cost that doesn’t show up on a budget line — it shows up in competitive position, hiring, and organizational capability that erodes while others are building.
Trap 2: FOMO. This looks like reactive adoption driven by competitive anxiety. Buying tools because peers are buying them. Rolling out AI because the board asked about it. Measuring success by how many people are using it rather than what’s changing as a result. Uber’s leaderboard strategy is a textbook example. Shopify’s CEO telling employees to use AI or risk getting fired is another. These are FOMO at institutional scale — adoption mandated by anxiety rather than analysis.
Source: Inc. — What Is ‘Tokenmaxxing’? (covers Shopify CEO mandate)
Both traps produce the same result: spending that isn’t connected to strategy, and organizations that can’t articulate what they got for the money.
The way through is deliberate investment — a slower, harder, more durable approach that most leaders already know how to do with other kinds of capital, and simply need to apply to AI.
What Deliberate AI Investment Actually Looks Like
Deliberate AI investment isn’t cautious investment. It’s informed investment. The difference is that you know what you’re buying, what it should produce, and how you’ll know if it’s working. Here’s what that requires in practice:
Understand the cost model before you approve the spend. Ask specifically: is this priced per seat or per consumption? If it’s consumption-based, what does usage look like at 10% adoption? At 50%? At 80%? What happens to the cost curve when engineers or agents run more complex, autonomous workflows? These are not technical questions. They’re finance questions, and they should be answered before any deployment decision is made.
Measure outcomes, not usage. Token consumption is not a productivity metric. It is a cost metric. The only number that matters is what the spending produced: hours recovered, cycle times reduced, errors caught, capacity created. If you can’t answer that question specifically, you’re measuring the input without tracking the output. As we’ll cover in a later article in this series, defining success criteria before deployment is what makes evaluation possible afterward.
Build financial governance before you scale. For consumption-based AI tools, this means usage caps, real-time cost monitoring, and spending alerts before bills arrive rather than after. Deloitte recommends treating AI economics with the same rigor as energy or capital allocation. That framing is right: you wouldn’t let an energy contract run uncapped and unmonitored. The same discipline applies here.
Source: Deloitte — AI Tokens: How to Navigate AI’s New Spend Dynamics
Plan for adoption to outpace your forecast. Uber built its budget on early usage projections and watched actual adoption blow past them. If you are actively encouraging AI adoption — and you should be — budget for the scenario where it works faster than you expect. The $150–$250/developer/month figure above is a reasonable planning baseline for agentic tool environments; power users will exceed it.
Keep the CIO and CFO in the same conversation. Deloitte’s research puts this plainly: sustainable AI adoption requires a CIO who thinks like a CFO and a CFO who thinks like a CIO. Most organizations still have these as separate conversations. The leaders who will navigate AI spending most effectively are the ones who close that gap before the budget surprises do it for them.
Source: Deloitte — AI Tokens: How to Navigate AI’s New Spend Dynamics
The Question Worth Asking Before Your Next AI Meeting
The Uber story is already being read two ways. Some leaders are reading it as a warning: AI costs more than we thought, be careful. Others are reading it as a validation: the tools work, people want to use them, figure out how to afford it.
Both readings are correct. And neither one is a strategy.
The more useful question isn’t “should we be spending on AI?” It’s “do we have the financial literacy, the governance infrastructure, and the outcome metrics in place to know whether what we’re spending is working?”
If the answer is yes, you’re in a strong position to invest confidently. If the answer is no, that’s the thing to fix first — before the next tool gets approved, not after the next budget conversation goes sideways.
WHIM works with organizations at exactly this inflection point. If you’d like help building the financial governance and outcome framework that makes AI investment sustainable rather than reactive, a Strategy Call is the right place to start.
About WHIM Innovation
WHIM Innovation helps organizations harness the practical power of AI, automation, and custom software to work smarter and scale faster. We combine deep technical expertise with real-world business insight to build tools that simplify operations, enhance decision-making, and unlock new capacity across teams. From AI strategy and workflow design to custom monday.com apps and fully integrated solutions, we partner closely with clients to create systems that are efficient, intuitive, and built for long-term success.