Today we're diving into a shift happening in the AI landscape right now — one that might surprise you (and perhaps even be worrying!) given all the hype we've been hearing. While tech giants continue pouring billions into AI infrastructure, many organizations are hitting a wall when it comes to actually implementing AI — particularly generative AI — in meaningful ways. Let's explore what the heck is going on.
The key piece of context is that Fortune 500 executives are increasingly expressing frustration and disappointment. "I don't know why it's taking so long," they say. "I've spent money on AI but not getting a return on the investment." This sentiment is spreading across corporate boardrooms, and I have data to back it up. According to survey results from S&P Global, 42% of companies are now abandoning most of their generative AI pilot projects. That's a huge jump up from just 17% of firms responding that they were abandoning most AI projects last year. Even Klarna, the Swedish buy-now-pay-later provider, recently admitted they went too far in replacing customer service jobs with AI and are now rehiring humans for those roles.
What we're witnessing is what Gartner calls the "trough of disillusionment" — that inevitable phase in their famous hype cycle that follows the initial euphoria around new technology. It's where reality meets expectation, and the gap between them becomes painfully clear.
Now here's what makes this particularly interesting: While businesses struggle, consumers are embracing generative AI like never before. Sam Altman recently revealed that ChatGPT is being used by 800 million people per week. That's double the usage from a few months ago, in February. The technology clearly has appeal, but translating that consumer enthusiasm into systematic business transformation? That's so far proving to be a much tougher nut to crack.
So why are companies struggling? Several reasons stand out. First, their data are often trapped in silos and legacy IT systems make integration challenging. Second, there's a serious shortage of technical talent with the skills needed to implement these systems effectively. Third — and this is crucial — companies have brands and reputations to protect. They can't afford to have an AI bot make damaging mistakes, expose customer data, or violate privacy regulations. The stakes are simply too high.
Meanwhile, the hyperscalers — Alphabet, Amazon, Microsoft, and Meta — continue their massive AI infrastructure investments. Pierre Ferragu from New Street Research points out that these hyperscalers’ combined capital expenditures are on track to hit 28% of revenues this year. That's more than double the 12% of revenue these firms were spending on capex a decade ago. The superscalers are betting big, but the question remains: Will they generate healthy enough returns to justify this unprecedented spending spree?
At recent developer conferences, tech leaders like Microsoft’s Satya Nadella and Google’s Sundar Pichai painted an optimistic picture (of course!). They talked excitedly about "platform shifts" and an emerging "agentic web" where semi-autonomous AI agents interact with each other on our behalf. They highlighted how AI models are getting better, faster, cheaper, and more widely available. They even introduced new metrics like the number of tokens processed to demonstrate booming usage. But notably absent from these presentations? Traditional business metrics like sales or profit growth from AI initiatives.
The reality is that most cloud revenues from AI are coming from AI labs and startups — many of which are actually funded by these same tech giants. It's a bit of a circular economy at this point. However, the hyperscalers are applying AI to their own operations with some success. Google has launched AI summaries in search results that reach 1.5 billion people monthly and integrated generative AI into its ad business. Meta has woven AI into its advertising platform using its open-source Llama models. Microsoft has embedded AI into its workplace apps and GitHub coding platform. Amazon is using it to improve product recommendations and optimize logistics.
These internal applications might even help reduce costs. Microsoft recently laid off 6,000 workers, many reportedly software engineers, as AI tools make certain programming tasks more efficient. If these efforts succeed, they might encourage other companies to keep experimenting until they too can make AI work effectively.
The cost of falling behind in the AI-investment race is already evident at Apple, which was slower to embrace generative AI. Their attempt to rebuild Siri around Large Language Models has been so bug-ridden that the rollout had to be drastically postponed. It's a cautionary tale about the risks of both moving too fast and moving too slow.
So, if AI’s currently in the “trough of disillusionment” when will we emerge from it? Analysts from Gartner predict it will last until the end of next year. In the meantime, there's serious work to be done. Microsoft's CTO Kevin Scott points out that for AI agents to fulfill their promise, they (for example) need better memory systems to recall past interactions. The web also needs new protocols to help agents access various data streams. Companies are working on standards like Model Context Protocol (MCP; which you can hear more about in Episode #884 of this podcast) to address these challenges.
Here's the key insight: Many companies say what they need isn't necessarily cleverer AI models but more practical ways to make the technology useful. This is what we can call the "capability overhang" — we have more AI capability than we know what to do with! The challenge isn't building better models; it's figuring out how to apply what we already have in ways that create real business value. (This opportunity, incidentally, is exactly why I founded my new consultancy, Y Carrot.)
At Microsoft’s Build conference last week, Anthropic's CEO Dario Amodei urged users to keep the faith, particularly given how quickly AI progress is happening. "Don't look away," Dario said. "Don't blink." It's good advice. While we're definitely in the trough of disillusionment right now, this is actually a good thing because troughs have two sides! What follows in Gartner's hype cycle is the "slope of enlightenment," where practical applications emerge and real value gets created.
The lesson here? Transformative technologies rarely follow smooth adoption curves. The internet went through similar phases. So did mobile computing. The current AI trough doesn't mean the technology has failed — it means we're entering a more mature phase where hype gives way to hard work, and real applications emerge from experimentation.
For you data science professionals and business leaders out there listening, that means this trough is actually an opportunity. While others grow frustrated and pull back, those who persist in finding practical, focused applications for AI — who solve the integration challenges, build the right teams, and manage the risks effectively — will be the ones who emerge strongest when we climb out of this trough.
The AI revolution is getting real. And that might be the best news of all.
The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.