Shockwave Is AI More Expensive Than Humans?! The “AI Budget Blowout” Hitting Microsoft and Uber

📰 News Overview

• Microsoft Limits Internal AI Usage: The company has canceled most licenses for the much-loved “Claude Code” among engineers and is forcing a transition to its own GitHub Copilot CLI.
• Uber’s Budget Depleted in Four Months: After encouraging employees to use AI, Uber found itself using its entire annual budget for AI coding tools in just four months.
• Emergence of the “AI Paradox”: While token prices are projected to decrease by 90% compared to 2025, the consumption by agent-based AI is skyrocketing, increasing 24-fold and pushing total corporate costs higher.

💡 Key Points

• Computational Costs Surpassing Labor Costs: Nvidia executives have stated that “computational costs far exceed employee costs.” The infrastructure costs to run AI are becoming the largest fixed expenses.
• The Tokenmaxx Trap: The “use tokens to the max (tokenmaxx)” culture advocated by Amazon and Meta ironically becomes a bottleneck that pressures corporate profits.
• The Cost of Agentification: The proliferation of agent-based models, which consume far more tokens than standard models, is outpacing efforts to reduce costs.

🦈 Shark’s Eye (Curator’s Perspective)

Welcome to the “AI Horror Story” of 2026! It’s ironic that Microsoft, a major investor in Anthropic’s “Claude Code,” has to limit it due to its own runaway costs. This is the true “sacrifice of success”!

What stands out is Nvidia’s Catanzaro highlighting the reversal of “computational costs > labor costs.” It used to be said that AI would replace humans cheaply, but now the battle is on how to tame this “super expensive digital workforce.” Just as Gartner predicted, even if unit costs drop to a tenth, if agents work 24 times more, the bill will jump 2.4 times. Solutions to tackle the “poor fuel efficiency” of agent-based AI will hold the key to future dominance!

🚀 What’s Next?

Companies will likely transition from a phase of “unlimited AI use” to a selection phase where they allocate AI only to tasks that yield a worthwhile ROI. Additionally, the race to develop “super lightweight, specialized agents” for cost reduction will intensify!

💬 One Shark’s Insight

Budget blowouts from overusing AI are like gorging at an all-you-can-eat buffet and then being unable to move! A pro shark knows to feast strategically! 🦈🔥

📚 Glossary

• Tokenmaxx: A culture or mindset that advocates maximizing the use of “tokens,” the building blocks of AI, to achieve thorough automation and efficiency in operations.

• Agent-based AI: Advanced AI systems that not only provide answers but autonomously plan and execute tasks to achieve goals.

• Inference Costs: The expenses incurred when querying a trained AI model for answers, involving computational resources (like GPU costs).

• Source: Microsoft reports AI is more expensive than paying human employees