AI is Not a Bubble

Understanding the Infrastructure vs Speculation Distinction

Here’s a topic that annoys me considerably. The mis-information around this is robbing you of a competitive advantage, which in turn frustrates my business of bringing the power of AI to your business.

Your board is asking about the AI bubble, and the clever answer—‘it’s a bubble but the technology is real’—is precisely the wrong response to give them. But yet I see this becoming a trend. I hear it on podcasts and read it in your post comments.

This sophisticated hedge serves no one. It’s the kind of nuanced position that sounds intelligent at dinner parties but paralyses decision-making in boardrooms.

Your directors don’t need commentary that makes them feel sophisticated whilst accomplishing nothing. They need clarity on whether the hundreds of millions being invested in AI represent strategic infrastructure or speculative excess.

The distinction matters enormously. To me and to you!

When railways consumed 4% of US GDP in 1872, observers cried bubble. When electricity infrastructure was built out, sceptics saw waste. Yet these investments created the foundation for entirely new economies.

The question your board should actually be asking isn’t “Is AI a bubble?” but rather “Are we building permanent capability or chasing temporary trends?”

The Illusion

Every transformative infrastructure build looks like a bubble to those who can’t see the future it’s creating. I can.

I dream about it at night. I take it onto the cricket pitch (village cricket). I eat it for breakfast, lunch and dinner. My poor wife, it is the only thing I ever talk about.

Consider what’s actually happening right now. Microsoft, Amazon, and Google are spending unprecedented sums—but they’re using their own cash, not leveraging debt like the telecom bubble of the 1990s. This isn’t financial engineering; it’s engineering, full stop. These companies generated over $200 billion in free cash flow last year. They’re not gambling with borrowed money; they’re investing earnings into permanent infrastructure.

The numbers tell a remarkable story. One-third of current US GDP growth traces directly to data centre construction. This isn’t speculation—it’s physical infrastructure being built at a scale we haven’t seen since the highway system. These data centres aren’t factories for a single product; they’re the computational substrate for the next economy. Just as roads enabled commerce, telecommunications enabled globalisation, and the internet enabled digital transformation, AI compute infrastructure will enable capabilities we’re only beginning to imagine.

Here’s what makes this different from historical bubbles: compute is becoming a utility. Not metaphorically, but literally. The US government now classifies AI compute as critical infrastructure. When Microsoft builds a data centre, they’re not betting on a single application or business model. They’re building capacity that will serve thousands of use cases, most not yet invented.

But here’s the controversy: Your ChatGPT subscription isn’t AI investment. Neither is buying AI software licenses. Real AI investment means building or securing permanent computational capacity and the expertise to deploy it. Everything else is just consumption.

How does your company think about infrastructure? Are you building permanent capability, or are you just renting temporary access?

The Speculation Trap Your Competitors Avoid

The enterprises spending £130 million on AI this year aren’t gambling—they’re building competitive moats.

KPMG’s latest data reveals something remarkable: enterprise AI investment is accelerating, not hesitating. The average billion-pound company plans to invest £130 million in the next 12 months, up 47% from last year. These aren’t speculative punts by exuberant startups. These are careful infrastructure decisions by companies that survived 2008, COVID, and countless other crises.

The token shortage tells the real story. Major enterprises can’t secure enough AI compute capacity. They’re signing contracts for data centres that don’t exist yet, committing capital before infrastructure is even built. When demand exceeds supply this dramatically, you’re not in bubble territory—you’re in the early stages of a platform shift.

Meta’s experience provides the template. Their AI implementation improved advertising conversions by 3-5%. That might sound modest until you realise it represents billions in additional revenue from existing operations. No new products, no new markets—just pure operational improvement. One FTSE 100 financial services firm we work with identified £50 million in annual savings from document processing alone, with implementation costs recovered in eight months.

The “curiosity revenue” critique—that companies are just experimenting—misses the point entirely. Of course they’re experimenting. That’s how enterprises learn to deploy new infrastructure. The difference between curiosity and commitment isn’t the initial spending; it’s whether value compounds over time. And the data shows it is.

How much of your competitor’s AI spend is actually reaching customers already? If you don’t know, you’re already behind.

The Three-Year Advantage

Your CFO thinks three-year depreciation is a bug. It’s actually the feature that protects your investment.

GPUs age in dog years—their useful life for frontier applications is perhaps three years. Compare this to railway tracks lasting generations or fibre optic cables still carrying data decades later. This terrifies traditional infrastructure investors. It should comfort enterprise executives.

Why? Because fast depreciation enforces a discipline that previous infrastructure booms lacked. Railway companies could hide bad investments for decades, sustained by long asset lives and creative accounting. The telecom bubble’s excess fibre lay dark underground for years, enabling denial. But three-year depreciation means AI investments must prove themselves quickly or fail fast.

This isn’t theoretical. One major bank discovered their initial AI implementation approach was wrong within six months, not six years. They pivoted, redeployed resources, and found success with a different architecture. Under traditional IT investment cycles, they’d have spent years committed to the wrong path. The fast feedback loop saved them millions.

The depreciation timeline also aligns with value capture windows.

Microsoft’s Azure learned this lesson: between 2015 and 2018, their capital expenditure represented 70-90% of revenues. It looked insane at the time. But that infrastructure investment created durable competitive advantage that compounds annually. The 70% rule—expect to invest 70% of early revenues in infrastructure—has become the template for AI deployment.

This changes everything about vendor negotiations. When infrastructure ages rapidly, vendors can’t lock you into decade-long contracts. Flexibility becomes standard. Risk shifts from buyers to builders. The entire procurement model adapts to rapid innovation cycles.

Traditional IT investment, with its five-to-seven-year depreciation schedules and lengthy implementation cycles, is actually riskier than AI infrastructure investment. At least with AI, you know quickly if you’re wrong.

Your System for Confident Investment

Stop asking ‘Is AI a bubble?’ Start asking ‘Are we building infrastructure or chasing speculation?’

I have a ‘telemetry’ system for this economy to test for the AI bubble. As a hobbyist economist, I think it is rather useful. I have five monitors for bubble detection which translates directly to enterprise decision-making:

1. Economic Strain Monitor: Is AI investment distorting your industry? Green: 0.9% of GDP
   1. What it measures: How much of the total economy’s resources are being pulled into AI investment.
   2. The data: Currently, AI infrastructure spending represents about 0.9% of US GDP. The “2% danger zone” comes from historical analysis - when railway spending hit 4% of GDP in 1872, it crashed. Telecom hit 1% before its bubble burst.
   3. Why it matters for enterprises: If too much of the economy depends on one sector, any slowdown cascades everywhere. Your customers, suppliers, and talent pool all become vulnerable to a single shock.
2. Revenue Reality Monitor: Are returns materialising? Green: 300-500% growth
   1. What it measures: Whether actual money is flowing into AI companies, not just investment.
   2. The data: Gen AI revenues are growing 300-500% annually. By comparison, railways grew 22% annually before the 1873 crash, telecoms grew 16% before their bubble burst.
   3. Why it matters for enterprises: Rapid revenue growth suggests real demand, not just speculation. It means customers are actually paying for AI services, validating the infrastructure investment.
3. Valuation Sanity Monitor: Are prices detached from reality? Green: P/E of 32
   1. What it measures: How many years of current profits investors are willing to pay for (Price-to-Earnings ratio).
   2. The data: The NASDAQ’s P/E ratio is 32, meaning investors pay 32 times annual earnings for stocks. During the dot-com bubble, it hit 72. Some internet stocks had implied P/E ratios of 605.
   3. Why it matters for enterprises: When valuations detach from reality, it suggests speculation rather than fundamental value. Current valuations, whilst elevated, aren’t in bubble territory.
4. Funding Quality Monitor: Who’s paying? Green: Strong balance sheets
   1. What it measures: Whether investments are funded by sustainable sources or risky debt.
   2. The data: Microsoft, Amazon, Google are using their own cash flows (over $200B generated last year) rather than borrowing. Compare to telecoms which doubled and quadrupled their debt before crashing.
   3. Why it matters for enterprises: Debt-funded booms collapse quickly when revenues disappoint. Cash-funded building can weather downturns.
5. Industry Strain Monitor: Is investment racing ahead of revenue? Yellow: 6x investment to revenue
   1. What it measures: The ratio between capital expenditure and actual revenues in the AI sector.
   2. The data: 370 billion in global data centre CapEx against 60 billion in Gen AI revenues = 6x ratio. Railways peaked at 2x, telecoms at 4x before their crashes.
   3. Why it matters for enterprises: This is the one warning flag. Investment is running far ahead of current revenues. However, given the 300-500% revenue growth rate, this gap could close quickly.

Three questions every board should ask quarterly:

1. What percentage of our AI investment builds permanent capability versus temporary access?
2. Are we seeing compound value from our AI deployments, or just one-time gains?
3. How does our AI infrastructure investment compare to our competitors’?

Warning signals that actually matter: If AI investment approaches 2% of GDP, if revenue growth drops below 100% annually, if companies start leveraging heavily to fund AI infrastructure, or if the P/E ratio exceeds 50—then reassess. We’re nowhere near these thresholds.

Your investment staging strategy should follow infrastructure maturity. Start with proven applications (document processing, customer service automation). Build towards frontier capabilities (autonomous agents, complex reasoning). Always maintain optionality for emerging use cases.

Which of these five gauges is your organisation actually monitoring? Most boards track none, leaving them vulnerable to both bubble hysteria and competitive disruption.

The Permanent Revolution

The bubble question is a distraction from the real issue: AI is becoming the infrastructure of competitive advantage.

History repeats: before personal computers transformed business, experts debated whether computing was a bubble. Before the internet became essential, analysts questioned web investments. Before mobile ate the world, smartphone adoption seemed speculative. The pattern is consistent—infrastructure debates always precede transformation.

The enterprises building AI capability now versus those waiting for “clarity” are writing the next chapter of competitive dynamics. Coca-Cola didn’t wait for bubble clarity before adopting refrigeration. Amazon didn’t wait for consensus before building cloud infrastructure. The winners move while others debate.

Infrastructure investments compound; speculation evaporates.

Every AI model you train improves your data quality. Every process you automate frees resources for innovation. Every capability you build becomes a foundation for the next. This isn’t like buying tulip bulbs or flipping houses. It’s like building roads—the railways—each one makes the next more valuable.

In 2030, we won’t debate whether AI was a bubble. We’ll wonder how anyone could have missed the infrastructure revolution happening in plain sight.

So I pose for your consider the following, it isn’t whether AI will transform your industry—it’s whether you’ll be driving that transformation or reacting to it.

The Choice Before You

The enterprises that will dominate 2030 will be making infrastructure decisions today, not waiting for bubble clarity.

The infrastructure versus speculation lens changes everything.

It transforms AI from a risky bet into a strategic imperative. It shifts the conversation from “Can we afford to invest?” to “Can we afford not to?”

The cost of waiting for “bubble resolution” compounds daily.

While you debate, competitors build. While you analyse, they implement. While you hedge, they commit. The six months you spend waiting for clarity is six months of capability your competitors are accumulating.

Your immediate action framework:

• Assess: Audit current AI spending—what builds capability versus what’s merely consumption?
• Plan: Define your infrastructure strategy—build, buy, or partner?
• Build: Start with proven applications, maintain optionality for emerging capabilities

This isn’t about being reckless.

It’s about recognising that the biggest risk isn’t investing in AI infrastructure—it’s watching your competitors build permanent advantages while you wait for perfect information that will never come.

Let’s move beyond bubble rhetoric. Connect with me to discuss how your organisation can distinguish AI infrastructure from speculation. Share this if you’re tired of bubble fear paralysing strategic decisions.

For a detailed framework on evaluating AI infrastructure investments, including our monitoring system and the 5 thresholds, including an implementation roadmap, reach out directly. The conversation your board needs to have isn’t about bubbles—it’s about building the infrastructure for competitive advantage.

#EnterpriseAI #AIStrategy #DigitalTransformation #AIInvestment #FutureOfWork