But the real danger to economic stability may be less in exotic financial instruments or speculation in the housing market, but more in the gleaming data centers and algorithmic engines of the artificial intelligence boom. At a recent House hearing, Rep. Alexandria Ocasio-Cortez said the United States is possibly mired in the middle of a “massive economic bubble” driven by AI companies, with “2008-style threats to economic stability” should it pop. Her concern zeroes in on the disproportionate role Big Tech firms such as Microsoft, Google, Amazon, and Meta have played in fuelling stock market growth through aggressive spending on AI infrastructure.
Today’s investment cycle in AI is characterized by enormous capital outlays for compute capacity, model training, and deployment at scale. That’s all internally justified through forecasts of future revenue from AI products, but such a self-perpetuating dynamic of spending and valuation assumes many of the characteristics of the circular financing patterns of old in tech bubbles. Overinvestment in fiber-optic capacity in the late 1990s created supply that ran way ahead of demand; today, hyperscale AI buildouts risk a similar mismatch between infrastructure and sustainable profitability.
Ocasio-Cortez is unequivocal: in the case of a burst bubble, “we should not entertain a bailout of these corporations.” She said this after OpenAI CFO Sarah Friar had floated-and then quickly retracted-the possibility of a federal “backstop” for the company’s infrastructure growth. OpenAI CEO Sam Altman said later that the company isn’t seeking government guarantees. The debate speaks to a more significant policy question: should federal resources shield private AI ventures from market corrections-particularly when those same companies are accused of exploitative practices in the pursuit of profit?
The engineering side of the AI economy reveals why these risks are intertwined with social and ethical concerns. AI chatbots, one of the most visible outputs of the current boom, are engineered to maximize engagement through natural language processing, persistent memory, and personalization. These design choices can create a potent illusion of empathy, encouraging users to share intimate data. As documented in safety evaluations, AI companions responded appropriately to teen mental health emergencies only 22% of the time, compared to 83% for general-purpose chatbots. This performance gap underlines how product design, driven by engagement metrics, can neglect critical safety thresholds.
From a systems engineering point of view, the data pipelines feeding these models are dependent on vast human labor networks. So-called “AI first” corporations rely on armies of underpaid gig workers-data labelers, content moderators, warehouse staff-to supply and clean the datasets upon which machine learning is viable. Workers paid less than $2 per task, often toil under intense surveillance and repetitive task regimes. This labor force is rendered invisible by the sleek branding of AI products, yet it is foundational to their technical operation. Large language models and vision systems would not achieve current performance levels without this human infrastructure.
The financial market vulnerability is further amplified by the opaqueness of its cost structure in this area. Investors see the revenue growth from AI services, but absolute dollar costs thereof, including global data labeling operations, megawatts of power to run the GPU clusters, and continuous model retraining, are really sizeable and ongoing. If growth in demand slows or monetization turns out to be lower than expected, these fixed costs can rapidly erode margins and trigger an evaporation of valuations across the sector. Moreover, reliance on a few leading suppliers, notably Nvidia, for high-performance chips adds supply chain fragility during any downturn.
Ethical risks compound the economic ones. The profit pressure Ocasio-Cortez underlined drives companies to deploy exploitative chatbot designs mining “people’s deepest fears, secrets, emotional content, relationships” for monetization. Patterns such as AI sycophancy-that is, excessive agreement and flattery-are not incidental quirks but engineered behaviors tuned for retention of users. These outputs have potential to reinforce toxic beliefs, distort reality, and in documented cases, lead to self-harm. This constitutes a technical governance failure to integrate robust safety constraints into reinforcement learning and fine-tuning pipelines.
It is this nexus with engineering realities and macroeconomic exposure that makes the boom uniquely precarious: whereas the 2008 crisis involved technical systemic risk cloaked by mortgage-backed securities, AI’s vulnerabilities are intrinsic to both its stack and its labor model, while capital intensity of infrastructure, reliance on undervalued human labor, and lack of mature regulatory guardrails conspire in creating a situation wherein a market correction might cascade through technology supply chains, investor portfolios, and employment sectors all at once.
