The AI Boom's Dark Fiber Moment: When Compute Becomes a Commodity

Machine-augmented insight on the inevitable AI bust and the innovation renaissance that follows

Executive Summary

The current AI boom bears striking parallels to the fiber optic boom of the late 1990s. Just as WorldCom’s false claim that “internet traffic doubles every 100 days” drove massive overinvestment in fiber infrastructure, today’s AI boom is fueled by equally speculative narratives about exponential growth and imminent AGI. The inevitable bust will create vast amounts of “dark compute”—underutilized AI processing power that will become the foundation for an unprecedented wave of innovation, much like how dark fiber enabled Netflix, YouTube, and the entire Web 2.0 revolution.

We live in the age of AI infrastructure mania. Hyperscalers are pouring hundreds of billions into data centers, NVIDIA’s market cap has soared past 3 trillion dollars, and every startup claims to be “AI-native.” The parallels to another infrastructure boom are impossible to ignore—the fiber optic frenzy of the late 1990s that ended in spectacular collapse, only to birth the digital world we inhabit today.

As someone who has witnessed multiple technology cycles, the pattern is unmistakable. We are in the midst of a classic boom-bust cycle, and understanding the lessons from the fiber era provides a roadmap for what comes next.

The Fiber Boom Playbook

The telecommunications boom of 1995-2000 was driven by a single, powerful narrative: internet traffic was doubling every 100 days. This statistic, popularized by WorldCom, justified a debt-fueled infrastructure buildout of unprecedented scale. Over 500 billion dollars was invested in laying millions of miles of fiber optic cable, with 800 billion dollars in M&A activity in 1999 alone.

The logic was seductive. Network effects meant that the largest networks would capture disproportionate value. Equipment vendors like Cisco, Lucent, and Nortel provided vendor financing to accelerate purchases. The Telecommunications Act of 1996 had deregulated the industry, creating a gold rush mentality among new entrants.

There was just one problem: the fundamental premise was false. Internet traffic was actually doubling once per year, not every 100 days. When reality collided with expectations in 2000, the result was catastrophic. Half a million people lost their jobs, 7 trillion dollars in market value evaporated, and 23 major telecom companies went bankrupt simultaneously.

But the most important outcome was not the destruction—it was what remained. Billions of dollars worth of fiber optic networks sat unused, creating what became known as “dark fiber.” This stranded infrastructure would become the foundation for the next wave of innovation.

The AI Boom Mirror

Today’s AI boom follows an eerily similar script:

Core Infrastructure • Fiber Boom: Fiber Optic Cable (Bandwidth) • AI Boom: GPUs / AI Accelerators (Compute)

Key Enabler • Fiber Boom: Deregulation (Telecommunications Act of 1996) • AI Boom: Breakthroughs in deep learning (Transformer architecture)

Demand Narrative • Fiber Boom: “Internet traffic doubles every 100 days” • AI Boom: “AGI is just around the corner” / Exponential AI growth

Primary Investment • Fiber Boom: Laying millions of miles of fiber • AI Boom: Building massive AI data centers and training larger models

Key Equipment Suppliers • Fiber Boom: Cisco, Lucent, Nortel • AI Boom: NVIDIA, AMD, Intel

Infrastructure Builders • Fiber Boom: WorldCom, Global Crossing, Level 3 • AI Boom: Google, Amazon (AWS), Microsoft (Azure), Meta

Speculative Startups • Fiber Boom: Hundreds of dot-coms and ISPs • AI Boom: Thousands of AI-native startups and feature companies

The parallels extend beyond surface similarities. Just as the fiber boom was driven by vendor financing from equipment manufacturers, today’s AI boom is fueled by cloud credits and compute subsidies from hyperscalers eager to sell their infrastructure. The same network effects logic that justified telecom consolidation now drives the race to build the largest AI models and data centers.

The Coming AI Bust and “Dark Compute”

The AI bust is not a matter of if, but when. The massive overinvestment in AI compute capacity, driven by speculative fervor and inflated expectations, will eventually collide with the reality of sustainable demand. When it does, the result will be the creation of vast amounts of “dark compute”—underutilized AI processing power that will trade at commodity prices.

This dark compute will emerge from three sources:

Hyperscaler Overbuilding: Companies like Google, Amazon, and Microsoft are engaged in an arms race to build AI infrastructure. When demand growth inevitably slows, they will be left with significant excess capacity that must be monetized at marginal cost.

Startup Failures: The current AI landscape is crowded with companies building similar products. Market consolidation will leave behind compute resources that need new homes and new purposes.

Hardware Depreciation: The rapid pace of AI hardware innovation means today’s cutting-edge chips will be tomorrow’s commodities, further driving down the cost of intelligence.

The Innovation Renaissance

The most important lesson from the fiber boom is that the true transformative potential of infrastructure is often realized not during the boom, but in the period of abundance that follows the bust. When bandwidth became cheap and plentiful after 2002, it enabled entirely new categories of services:

Netflix could only pivot to streaming in 2007 because the cost of bandwidth had collapsed. YouTube launched in 2005 with the premise that user-generated video content was economically viable. Amazon Web Services emerged in 2006, leveraging not just cheap bandwidth but the broader infrastructure overcapacity. The entire Web 2.0 movement—characterized by social media platforms like Facebook—was enabled by the dramatic reduction in infrastructure costs.

When AI compute becomes abundant and cheap, we can expect a similar explosion of innovation. The companies that will define the next era will not be those building incrementally better models, but those that can creatively leverage commodity intelligence to build entirely new categories of products and services.

Consider what becomes possible when the cost of intelligence approaches zero:

Hyper-Personalization at Scale: AI-powered services that are deeply and continuously personalized to each individual user, from adaptive education platforms that tailor curriculum in real-time to personal assistants that manage every aspect of your digital life.

Ambient Computing: Intelligence embedded in every device and environment, creating seamless and intuitive experiences. Smart homes that anticipate your needs, truly autonomous vehicles that communicate with city infrastructure, retail environments that offer completely personalized shopping experiences.

Generative Everything: The ability to generate not just text and images, but complex software, scientific hypotheses, and even physical products. AI systems that can design and code entire applications from natural language prompts, drug discovery platforms that simulate molecules in silico, generative design tools that create optimized physical objects.

New Forms of Entertainment: Immersive experiences generated in real-time, with truly intelligent NPCs, dynamic procedurally generated worlds, and entirely new forms of interactive storytelling that adapt to each viewer.

Democratization of Science: Powerful AI tools accessible to anyone for conducting research, analyzing large datasets, automating experimentation, and tackling humanity’s most pressing problems.

The Strategic Imperative

For investors, entrepreneurs, and policymakers, the lesson is clear: prepare for the bust, but position for the renaissance. The companies that will thrive in the post-bust era will be those that can leverage abundant, cheap compute to create entirely new forms of value.

This means focusing not on building bigger models or more data centers, but on imagining what becomes possible when intelligence is as cheap and abundant as bandwidth became after the telecom crash. The winners will be those who can see beyond the current hype cycle to the transformative potential that lies on the other side of the inevitable correction.

The AI boom will be exciting, but the post-bust era of dark compute will be truly revolutionary. Just as the fiber glut gave birth to the digital economy we know today, the coming AI capacity glut will unleash innovations we can barely imagine.

The question is not whether the AI bust will come—it’s whether we’ll be ready to capitalize on the abundance that follows.

This analysis represents a collaboration between human insight and machine execution, exploring the intersection of technology cycles, market dynamics, and innovation patterns. As always, use discretion and independent judgment when engaging with these ideas.