Brace for a Crash Before the Golden Age of AI

AI is in the midst of its manic “installation phase”—a period marked by hyper-investment, speculative overreach, and financial instability. Tech giants like Google, Amazon, Microsoft, and Meta are pouring an eye-watering $750 billion into AI infrastructure over the next two years, while Morgan Stanley projects a staggering $3 trillion in global AI infrastructure spend by 2029.

Yet beneath the frenzied capex, the fundamentals are shaky. A recent MIT report found that a whopping 95% of surveyed companies are seeing no return on their generative AI investments. Historical analogies, drawn from Carlota Perez’s study of previous technological revolutions, suggest that a crash often precedes any transformative "golden age"—a period of sustained productivity and social benefit.

So what does this mean for the security and enterprise community? First, financial volatility could expose AI-powered systems to heightened cyber risk during economic downturns. Budget cuts may strain security controls, patching timelines may slip, and monitoring for prompt-based exploits could fall by the wayside. In worst cases, speculative infrastructure investments might collapse, leaving critical AI-driven services under-resourced or unavailable when needed most.

Additionally, large-scale AI rollouts—particularly in domains such as autonomous vehicles, biotech, or healthcare—amplify systemic risk. A bubble-induced crash could bring disruption not only to financial markets but also to essential public services that rely on AI. Without proactive governance, monopolistic dynamics, combined with poorly audited AI deployments, can magnify both technical and societal vulnerabilities.

Mitigations require a security-first strategy: anticipate infrastructure fragility, enforce AI-specific resilience testing (including prompt-injection and model misuse), and decouple critical systems from volatile funding cycles. Enterprises should prioritize establishing minimal acceptable security baselines that persist regardless of market conditions.

In short, the "installation phase" of AI isn't simply about scaling; it's also about growing responsibly. Brace for instability—but let it not blind you to the imperative of building AI systems that are not just powerful, but also resilient and secure.