AI: When the Bubble Bursts

Artificial intelligence is already being threaded through government administration, health, finance, defence, cybersecurity, communications, logistics, education, corporate management and the networks connecting them. The speed matters. This is not simply the orderly adoption of a useful technology, but deployment within a political economy organised around investment, growth, competitive advantage and accelerated returns. Capital rewards rapid uptake, market capture and dependency formation; redundancy, maintainability, interoperability and long-term resilience are comparatively easy to treat as costs.

AI did not create this dynamic, and another sufficiently promising technology would eventually have attracted similar pressures. But AI is not merely an interchangeable speculative object. It is unusually powerful and general, capable of entering the informational, communicative and decision-making relations through which other systems persist. The problem is the conjunction of an exceptional technology with a recurrent economic logic: a new technical layer is being installed inside the organisational machinery of society before the persistence of the industrial and financial structures sustaining it has been demonstrated.

If the AI bubble bursts, the technology does not disappear with the money. A severe contraction in investment could bankrupt vendors, cancel infrastructure projects, eliminate technical teams, discontinue products and services, force acquisitions and strip away the continuing support on which deployed systems depend. This pattern already exists. In April 2022, the financially distressed smart-home company Insteon abruptly shut down its cloud servers after failing to secure a buyer. Customers lost app control, automation and scheduling, and some cloud-dependent devices became effectively unusable until a group of users later acquired and revived the company. The stakes become considerably higher when the same dependency is embedded in hospitals, governments, networks or security systems.

Models, libraries, APIs, medical tools, administrative systems and network components can remain physically or organisationally embedded long after the company responsible for maintaining them has failed, merged or moved on. The result is not technological absence but technological residue: useful systems continuing to operate, or remaining necessary, after the commercial structures required to sustain them have weakened or disappeared. The mismatch is structural. The economic lifetime of a company can be much shorter than the operational lifetime of its software, while institutions reorganised around that software may no longer retain the expertise, procedures or fallback systems required to absorb its failure.

Cybersecurity and intelligence make the contradiction especially clear. Unsupported software does not remain at a fixed level of security simply because it continues to run. The 2017 WannaCry attack on the English NHS demonstrated the broader problem. At least 81 of 236 trusts and hundreds of other NHS organisations were affected; infected organisations were running systems that were either unpatched or unsupported, while some medical equipment, including MRI scanners, still contained embedded Windows XP that local trusts could not readily update themselves and for which vendor support was reported to be poor. Vulnerabilities are discovered, dependencies change, operating environments evolve and adversaries adapt while inadequately maintained systems progressively freeze in time.

AI embedded in intrusion detection, authentication, surveillance, network management, intelligence analysis, healthcare, government and critical infrastructure could remain useful enough to preserve dependency while becoming progressively harder to update, understand and defend. A speculative boom can therefore leave behind more than failed companies and depreciated investments: it can leave a distributed layer of operational dependency whose maintenance was never secured beyond the conditions that produced it. The problem is not ultimately AI. It is a system that accelerates the formation of technological dependencies in pursuit of immediate returns while externalising the work of sustaining those relations into an uncertain future. The speculative object will change. The organising logic that repeatedly generates such objects will remain.