The hidden problem in AI

Key Takeaways

It's finally happening, some reckoning. So, companies are set to spend $2.5 trillion in AI this year. But, some of these projects are starting to fail and people are starting to realize. So, this this is from a new report from Gartner. 57% of AI leaders have had one AI project fail outright recently. Only 28% of AI incentives and infrastructure and operations are delivering on the AI that they promised. And according to Gartner, AI is officially in this >> [music] >> moment of disillusionment in 2026. The phase of the hype cycle where reality catches up with the promise. But, what's actually going on is that companies aren't failing because the AI is bad. They're failing because they're automating broken processes and the AI does exactly what it's [music] told. It it just turns out nobody had clearly defined what it was. So, so skill gaps are a factor in 38% of failures. Poor data quality is cited just as often. And the most complex use cases like self-healing infrastructure or AI-led workflow automation, that's where failure rates are highest because those systems they don't have predictability in the real world. Meanwhile, the spending just keeps going, right? So, Microsoft, Google, Amazon, Meta, they're on track to spend $700 billion in AI infrastructure this year alone, roughly double what they spent in 2025. Wall Street is demanding proof that these investments are working, but the proof is not coming from internal tools that are quietly being shut down. It's coming from enterprise contracts and the numbers that companies are reporting publicly. So, the pattern here is familiar. Every major technological breakthrough has this moment, has this gap. First the internet in 2001, then mobile apps in 2012, then the blockchain in 2018, and I guess a little bit now. The companies that survive aren't the ones who spent the most, they're the ones who figured out what actually works before the money runs out.