What the Study Actually Said
In late 2025, MIT researchers did something almost no one in the AI industry wanted them to do: they looked at the evidence. Not the projections. Not the pitch decks. The actual, documented outcomes of 300 public AI implementations in real businesses, measured against real profit metrics.
What the data showed was a funnel that should be pinned above every executive's monitor. Sixty percent of companies evaluated AI tools — a broad, low-commitment activity that costs little and proves less. Twenty percent of those companies took a project to the pilot stage, where real resources and real processes are involved. And of those pilots, only 5% made it to full production deployment on the service line. Ninety-five percent of AI investment, in other words, dissolved before it produced a measurable outcome.
The finding is not a verdict against AI as a technology. It is a verdict against the way most companies are currently deploying it: with haste, without clear success metrics, and with the assumption that the efficiency gains will be self-evident once the tool is in place. That assumption, the data shows, is wrong in 95 cases out of 100.
The Layoff Math That Doesn't Add Up
The official story of 2025 was that AI was eliminating jobs. Consulting firm Challenger, Gray and Christmas tracked 55,000 US layoffs that companies explicitly attributed to AI investment — to automation, to efficiency, to "doing more with fewer people." That number traveled widely. It became the shorthand for a new economic reality.
What traveled less widely was the total layoff figure: 1.17 million jobs lost in 2025, the worst number since the COVID-19 pandemic. The gap between 55,000 and 1.17 million is where the real story lives. The vast majority of those job losses had nothing to do with AI efficiency. They were budget cuts. Restructurings. Bets gone wrong, markets shifted, revenue missed. Companies reached for the AI narrative because it sounded strategic rather than desperate. Cutting staff because you're losing money is embarrassing. Cutting staff to invest in the future sounds like vision.
The conflation matters because it distorts the public understanding of what AI is actually doing to the labor market — and it lets companies avoid accountability for decisions that had little to do with genuine efficiency gains. When the reversal comes, as Forrester predicts it will, the same rebranding instinct will be applied in the other direction.
"The vast majority of money invested in AI just fades away. Right now, the entire AI economy is balancing on the head of a pin."
Source transcriptThe Energy Cost Nobody's Counting
There is a second ledger that almost no AI efficiency calculation includes: energy. As of July 2025, ChatGPT was processing approximately 2.5 billion queries per day, with Gemini and other major AI platforms handling comparable volumes. Keeping those systems running requires the energy equivalent of roughly one full nuclear reactor operating continuously, every single day.
The numbers become more alarming when you look upstream. Training the next generation of foundation models — the large-scale supercomputer runs that produce the AI tools companies are buying — requires the energy equivalent of up to 10 nuclear reactors per day, per training run. This is not a marginal cost. It is a structural one, and it is growing as model scale increases.
When a company calculates that it saved money by replacing a human worker with an AI system, that calculation almost never includes a proportional share of the energy infrastructure required to run the AI at scale. The savings look clean on a headcount spreadsheet. They look considerably less clean when the full resource picture is drawn. This is not a reason to reject AI — but it is a reason to be honest about what the true cost of "efficiency" actually is.
Where AI Genuinely Delivers
The 5% of companies that showed real profit impact from AI deployment have something instructive in common. They did not try to replace human judgment. They identified specific, bounded tasks where the input was well-defined, the output was measurable, and success could be evaluated objectively — before deployment, not after.
Content drafting is a legitimate category: companies using AI for first-draft generation of marketing copy, internal documentation, and routine communications are seeing time reductions of roughly 60% on those specific tasks. Code review and bug triage — where AI can scan large codebases for known error patterns faster than any human reviewer — is producing consistent, measurable time savings in engineering teams. Data summarization for structured reports, customer FAQ handling for the narrow band of genuinely repetitive queries, and categorization tasks at high volume all represent areas where the tool's speed advantage is real and the failure mode is contained.
What these use cases share is precision of scope. The companies that benefit from AI are not asking it to think. They are asking it to execute — faster and at lower marginal cost — a task that is already well understood. The companies that are suffering are asking AI to substitute for the understanding itself.
What This Means for Your Business
The most dangerous benchmark in AI adoption right now is headcount. "We replaced X employees with AI" is not a measure of efficiency. It is a measure of reduction. Efficiency means the same output at lower cost, or more output at the same cost — and those outcomes need to be tracked at the level of individual tasks, not org charts.
The companies that will emerge from the current AI investment cycle with genuine competitive advantage are those measuring at the task level: did this specific workflow get faster? Did it get cheaper? Did quality hold? Goldman Sachs, cautious about the AI investment wave, suggested the technology would not have a significant economic impact until 2027 — which is, notably, exactly when Forrester predicts the reversal of AI layoffs will be in full swing.
The two-year window between now and that inflection point is not empty time. It is the window in which companies that are honest about what their AI investments are actually producing will build a foundation that the headline-chasers won't have. Measure the task. Track the outcome. Ignore the narrative.