History Doesn’t Repeat — But It Rhymes. Here’s What That Means for AI.

In my previous post, I introduced a pattern that has repeated across every major technology wave of the past 50 years: investment flows first to hardware, then to software, and finally—and most enduringly—to services.

A fair question surfaced in the response: How do we know the pattern will hold this time?

The honest answer is that we don’t—not with certainty. But history offers more than vague analogies. It offers specific, quantifiable evidence of how technology investment actually evolves. And two waves in particular—the internet and public cloud—provide instructive parallels for what’s unfolding with AI today.

The Internet Wave: Infrastructure First, Value Later

When we remember the dot-com era, we tend to think of Pets.com, astronomical valuations, and the spectacular crash of 2000. What we often forget is the massive physical infrastructure buildout that made the internet possible in the first place.

Before a single webpage could load at broadband speeds, someone had to lay the fiber.

The Hardware Phase: Digging Trenches

Between 1996 and 2001, telecommunications companies averaged more than $85 billion annually in U.S. communications infrastructure investment. Level 3, Global Crossing, WorldCom, Qwest—they raced to bury cable across continents and under oceans. Cisco and Lucent sold billions in networking equipment. Sun Microsystems shipped servers as fast as it could build them.

This was the hardware phase—capital-intensive, infrastructure-focused, and essential. Without it, nothing else could follow.

The investment was so aggressive it created massive overcapacity. When the bubble burst, many of these companies collapsed. Global Crossing and WorldCom filed for bankruptcy. Hundreds of billions in shareholder value evaporated.

But here’s what’s often overlooked: the infrastructure remained. The fiber in the ground didn’t disappear. It became the foundation for everything that followed—YouTube, Netflix, cloud computing, the mobile internet. The hardware phase overbuilt, overcorrected, and ultimately enabled two decades of innovation.

I have a personal connection to this story. My uncle actually discovered abandoned fiber in California that had originally been laid by Global West Network—infrastructure that cost an estimated $100 million to deploy, left sitting unused when the company collapsed. That fiber didn’t disappear. It eventually found its way into the networks we rely on today.

The Software Phase: Platforms Emerge

As bandwidth became abundant, value shifted to software. The mid-2000s brought web applications, e-commerce platforms, and early SaaS pioneers. Salesforce launched in 1999 and spent a decade proving enterprise software could be delivered over the internet. Google, Amazon, and Facebook built platforms that would become some of the most valuable companies in history.

The Services Phase: The Long Tail

By the late 2000s, the services opportunity had fully emerged. Enterprises needed help building websites, optimizing for search, managing digital marketing, integrating e-commerce. Agencies like Razorfish and Sapient built massive practices. Managed hosting became a serious business. Digital transformation consulting grew into a multi-billion-dollar category.

And here’s the key: the services phase didn’t end. It evolved, expanded, and eventually converged with the next wave.

The Cloud Wave: The Pattern Repeats—With a Twist

If the internet wave feels like ancient history, cloud computing offers a more recent and more precise parallel to AI.

The Hardware Phase: Hyperscale Infrastructure

When AWS launched in 2006, it was essentially a hardware play dressed in software clothing. Amazon had built massive data center infrastructure for e-commerce and realized it could rent that capacity to others.

The first decade of cloud was defined by infrastructure buildout. AWS, Azure, and Google Cloud invested tens of billions constructing data centers worldwide. Amazon’s CapEx spending grew from $3.8 billion in 2012 to over $60 billion by 2022—a large and growing share of which supported technology infrastructure.

The Software Phase: Up the Stack

As infrastructure commoditized, value migrated to software. The 2010s saw rapid evolution from IaaS to PaaS to SaaS. Kubernetes became the orchestration standard. Snowflake, Databricks, and other cloud-native companies built multi-billion-dollar businesses on hyperscaler infrastructure.

Traditional vendors—Microsoft, Oracle, SAP—raced to transform for cloud delivery. The ones who moved fast captured enormous value. The ones who hesitated lost relevance.

The Services Phase: Where We Are Now

Today, we’re firmly in the services-dominant phase of cloud.

Global cloud spending—a market that barely existed 15 years ago—reached nearly $600 billion in 2024.  But the fastest-growing segments aren’t raw infrastructure—they’re the services that help enterprises actually use it: migration, managed services, FinOps, security.

The systems integrators and managed services providers that built cloud practices in 2012-2015 are the ones dominating today. Timing mattered.

What Cloud Teaches Us About Cost vs. Growth

But here’s something that doesn’t get discussed enough when we talk about cloud: the companies that treated it purely as a cost reduction play missed the real opportunity.

Many enterprises approached cloud migration as “lift and shift”— take your existing applications, move them to AWS or Azure, and save money on data center costs. And yes, some achieved modest savings. But they missed the forest for the trees.

The transformational value of cloud wasn’t about reducing your existing infrastructure bill. It was about what the new economics made possible.

Cloud fundamentally changed the cost structure of innovation. The shift from CapEx to OpEx, from fixed to variable costs, meant the marginal cost of experimenting with a new application dropped dramatically. You didn’t need to provision servers for peak capacity and let them sit idle. You didn’t need a six-month procurement cycle to test an idea. You could spin up infrastructure in minutes, test a concept, and shut it down if it failed—paying only for what you used.

Netflix understood this.

When Netflix shifted from DVD-by-mail to streaming, they didn’t just migrate their existing infrastructure to the cloud to save money. They used AWS to fundamentally reimagine their business. The elasticity of cloud allowed them to scale globally almost overnight. They could handle massive demand spikes—think Friday night viewing peaks—without maintaining idle capacity the rest of the week. They could enter new markets without building data centers first. They could iterate on their product continuously.

Netflix didn’t use cloud to cut costs. They used it to build a $250 billion-plus company that transformed an entire industry.

Contrast that with the enterprises that spent years on lift-and-shift migrations, carefully replicating their on-premises environment in the cloud, and then wondered why their cloud bills weren’t lower than their old data center costs.

They were optimizing for the wrong thing.

The cloud winners—Netflix, Airbnb, Spotify, and the enterprises that truly transformed—used cloud as an engine for growth, not just efficiency. They built new products, reached new customers, and created business models that couldn’t have existed in the CapEx-heavy world of traditional infrastructure.

I made this point directly in a recent conversation with John Furrier at TheCUBE, and I think the Ensono example gets at something most companies are currently missing:

That shift — from monitor, react, repair to predict, prevent, optimize — is what AI-driven revenue growth actually looks like in practice. Not headcount reduction, but a fundamentally different service model, built on data we already had.

The AI Parallel: Are We Making the Same Mistake?

I see the same pattern emerging with AI—and it concerns me.

Listen to how most enterprises talk about AI today. The dominant framing is cost reduction: automating customer service to reduce headcount, streamlining document processing to cut labor costs, replacing manual tasks with AI agents.

There’s nothing wrong with efficiency. Cost reduction is a legitimate use case. But if that’s all we see in AI, we’re making the same mistake enterprises made with cloud.

The companies that will win with AI aren’t the ones cutting costs. They’re the ones growing revenue.

Think about what AI actually makes possible:

• Personalization at scale. You can now deliver individualized experiences to millions of customers without proportionally scaling your workforce. What new markets does that open?
• Products that couldn’t exist before. AI enables entirely new categories of products and services—from AI-powered diagnostics to generative design tools to intelligent assistants embedded in every workflow.
• Speed to market. AI dramatically compresses the time from idea to product. Code generation, automated testing, rapid prototyping—what could you build if development cycles were 50% shorter?
• Reaching underserved markets. AI can make it economically viable to serve customer segments that were previously too expensive to address—think personalized financial advice for mass-market customers, or healthcare guidance in regions with provider shortages.

The enterprises treating AI as “a way to do the same things with fewer people” are optimizing for the wrong outcome. The winners will be those who ask: What can we do now that we couldn’t do before?

What History Tells Us About What’s Coming

So what does this mean for AI? The parallels are striking.

Like the internet, AI is seeing massive infrastructure investment that may eventually prove excessive. NVIDIA’s GPUs are today’s equivalent of Cisco’s routers or Level 3’s fiber—essential, valuable, and potentially overbuilt relative to near-term demand.

For perspective, a recent Axios C-Suite newsletter framed the scale two powerful ways. First, it noted that JPMorgan, Bank of America and others now project total AI capex to top $1 trillion a year starting in 2027 (by my estimate, roughly six times in today’s dollars what telecom carriers spent annually to build out the internet’s physical infrastructure during the fiber boom.)

Second, it pointed out that a single year of that spending (again, in 2026 dollars) is roughly the Manhattan Project, the Apollo moon landings, the Interstate Highway System and the Human Genome Project combined. Those were the defining moonshots of the 20th century, spread across six decades. The AI buildout is about to spend that much every. Single Year.

Like cloud, AI is transitioning through phases faster than previous waves. The foundation models that seemed miraculous two years ago are already commoditizing. The software layer—AI platforms, MLOps tools, enterprise applications—is maturing quickly.

And like both waves, the services opportunity is just emerging.

Enterprises are moving from experimentation to production. They’re discovering that implementation requires integration with existing systems. That models require ongoing management. That costs require governance. That skills are scarce.

If history is any guide:

• The hardware phase will peak and correct. We may already be seeing early signs with compute capacity coming online.
• The software phase will produce category-defining companies, but also rapid commoditization
• The services phase will be larger, longer, and more broadly distributed than most expect

But there’s one more lesson from history, and it may be the most important:

The companies that use AI purely for cost reduction will capture a fraction of the value. The ones that use it to grow—new products, new markets, new capabilities—will define the next era.

For CFOs, that distinction has a direct translation: ROI models built only around cost reduction are leaving value on the table.

This is Part 2 of a multi-part series on AI investment patterns. Read the first post here. Next in the series: The CFO’s AI Challenge—navigating visibility, forecasting, and governance as AI spend accelerates.

Frequently Asked Questions

What does history suggest about AI investment?

History suggests AI may follow the same investment pattern as earlier technology waves: infrastructure first, then software, and finally services that help enterprises implement, manage, and optimize the technology.

How is AI similar to the internet and cloud waves?

Like the internet and cloud, AI is beginning with massive infrastructure investment. Over time, value may shift toward software platforms, enterprise applications, and services that make AI useful in production.

Why is cost reduction the wrong way to think about AI?

Cost reduction is a valid AI use case, but it is not the full opportunity. The larger value may come from using AI to create new products, enter new markets, personalize customer experiences, and grow revenue.

What lesson does cloud computing offer for AI?

Cloud showed that companies gained the most value when they used new technology to innovate and grow, not simply to reduce infrastructure costs. The same lesson applies to AI investment.

How should CFOs evaluate AI ROI?

CFOs should evaluate AI ROI through both efficiency and growth. Strong AI business cases should include cost savings, but also revenue expansion, faster innovation, new capabilities, and competitive advantage.