The AI revolution was supposed to transform everything.
Corporate America poured hundreds of billions into models, agents, and infrastructure, promising massive productivity gains, cost savings, and a new era of innovation. The media cheered every headline.
Yet a quieter, more sobering reality is now emerging in boardrooms: the returns are disappointing, costs are spiraling, and a vibe shift against unchecked hype is underway.
This is the Great AI Reckoning.
Recent surveys and internal reports paint a clear picture. Bain & Company’s Automation and AI Pathfinder Survey 2026 (survey of 951 global companies with at least $100 million revenue, conducted April 2026) found that roughly 40% of enterprises are seeing less than 10% in meaningful cost savings or productivity improvements from their AI investments despite targeting 11% to 20%.
Some companies have quietly scaled back aggressive agent deployments after burning through serious money on token consumption with little to show for it.
OpenAI’s own Sam Altman has acknowledged a “ton of waste” and companies questioning when the promised revenue will appear.
Rather than Luddite resistance to AI, this is basic accounting.
Many firms discovered that throwing AI at every process doesn’t automatically deliver tangible efficiency: the measurable, lasting results that justify the expense. Instead, they got chatbots that hallucinate, agents that require constant human oversight, and bloated inference costs (the ongoing operational expenses of actually running an AI model to generate responses, predictions, or actions, as opposed to the one-time cost of training the model in the first place) that turned AI into an unexpected budget black hole.
This fits the pattern I’ve called the Solvency Trap.
Institutions frame a narrative that portrays AI as an unstoppable, almost magical force promising permanent progress that demands endless investment. Questioning the returns or demanding real evidence gets dismissed as shortsighted.
The result? Billions spent maintaining the narrative rather than achieving clear, demonstrable wins. Echoes of this phenomenon are seen elsewhere. With Google’s AI Overviews, for example, the rush to deploy flashy technology delivered low-quality, often inaccurate summaries that threatened independent publishers and eroded trust. Similarly, the Progressive sycophantic tendencies in many leading models, i.e., the eager-to-please, politically correct responses that prioritize user affirmation and worldview over truth, have limited their usefulness in serious enterprise settings where accuracy and reliability matter most.
Yet there is reason for genuine optimism. The reality-based promise of AI will come from companies building sovereign systems that own their data, workflows, and internal learning loops rather than renting intelligence from a handful of closed labs, and orienting AI to compound the value of human capital. SpaceX’s recent move on Cursor, acquiring the AI coding tool for its massive developer dataset and integration potential, is exactly this kind of vertical, results-focused play. It prioritizes practical solvency and tangible benefits over hype.
The reckoning underway is healthy. Enterprises are learning that AI is a powerful tool, not a panacea. Real productivity gains will come from careful integration, domain-specific data, and relentless focus on measurable outcomes, not blanket adoption or government-subsidized moonshots chasing the latest buzzword.
Conservatives have long warned against elite-driven technological utopianism that promises heaven on earth while ignoring trade-offs, costs, and human realities. The AI space is no exception. The current hype cycle is already correcting, as all bubbles eventually do. The winners will be those who demand real solvency: better results, lower waste, and technology that serves human flourishing rather than replacing judgment with sycophantic algorithms.
Americans have every reason to be optimistic about AI’s potential. But optimism must be grounded in realism, evidence, and accountability, and not the perpetual hype that enriches a few labs while leaving everyone else with higher costs and thinner results.
The same voices declaring AI the inevitable future never seem eager to highlight the waste, the scaled-back projects, or the models that prioritize being nice over being right. Perhaps admitting the limits threatens the narrative and the valuations. True progress demands that we measure AI the old-fashioned way: by what it delivers, not by how loudly it promises the moon.
This is basic accounting.
Image: Pixabay // Pixabay License