How I Think the AI Bubble Deflates

⊛ February 17, 2026

So in looking at the AI bubble my thesis on how it deflates in the US is straightforward in that the uncapped exponential demand for AI comes up against the limitations of resources, both financial and physical leaving a few players in monopolistic competition with each other.

From the financial side of things I think it’s becoming increasingly clear that the head start lead from OpenAI is starting to erode as noted by Microsoft losing confidence in them by pursuing other options for AI models, along with Nvidia backing off some of its funding agreements. This in conjunction with the opportunity costs of layoffs, ostensibly due to AI, means that orgs are prioritizing what they see as potential short term promises of gains, when they’ve gutted large swaths of their workforce for nothing. Incessant cutting is not a strategy for growth, it’s a strategy for financialization and looking good to investors. The name of the game at this point is sustainability, and most orgs don’t seem to have a path to profitability or solid revenue.

On the physical side of things, power, HVAC, cooling, and, memory, even public sentiment are going to be major headwinds. So many of the AI providers are looking to get power connections, contracts, and power availability that NoVA (which has the most data centers per capita) is turning against data centers with politicians, neighborhoods, and environmental groups all aligned against new data centers going up. Even in places outside of NoVA residents and politicians are skeptical of data centers as the tax breaks lead to a temporary increase in construction jobs, and then relatively few permanent data center jobs. Having to get so many skilled contractors in HVAC and cooling not to mention fresh water resources to carry away all of the heat is proving to be yet another bottleneck. All of this doesn’t include that DRAM, HDD, and graphics chips contracts extend out for months to where capacity is already obligated for years in some cases increasing prices of a number of goods, further turning public sentiment against AI.

Bottom line: There will be a “Minsky moment”(1), at which point cash will drain from the AI market leaving us to see who is sustainable (Warren Buffet put this in far more colorful terms), but overall there will be something useful to come out of this be it code or offboarding some level of knowledge work. The question is, how much longer can this rampant speculation keep up? My uneducated guess and timeline puts it at 18 to 36 months from early 2026, based on nothing but speculation, much like the AI market.

  1. A Minsky moment is a sudden, major collapse of asset values marking the end of an extended period of bullish speculation, excessive risk-taking, and high leverage. Coined from economist Hyman Minsky’s theory, it describes how long periods of economic stability encourage unsustainable debt (Ponzi borrowing), leading to a inevitable market crash.