I've been building technology businesses since the 1990s, and I've survived enough bubbles to recognise the warning signs. The dot-com crash taught me that transformative technology and sustainable valuations are two entirely different conversations. Today, as I watch the AI market evolve, I see something both familiar and fundamentally different unfolding.
Every major bubble I've witnessed shares a common architecture. In the dot-com era, we convinced ourselves that clicks would translate to cash, eventually. During the 2008 financial crisis, we believed housing prices only went up and that securitization had eliminated risk. The crypto boom promised decentralization would replace traditional finance, right up until it didn't.
Each time, the technology or innovation was real. The internet did transform commerce. Digital assets have found legitimate use cases. But in each instance, the financial structure built around the innovation collapsed long before the innovation itself proved its worth. Amazon and Google survived the dot-com crash. Plenty of genuinely useful internet companies did not.
What makes the current AI moment particularly precarious is the mismatch between the investment timeline and the revenue reality. We're witnessing an unprecedented capital deployment into infrastructure with extraordinarily uncertain payback periods.
Consider the financial architecture: companies are taking on long-duration debt to finance data centres, chip fabrication facilities, and energy infrastructure that may take years to generate returns. They're betting these assets will remain competitive and relevant even as the technology evolves at breakneck speed. Meanwhile, the demand signals they're receiving are heavily circular, with much of the AI investment fuelled by other AI companies buying AI services to build AI products.
This is different from previous bubbles. The dot-com bubble was primarily an equity story. Venture capital poured into companies with minimal fixed assets. When the music stopped, these companies simply ceased to exist, but they didn't leave behind billions in stranded infrastructure. The 2008 crisis involved real assets (housing), but those assets retained fundamental utility and value even after repricing.
AI infrastructure represents something else entirely: massive, fixed capital investments in rapidly depreciating technology. A state-of-the-art GPU facility today may be obsolescent in three years. A $10 billion data centre built for today's models may be unsuitable for tomorrow's architectures. This creates a unique vulnerability where companies must continue running forward simply to avoid the write-downs from standing still.
The energy requirements alone tell the story. We're building power infrastructure to support computational demands that didn't exist five years ago, financed with debt that assumes those demands will not only continue but accelerate. If adoption curves flatten or if more efficient architectures emerge, that infrastructure becomes a burden rather than an asset.
Here's what keeps me up at night: enterprise AI adoption is following a very different trajectory than consumer internet adoption did. In the dot-com era, consumers eagerly adopted email, search, and e-commerce because the value proposition was immediate and obvious. Today's enterprise AI implementations are more complex, requiring significant organisational change, retraining, and infrastructure investment before value can be realised.
Many companies are experimenting with AI, but moving from pilot programs to full-scale deployment is proving slower and more challenging than the investment thesis assumes. This creates a dangerous gap between the pace of infrastructure buildout and the pace of revenue generation.
AI-related stocks have driven nearly half of market gains over the past year. Whether you've consciously chosen to bet on AI or not, your portfolio likely has significant exposure.
Unlike previous bubbles, the AI reckoning will likely involve four things at once. Stranded physical assets at unprecedented scale — we're not just writing down goodwill and intangibles, but potentially billions in specialised hardware and facilities with limited alternative uses. Sovereign and strategic implications — governments have made AI infrastructure a matter of national competitiveness, which may cushion the fall but also creates moral hazard. Energy and environmental constraints becoming binding — previous tech buildouts didn't face hard physical limits on power generation and cooling capacity; AI does. And talent markets that may reverse sharply, as the premium paid for AI expertise assumes sustained demand that will close quickly and painfully once projects get cancelled and budgets contract.
I remain convinced that AI will transform the economy. The technology is real. The applications will emerge. But I'm equally convinced that the current financial architecture around AI is unsustainable. Short-duration assets financed with long-duration debt, circular demand signals masquerading as product-market fit, and valuations that assume exponential growth curves continuing indefinitely are all heading toward repricing.
The companies with genuine competitive advantages, sustainable unit economics, and balance sheets that can weather a valuation reset will emerge stronger. Those relying on continued access to cheap capital and ever-rising valuations will struggle. The pattern holds across every bubble I've seen: the technology finds its place, but the financial excess gets wrung out, often brutally.
Understanding the fracture points before they break is how we position for what comes next. The question isn't whether AI will transform business and society. The question is whether the financial structures being built today will survive long enough to see it happen, and whether we'll recognise the warning signs in time to act. I've seen this movie before. The ending is rarely what the protagonists expect.