Most blue-chip companies, the FT found, were discussing and pushing AI initiatives at an ever-greater pace. But tangible business success stories—improved margins, higher revenues, etc.—were almost non-existent. This comes on the heels of findings by MIT’s Media Lab that 95% of generative AI investments produced no measurable return. That report sent stocks down the day it came out. Investors worried that AI hype was just that.
What these signals really prove is not that AI is overhyped, but that the wrong firms are being treated as bellwethers. Large corporations don’t adopt disruptive tech at speed. They rarely lead. It’s the midmarket and down who will seize opportunities first with AI, and that’s something we see live, every day.
Previous technology breakthroughs exhibited the same pattern. Take cloud computing: startups and digital natives went first while enterprises dragged their feet for a decade, insisting their old data centers were fine. Streaming: Netflix proved the model while Blockbuster and cable operators clung to legacy revenue streams until it was too late. Ridesharing and home-sharing: Uber and Airbnb built categories while incumbents tried to litigate them out of existence. Electric vehicles: Tesla went all in; Detroit clung to pick-up trucks.
As for the MIT Study: why are 95 percent of pilots failing? Not because of hollow technology, but because of hollow deployments. Companies are tossing ChatGPT licenses to employees and calling it a strategy. They’re buying point solutions that may not fit their actual gaps and expecting magic. What they aren’t doing is the hard stuff: re-plumbing their data, redesigning processes with AI in the middle, retraining workforces, rewiring incentives.
Our own analysis of S&P 500 earnings calls from the latest quarter show that ‘AI’ was mentioned 2,800 times in the latest quarterly earnings calls, an increase of 25% vs previous quarter. But 84 of the S&P 500 didn’t mention AI or related subjects even once. The full table of this data is here.
And that brings us to the bigger question: what should companies do? If 95 percent of pilots are failing, what separates the five percent that succeed? Success doesn’t come from sprinkling AI on the surface. It comes from meeting the business where it’s at now and mining a business’ operations for the ripest AI and data opportunities. One by one, each success helps build a culture of adaptation and a belief within the company that AI can deliver. The best AI practitioners don’t show up with a baked solution, they show up looking for problems to fix and they shape solutions accordingly.
Some big companies will certainly make plenty of money from AI, especially ensconced hyperscalers, plus OpenAI and Anthropic. But it’s not clear that those margins won’t be tightly contested via competition. The leading LLMs themselves—Claude, GPT, Gemini—already seem to be converging in terms of utility. The largest benefits may well accrete downhill to the companies who find ways to use AI best, rather than those who are running the compute.
In addition, small, focused LLMs and straight machine learning models, running cheaply on company-owned instances, often offer companies the best ROI when imbuing their operations with AI. The projects offering the quickest paybacks have been clustering here, in our experience.
In my 2012 book Automate This: How Algorithms Came to Rule Our World, I tracked how algorithms crept from Wall Street into music, hiring, law, and everywhere else. Three themes stood out:
- Early movers always beat incumbents because they don’t carry the baggage of legacy systems and entrenched habits.
- Disruption often happens in the shadows long before — by the time anyone’s paying attention, the “invisible takeover” is already underway.
- Automaton technology might take years to prove itself, but the last wave of adoption often overwhelms those who didn’t tack early enough—incumbents don’t have as much time as they assume.
We’re seeing the same patterns with AI, the natural extension of algorithms. The revolution isn’t necessarily coming from Wall Street this time, but it’s still rooted in the mid and small markets. Investors looking for the cruise ships of the S&P 500 to turn on a dime may end up disappointed.
Enterprise companies breed risk intolerance. This aversion to risk favors small, incremental “proofs of concept” that never scale and therefore get classified as failures. In the midmarket, the move toward AI is real and we’re seeing results every day. With that said, midmarket companies who move slowly in AI are at even more of a disadvantage than enterprise companies, who can lean on the strengths of incumbency longer.
