Michael Novogratz Calls AI the Biggest Bubble of Our Lifetimes

Key Takeaways

• Michael Novogratz says AI valuations are overheated, but the investment cycle has not reached its dramatic conclusion.
• Spending on chips, data centers, cloud capacity, and AI systems continues to rise despite uncertain enterprise returns.
• Business leaders face growing pressure to connect AI experimentation with measurable financial performance and stronger governance.

Michael Novogratz has joined the expanding group of prominent investors warning that artificial intelligence markets may be running too hot. Speaking at the Greenwich Economic Forum, he called AI "the biggest bubble of our lifetimes," while making a less bearish point alongside the warning: investors should still participate because the cycle has not reached its dramatic end.

That combination captures the strange state of the AI economy in 2026. Valuations and capital commitments increasingly resemble bubble conditions, yet demand for computing infrastructure remains substantial. Enterprises are also pushing ahead with deployments, even when the financial results are difficult to measure.

In other words, this is not simply a story about speculative enthusiasm detached from commercial activity. There is real spending, real adoption, and real capacity being built. The harder question is whether future revenue and productivity gains can justify the speed and concentration of that investment.

According to Gartner, global AI spending is forecast to reach $2.59 trillion in 2026, up 47% year over year. AI infrastructure accounts for roughly $1.43 trillion of that total. Those figures show how much of the current cycle is tied to the physical and cloud foundations required to train, host, and operate increasingly large models.

NVIDIA, Microsoft, and OpenAI sit near the center of this investment loop. NVIDIA supplies chips used across AI data centers. Microsoft provides cloud infrastructure while investing heavily in model access and enterprise distribution. OpenAI supplies foundation models that have helped create demand for both computing capacity and new software products.

Infrastructure can be valuable even when the companies financing it overestimate near-term demand. Railroads, telecommunications networks, and internet infrastructure all produced lasting economic assets while inflicting losses on some investors. AI could follow a similarly uneven path, with useful capacity surviving a broader repricing.

The amount of capital entering the market makes that possibility difficult to dismiss. Stanford HAI reported that global corporate AI investment reached approximately $581.7 billion in 2025, an increase of 130% from the previous year. Such concentrated flows can accelerate technical progress, but they can also encourage duplicated infrastructure, generous startup financing, and business plans built around optimistic adoption assumptions.

Enterprise results remain mixed. McKinsey found in its 2025 survey that 62% of organizations were experimenting with AI agents, while only 39% reported measurable EBIT impact. That gap is central to Novogratz's warning. Organizations may be using AI, but usage alone does not establish that the investment will generate adequate returns.

For technology executives, the distinction between experimentation and production economics is becoming more important. A pilot can demonstrate that an AI agent summarizes documents, writes code, or handles customer inquiries. Production deployment introduces other costs: model inference, data preparation, security controls, human review, system integration, monitoring, and regulatory compliance.

Can revenue growth or labor productivity consistently outrun those expenses? In some applications, perhaps. Across an entire enterprise portfolio, the evidence is still developing.

That said, executives do not need to choose between uncritical enthusiasm and abandoning AI. A more measured approach would tie each project to a defined operational metric, such as reduced handling time, higher conversion, fewer processing errors, or lower software maintenance costs. Programs that cannot show progress can be redesigned or closed before they turn into permanent spending commitments.

Governance also matters more as experiments become operational systems. Companies increasingly need documented accountability for model selection, data use, testing, human oversight, and incident response. These controls may slow some deployments, but they can also expose weak business cases before those projects scale.

Novogratz's position is therefore less contradictory than it initially sounds. A market can be overheated and still move higher. It can also fund infrastructure that becomes economically important after valuations fall. For businesses, the practical challenge is not predicting the exact moment sentiment changes. It is building AI programs that remain useful when capital becomes more selective, growth assumptions are questioned, and measurable returns matter more than participation itself.