Uber burned through its 2026 AI budget in 4 months: the real lesson on costs
Here’s a fact that should interest you: Uber deployed Claude Code to its engineers in December 2025. Four months later, in April 2026, the company had exhausted its entire annual AI budget. Not because the tool failed. Quite the opposite: 95% of engineers were using it actively, and productivity skyrocketed.
What happened is that no one had planned for rapid, massive adoption. Each engineer cost between $500 and $2,000 per month in AI credits. With thousands of engineers on the payroll, the math was simple: the budget evaporated.
This sends a strong signal on two fronts. First, quality AI tools create real, immediate demand. People don’t leave them sitting on the shelf; they use them. Second, companies massively underestimate operating costs when they scale deployment. Uber didn’t budget for success.
For a small business, the lesson is different but just as clear: if you invest in serious AI tools (Claude, GPT, Gemini), you’ll need to budget API or licensing costs progressively and realistically. A quality tool costs more in actual usage than you’d expect, but it also generates more value. You just need to plan for it.
What this means for your business
What this means for your small business
If you’re thinking about giving your team (dev, marketing, content, support) access to AI tools, assume adoption will be faster and more expensive than planned. Don’t base your AI budget on theoretical formulas.
Two approaches: (1) Start small, with a pilot team, measure actual costs per user and per use case, then scale up progressively. (2) Opt for fixed-cost solutions (Copilot Pro subscriptions, ChatGPT professional plans) rather than pay-as-you-go APIs, to avoid surprises.
The real risk for a small business: having a budget too tight to absorb exploding demand. Plan for 30-50% contingency on your AI budgets.
In brief
Microsoft launches dedicated AI agent for legal teams
Microsoft is integrating an agent into Word specialized in contract management, negotiation history, and editing complex documents. The agent understands legal context better than generic models. SMB angle: if you have high legal turnover or few in-house lawyers, this tool can cut your contract review timelines.
The real problem with AI voice synthesis: integration into your workflows
AI voice quality is improving, but the real bottleneck remains production integration. Tools exist in isolation, disconnected from your existing systems. SMB angle: before investing in an AI voice solution (enhanced IVR, customer assistant), verify it integrates with your current CRM or phone system.
AI agents: execution cost is becoming a design factor
Developers are discovering that every API call costs real money. This is changing how agents are designed: fewer unnecessary loops, more upfront testing. A poorly calibrated agent can burn your budget without creating value. SMB angle: if you’re building a custom AI agent, ensure its operating cost (per transaction, per user) is viable.
Pentagon approves 7 AI vendors for classified environments, not Anthropic
OpenAI, Google, Microsoft, Amazon, Nvidia, xAI, and Reflection are approved for classified defense tools. Anthropic is excluded. This reflects growing geopolitics around AI vendors. SMB angle: if you operate in sensitive sectors (defense, critical energy), verify government certifications and approvals for your AI vendors.
OpenAI also restricts its sensitive tools (Cyber) to ‘critical defenders’
After criticizing Anthropic for its restrictions, OpenAI is doing the same with GPT-5.5 Cyber. Tool access is limited. Emerging pattern: AI vendors are locking down powerful capabilities. SMB angle: expect growing restrictions on powerful AI tools, even if you’re a paying customer.
Get The AI Brief in your inbox
3x per week, the essentials of AI decoded for business leaders.