Uber burned through its 2026 AI budget in 4 months: €500-2000 per engineer, what lessons for your company?

Uber just revealed a figure that should alert all business leaders: the company consumed its entire 2026 AI tools budget in just four months. The cost per engineer? Between $500 and $2,000 monthly—far more than a typical SaaS subscription.

This budget overrun isn't an isolated case. It illustrates a phenomenon we observe at AISOS among French SMEs and mid-market companies: AI tool consumption is exploding without control, driven by team enthusiasm and the absence of clear governance.

This article breaks down the Uber case to extract concrete lessons. You'll discover how to anticipate your AI costs, avoid the overconsumption trap, and build a realistic budget adapted to your company's size.

What happened at Uber: anatomy of a budget overrun

In January 2024, Uber massively deployed generative AI tools for its engineering teams. GitHub Copilot, ChatGPT Enterprise, Claude, and various coding assistants were made available to thousands of developers. The stated objective: accelerate productivity by 20 to 30%.

The financial results surprised management:

• Annual 2026 budget exhausted by April: consumption projections were four times lower than reality
• Average cost per engineer: between $500 and $2,000 per month depending on profile and usage intensity
• Unexpected API bills: calls to GPT-4 and Claude models generated massive unanticipated overcharges
• Cascade effect: each team developed its own integrations, multiplying subscriptions and consumption

The central problem: Uber budgeted its AI expenses like traditional SaaS licenses, with a fixed cost per user. However, generative AI tools mostly operate on a consumption-based model. The more you use, the more you pay.

Why AI budgets explode: three traps to avoid

Trap number one: underestimating the usage-based pricing model

Most AI tools charge by token, request, or compute time. A developer who intensively uses GitHub Copilot and queries GPT-4 to debug code can easily generate $50 to $100 in daily consumption without realizing it.

For an SME with 50 people where 10 actively use AI tools, monthly costs can jump from an estimated EUR 500 to EUR 3,000 actual. The gap widens each month if no monitoring is in place.

Trap number two: uncoordinated adoption

At Uber, as in many companies, each team chose its own tools. Marketing uses Jasper, developers prefer Cursor, customer support tests Intercom AI. Result: multiplied subscriptions, no volume negotiations, and inability to measure overall ROI.

AISOS audits reveal that companies with 100 to 500 employees use an average of 7 to 12 different AI tools, often with redundant features. Consolidation typically reduces costs by 30 to 40%.

Trap number three: absence of value measurement

Uber measured its expenses but not its gains. How much time saved per engineer? How many bugs avoided? What impact on delivery timelines? Without these metrics, it's impossible to know if the $2,000 monthly per developer generates positive returns.

A simple rule: every euro spent on AI should generate at least 3 euros of value, whether in time saved, errors avoided, or additional revenue.

How much to spend on AI according to your company size

There's no universal answer, but market data allows establishing realistic ranges for 2024-2025.

Very small businesses with 1 to 10 employees

Recommended monthly budget: EUR 100 to 500. At this stage, favor one or two versatile tools rather than specialized solutions. ChatGPT Plus or Claude Pro for general tasks, possibly one specific business tool.

Average cost per active user: EUR 20 to 50 per month.

SMEs with 10 to 50 employees

Recommended monthly budget: EUR 500 to 2,500. The governance question becomes critical. Designate an AI tools manager, centralize purchases, and negotiate enterprise licenses.

Average cost per active user: EUR 50 to 150 per month depending on usage intensity and departments involved.

Mid-market companies with 50 to 500 employees

Recommended monthly budget: EUR 5,000 to 50,000. At this scale, the risk of Uber-style overruns becomes real. An initial usage audit and clear adoption policy are essential before any mass deployment.

Average cost per active user: EUR 100 to 300 per month, with strong variations between technical versus administrative profiles.

Seven recommendations to control your AI budget

First recommendation: audit before deploying

Before opening access to AI tools, identify priority use cases. Which repetitive tasks consume time? Where can AI actually create value? A two-week audit avoids months of unnecessary spending.

Second recommendation: centralize purchases and access

Create a single enterprise account for each selected tool. Negotiate volume pricing. Assign access based on real needs, not spontaneous requests. An engineer doesn't need GPT-4 Turbo if GPT-3.5 suffices for their tasks.

Third recommendation: set consumption limits

Most platforms allow configuring alerts and spending limits. Use them systematically. A EUR 200 monthly limit per user forces prioritization of high-value use cases.

Fourth recommendation: measure ROI by use case

Document time spent on a task before and after AI. Calculate the hourly cost of affected employees. If a EUR 50 monthly tool saves 10 hours for an employee billed at EUR 50 per hour, the ROI is 10x. If the gain is 2 hours, reconsider the tool.

Fifth recommendation: train rather than stack tools

A trained user extracts 3 to 5 times more value from the same tool than a self-taught user. Invest in training before adding new subscriptions. Often, one well-mastered tool replaces three superficially used tools.

Sixth recommendation: review quarterly

AI tools evolve rapidly. Prices change, new alternatives appear, your teams' needs become clearer. Quarterly subscription reviews eliminate underused tools and renegotiate contracts.

Seventh recommendation: anticipate hidden costs

Beyond subscriptions, factor in integration time, training, internal support, and potential custom development. For every euro of subscription, budget 50 cents to 1 euro in ancillary costs for the first year.

The most common mistakes of French SMEs

Field experience reveals recurring patterns among companies that overspend on their AI budget.

Common mistake: confusing experimentation with deployment. Testing ChatGPT with a few employees costs little. Scaling without governance costs dearly. The transition from one to the other requires a framing phase that's often neglected.

Common mistake: neglecting open source solutions. For certain use cases, models like Llama 3 or Mistral hosted locally offer comparable performance to commercial solutions at marginal cost. This option remains under-explored by SMEs.

Common mistake: buying maximum power by default. GPT-4 costs 20 times more than GPT-3.5 per request. For 80% of common uses, the less powerful model suffices. Reserve power for tasks that truly require it.

Common mistake: ignoring bundled offers. Microsoft 365 Copilot, Google Workspace with Gemini, or OpenAI enterprise plans often include multiple AI features for less than the sum of separate tools.

Building a realistic AI budget for 2025

Here's a four-step method to establish your budget envelope without surprises.

Step one: identify potential active users. Not all your employees will use AI daily. Estimate the number of intensive, regular, and occasional users. Apply coefficients of 1, 0.5, and 0.2 respectively to weight costs.

Step two: list priority use cases. Writing, coding, data analysis, customer support, document research. Each use case has a different consumption profile. Coding and analysis consume more tokens than simple writing.

Step three: apply a safety coefficient. Multiply your initial estimate by 1.5 to 2 to absorb usage peaks and new needs that will emerge during the year. The Uber case shows a factor of 4 may be necessary without governance.

Step four: plan an experimentation reserve. Allocate 10 to 15% of your budget to test new tools without impacting the operational envelope. This reserve avoids painful trade-offs during the year.

What the Uber case reveals about the future of AI spending

Uber's overrun isn't a failure but a signal. It announces what all companies will experience as AI integrates into daily processes: growing costs, value that's difficult to quantify, and urgent need for governance.

Companies that will thrive are those treating AI as a strategic investment, with a decision framework, performance metrics, and regular reviews. Not as another IT expense buried in the overall IT budget.

For SMEs and mid-market companies, the advantage is paradoxically size. You can deploy AI governance in weeks, where Uber will take months to regain control. The key: start now, before uncontrolled consumption habits take hold.

Take time to audit your current usage, define your priorities, and build an adapted budget framework. Your 2025 income statement will thank you.