$375 billion in AI investments. But 95% of projects are failing to deliver. Which raises the question: Are we really in an AI bubble?
The bubble debate misses the point. What matters: understanding which conditions create the 5% that succeed. AI infrastructure investments play out over decades, not quarters. Therefore, success requires years of foundational work.
Let me break down the core arguments raised in the AI bubble discussion.
The Bubble Signals
Productivity gaps: Despite $4.4 trillion in predicted gains, 95% of corporate AI projects aren't delivering expected results.
Declining improvements: Investments in new LLMs show declining improvements. The path from today's AI to artificial general intelligence, that would justify these valuations, stays unclear.
Circular spending: One example: Microsoft invests $13B into OpenAI → OpenAI buys Microsoft cloud → Microsoft books it as revenue growth. Real growth requires enterprise adoption beyond these closed loops.
Why it isn't 2000
Cash flow funded: The Magnificent Seven fund AI investments from profitable core businesses with significant free cash flow. Unlike the dot-com era, failure won't bankrupt them.
Long-term horizon: These are infrastructure plays expected to deliver ROI over years and decades, not quarters.
Public vs. private risk: Listed companies (like Nvidia, Microsoft) have profitable models backing their valuations. The real risk: An estimated $800B in private credit markets funding companies (like OpenAI, Anthropic) still seeking profitability.
What Determines Success
Positive examples of succeeding AI projects have one thing in common: years of investment in data infrastructure, quality and expertise.
The reality behind the AI bubble discussion is much more complex than any post could reflect. But what matters: AI infrastructure investments are a marathon, not a sprint. The winners are already years into their journey.