It's an old rule of journalism: if it bleeds, it leads. In the AI world, the same principle seems to apply. When MIT released a headline-grabbing claim that "95% of enterprise AI pilots fall," it went viral. Within hours, it was everywhere: Fortune, Forbes, Axios. Stock markets reacted instantly, and NVIDIA dropped 3.5%; AI-focused companies lost billions in market value.
The truth is, the study was flawed. It was based on just 29 interviews, all with companies that had volunteered to discuss their "AI challenges". They spoke only to those struggling, then declared that everyone was struggling. Conveniently, the report ended by promoting an MIT programme charging £200,000 for "corporate memberships". This wasn't research; it was marketing.
A more rigorous, three-year study from Wharton tracked 800 enterprise decision-makers. Its finding was significant: 74% of organisations report a positive ROI from generative AI. That research barely made a headline; success stories rarely do. Wharton's data shows AI has shifted from an IT experiment to a boardroom priority. Nearly half of executives now use it daily, and 67% say their company's AI strategy sits at the executive level. The most telling detail: around 30% of AI budgets go to Internal R&D - proof that organisations are building, not just buying. The demand for compute isn't a short-term fad; it's a long-term strategic shift.
The winners are the boring ones, and that’s precisely why they’re winning. They’re not chasing moonshots. They’re automating measurable workflows and proving value quickly. The highest-performing use cases are back-office tasks:
- Data analysis (73%)
- Document summarisation (70%)
- Document editing (68%)
Meanwhile, the failures are chasing complexity too soon, trying to deploy 50 autonomous agents before teaching staff basic prompt engineering.
The Real AI Story: What We See at DataVita
This is what we see firsthand at DataVita. We build the high-performance, high-density data centres that power AI. The technology works. The blockers are human.
The Wharton study confirms it. There’s a perception gap between executives (56% optimistic) and middle managers (28%). Training investment is falling even as tools get more complex. And 43% of leaders fear “skill atrophy”, that automation might dull human ability, which quietly breeds resistance.
By 2026, Wharton predicts the AI winners will permanently pull away. The rest, paralysed by the “95% failure” myth, will be left behind. The losers think infrastructure ends with hardware - the latest GPUs in a rack. Winners know their real infrastructure is human. A million-pound cluster is worthless if the people, managers, and culture around it aren’t aligned. The future advantage won’t go to those with the fastest tech, but to those with the most capable people using it.
The AI failure narrative is easy. The real story is harder, but it’s happening now.
Further Reading
- Scotland’s AI Crossroads: Two Futures, One Choice – Exploring how strategic choices today will shape Scotland’s AI future.
- High-Performance Computing at DataVita – How our HPC infrastructure supports real-world AI performance.