In April, Uber’s CTO admitted the company had spent its full 2026 AI budget. Not in twelve months. In four.
The culprit was AI coding tools. Engineers loved them so much that monthly costs ran $500 to $2,000 per person, and the bill came due faster than anyone modeled. Then the part that should stop every executive cold: Uber’s own COO said it was hard to draw a straight line between all that spending and more value reaching customers.
A company with a $3.4 billion R&D budget and some of the best engineering talent on earth got blindsided by its own AI adoption. So let me ask the uncomfortable question. If it happened to Uber, what makes you think your AI rollout has guardrails Uber didn’t?
This Is Not a Tool Problem
The easy read on the Uber story is “AI is expensive, watch your spend.” That read is wrong, and it will cost you.
The tools worked. Engineers used them. Code shipped. The problem wasn’t the technology being bad, it was the technology being deployed into an environment with no controls around it. No usage limits. No model-matching. No way to measure whether a dollar spent produced a dollar of value. Adoption ran ahead of the infrastructure built to govern it.
I’ve watched this exact pattern for twenty years, long before anyone called it AI. A capability gets introduced. People adopt it fast because it’s genuinely useful. The organization discovers months and millions later that it never built the plumbing to control, measure, or scale the thing it just turned loose.
That’s not a budgeting mistake. It’s a foundation mistake. And foundation mistakes are the only kind that scale.
The Part Everyone Skips
Here’s what almost nobody does before they roll out an AI tool: they don’t decide, in advance, how they’ll know if it’s working.
Sounds basic. It’s the thing Uber’s COO was circling when he said the link between spend and value “is not there yet.” They had the usage data. What they didn’t have was the connective tissue between usage and outcome. The measurement layer that tells you a tool is earning its keep instead of just being popular.
You cannot govern what you cannot see. If you deploy AI without instrumentation, you’re not running an AI program. You’re running an experiment with a corporate credit card attached and no one reading the meter.
This is the boring part. It’s also the part that separates the companies that get ROI from the ones that get a surprise invoice.
What “the boring part” actually means
When I ran global operations teams of 200-plus engineers, the discipline that kept us out of trouble was never the exciting new capability. It was the unglamorous infrastructure around it: clear thresholds, real visibility, the ability to match the right resource to the right task instead of throwing the most expensive option at everything.
That same discipline is why one account I led at Kyndryl produced over $2 million in documented cost savings and a 38% improvement in incident detection. None of that came from a flashier tool. It came from building the foundation first, then putting the capability on top of it. The order is the whole game.
I see the inverse constantly. A business gets excited about a capability, signs the contract, and turns it on across the whole company in a week. The rollout feels like progress. Six months later, finance asks a simple question. What did we get for this? Nobody in the room can answer it with a number. The energy that felt like momentum was just spend with no scoreboard. By then the fix is expensive, political, and slow. Done on day one, the same fix is a checklist.
Why Small Businesses Should Read This as Good News
If you run a small or mid-sized business, the Uber story can feel like a warning that AI is a money pit only giants can afford to play in. It’s actually the opposite.
Uber’s problem was scale without control. You don’t have Uber’s scale, which means you don’t have Uber’s blast radius. A small business that sets up the right guardrails before deploying AI can move faster and waste far less than an enterprise that bolted AI onto a sprawling, ungoverned mess.
Smaller footprint, tighter feedback loop, fewer places for spend to hide. That’s an advantage. But only if you treat the foundation as step one instead of an afterthought you’ll “get to later.” Later is where the surprise invoices live.
The Three Questions to Answer Before You Spend a Dollar
Before you green-light any AI tool: coding assistant, customer-service agent, internal copilot, anything, answer these three. If you can’t, you’re Uber in April.
Can I see what it costs in real time? Not at the end of the quarter. Now. If your only visibility is the monthly bill, you’ve already lost the ability to steer.
Can I tell whether it’s producing value, not just activity? Usage is not outcome. Define what a win looks like before you deploy, or you’ll end up defending spend you can’t connect to results.
Can I match the tool to the task? The most expensive model for every job is how budgets evaporate. Disciplined deployment means using the right-sized capability for each use case and knowing the difference.
These aren’t AI questions. They’re infrastructure and governance questions. Which is exactly the point. The failures that look like AI failures almost always started as foundation failures that nobody addressed when they were cheap to fix.
The Pattern Won’t Stay Contained to Uber
Uber went public because the number was eye-catching and the CTO was candid. Plenty of companies are living the same story quietly right now and hitting their annual AI budget in a quarter, watching spend double, scrambling to ration usage after the fact.
The ones who get ahead of it won’t be the ones who spend the least. They’ll be the ones who built the controls before they needed them. Budget discipline beats budget size every single time, and discipline is an infrastructure decision you make on day one, not a fire you fight in month four.
The companies still treating AI infrastructure as something to sort out after deployment are writing their own version of the Uber headline. They just haven’t gotten the invoice yet.
Build the Foundation First
If reading this made you slightly nervous about an AI tool you’ve already deployed or one you’re about to that instinct is worth acting on.
Start with an AI Infrastructure Assessment. Before you spend another dollar on AI, find out whether you can actually see, measure, and control it. We’ll map exactly where your controls are strong and where the surprise invoices are hiding: https://www.summitaibs.com/contact
See how we sequence it. Our infrastructure-first approach exists precisely so you don’t become the next budget-burn story. Here’s how the engagements work: https://www.summitaibs.com/services
Keep reading. More on why AI projects fail before the AI ever shows up, over on the blog: https://www.summitaibs.com/blog
Russell Love is the Founder & CEO of Summit AI Business Solutions, based in Browns Summit, NC. With 20+ years of enterprise transformation experience at IBM and Kyndryl, Russell helps businesses build the foundations that make AI actually work.