The AI bubble: 7 trillion looking for returns

TL;DR

• $7 trillion projected in AI infrastructure by 2030, yet 95% see no returns
• Who wins for sure: NVIDIA (GPUs), cloud providers, consultants
• Who loses: companies buying without understanding, undifferentiated startups, late investors
• Won’t disappear, but correction is coming: think dot-coms (Amazon survived, Pets.com didn’t)

---

The scary numbers

• McKinsey projects 7 trillion dollars in AI infrastructure investment by 2030
• Meta, Microsoft, and Amazon have spent tens of billions this year alone
• 95% of organizations see no measurable returns yet

7 trillion going in. 95% with no visible return.

Does that smell like a bubble? It does to me too.

Who’s winning (for sure)

NVIDIA

Every data center needs GPUs. NVIDIA has the monopoly. They win no matter what.

It’s like selling pickaxes during the gold rush. Doesn’t matter if you find gold or not. The pickaxe seller always gets paid.

Cloud providers (AWS, Azure, GCP)

All that AI needs servers. They have them. They win on usage, not results.

If your AI model works → you pay. If your AI model fails → you still pay.

Consultants

“We’ll help you implement your AI strategy.”

They charge by the hour. If implementation takes longer because it doesn’t work, they charge more.

Who’s losing (probably)

Companies buying without understanding

• Buy 6-figure “enterprise AI” licenses
• Don’t have clean data
• Don’t have technical teams
• Expect magic

Money goes out. Results don’t come in. They blame “AI” instead of their strategy.

AI startups without differentiation

“We’re like ChatGPT but for [industry].”

OpenAI and Anthropic can do the same thing tomorrow. And they will.

If your only advantage is a wrapper over GPT, you have no advantage.

Late investors

Those who got in 2022-2023 are fine.

Those getting in now at 100x revenue valuations… good luck.

The bubble signals

1. Nonsense valuations

Companies with no revenue valued in billions. “But the potential…”

2. Everyone’s an expert

In 2021 everyone was a crypto expert. In 2025 everyone’s an AI expert.

3. Corporate FOMO

“If we don’t add AI, we’ll fall behind.”

That’s not strategy. That’s panic.

4. Incumbents winning more than innovators

Microsoft, Google, Amazon are winning more than pure AI startups.

When incumbents win the “revolution,” it’s not a revolution. It’s a feature.

Will it burst?

Probably yes. But not all at once.

What will happen:

• Correction in startup valuations
• Consolidation (big ones buy small ones)
• Many companies will abandon “failed” AI projects
• Those who persist with real strategy will gain advantage

What will NOT happen:

• AI doesn’t disappear
• Real use cases keep working
• Built infrastructure stays

It’s like the dot-coms. Amazon and Google survived. Pets.com didn’t.

What to do if you’re a normal company

1. Don’t panic

You don’t need “AI” tomorrow. You need to solve business problems.

Sometimes the solution is AI. Sometimes it’s a well-made Excel.

2. Start with the problem, not the technology

Bad: “We want to use AI” → for what?

Good: “We want to reduce customer response time” → how? → maybe AI helps

3. Realistic budget

If you can’t afford to fail with the project, don’t do it.

AI projects have high failure rates. Build that into the budget.

4. Measure results, not activity

“We implemented 3 AI models” isn’t a result.

“We reduced process time from 2 days to 2 hours” is.

Conclusion

7 trillion will go into AI.

Part of that money will generate real value. Part will evaporate.

NVIDIA and cloud providers win no matter what. They’re selling the pickaxes.

The companies that win will be those who understand what problem they’re solving, not those with more “AI” in their PowerPoint.

If it smells like a bubble, it probably is. But bubbles also leave useful infrastructure when they pop.

The fiber optic from the dot-coms is still there. The data centers from this bubble will stay too.

The question is: will you be on the side of those building real value, or on the side of those buying smoke?

---

For a broader view of where the industry is heading, read my AI predictions for 2026.