Why the AI bubble metaphor is breaking down
The word bubble has become shorthand for the frenetic surge in interest, investment and headlines around artificial intelligence since ChatGPT’s public debut on November 30, 2022. That launch, coupled with OpenAI’s March 14, 2023 release of GPT-4, sent startups, public markets and legacy tech firms racing to rewire products and balance sheets around large language models and generative AI.
But calling it a single bubble risks flattening a far more complex picture. Instead of one transient mania that inflates and pops, the AI story looks like overlapping cycles: model breakthroughs, infrastructure buildout, and real-world adoption. Each cycle has different time horizons, economics and risk profiles.
Three overlapping cycles to watch
1. Models and algorithmic progress
Model breakthroughs drive headlines and user enthusiasm. Advances from transformer architectures to scaling laws powered products like ChatGPT and Google Bard and set off massive demand for developer attention and consumer trials. The immediate risk here is hype: models promise broad capabilities before most businesses know how to integrate them cost-effectively.
2. Infrastructure and compute economics
The second cycle is infrastructure. The surge in model size and inference demand created a multiyear buildout in GPUs, datacenters and specialized silicon. Nvidia, whose GPUs became the de facto platform for many training and inference tasks, surpassed $1 trillion in market value in 2023 as cloud and enterprise customers raced to secure capacity. Infrastructure is capital intensive and lumpy; it can sustain long-term value if utilization rises, but it can also leave overhang if demand softens.
3. Adoption and product-market fit
Finally, adoption is the slowest and most durable cycle. This is where regulatory constraints, enterprise sales cycles and human workflows determine whether AI delivers economic value. McKinsey estimated that AI could add up to about $13 trillion to global GDP by 2030, but that potential depends on real adoption across industries, not just proof-of-concept demos.
Implications for investors, founders and policymakers
Thinking in cycles changes the signals you track. In the models cycle, look at published benchmarks, open-source momentum and developer engagement. In infrastructure, monitor utilization rates, spot pricing and supply constraints for H100 and similar accelerators. For adoption, measure revenue retention, gross margins on AI-powered features, and the time it takes for automation to translate into cost savings or new revenue.
For venture capital, that means differentiating between companies riding model hype and those building durable distribution, data moats or hardware-backed advantages. For public investors, it means separating headline-driven valuations from indicators like free cash flow and margin sustainability.
Expert perspectives
As AI pioneer Andrew Ng has long put it, ‘AI is the new electricity’ — a reminder that the tech can be foundational but also uneven in how and when it rewires industries. An early-stage AI investor summarized the new framing this way: ‘Treat the recent surge less like a single spike and more like an overlay of booms. Betting on models alone is different from betting on scalable infrastructure or on a company that can actually monetize AI with real customers.’
Analysts at major banks and research firms also emphasize regulatory risk. The European Union reached a provisional agreement on the AI Act in December 2023, signaling tougher rules on high-risk systems. That regulatory backdrop will shape adoption timelines for sectors like healthcare and finance, where compliance matters more than raw model performance.
What to watch next
Key signals that distinguish transient hype from structural change include sustained enterprise adoption metrics, improvements in cost-per-inference, and more diverse hardware ecosystems that reduce single-vendor risk. Watch cloud providers’ reported AI margins, chip vendors’ supply plans, and startups that show real net-new revenue attributable to AI features.
Related coverage worth digging into includes our reporting on Nvidia’s GPU strategy, the EU AI Act’s implications for compliance, and VC funding trends across generative AI startups.
Conclusion: a more useful lens
Reframing the AI bubble as overlapping cycles tempers both pessimism and unchecked exuberance. It narrows the question from ‘will AI crash’ to ‘which layers are mispriced, and for how long’. That shift helps founders prioritize durable advantages, helps investors calibrate risk across very different business models, and helps policymakers target interventions where they matter most.
Ultimately, some parts of the market may correct quickly; others will take years to mature. Distinguishing between hype and durable value is the hard work ahead.
About the Author
