Has the AI Bubble Burst?
Benefits aren't apparent, costs are massive, and clients and investors are growing weary
LLMs, machine learning, and general computational analysis approaches that don’t get fancy acronyms, public notice, or massive infusions of speculative funding — for example, neural networks, good Python programs, and simple, well-constructed queries into well-managed datasets — will continue to provide good value at relatively low cost. Computer analyses are only going to get better, be used by more people, and help in myriad ways.
However, the “AI” that has left so many breathless and agog may have already proven it’s a flop — a mile wide, a 1/4” deep, and unreliable.
Here are some reasons to believe this is so, many of which are courtesy of a stalwart analyst of the space, Gary Marcus. The list is surprisingly long:
- Nobody has a moat, and a moat may not be feasible, making it unlikely investments will pay off with market dominance and pricing power.
- The most likely moat — hardware — isn’t very likely now, as spending by multiple players is high enough, while the energy expenditures needed to run the “compute” are so massive that until and unless there is real value to show, all this build-out may create societal backlash and glaring costs that repel investors
- The purported value is not manifesting:
- According to a recent survey, 47% of employees using AI say they have no idea how to achieve the productivity gains their employers expect, and 77% say these tools have actually decreased their productivity and added to their workload
- Microsoft’s CFO is asking investors to not expect returns on AI investments for 15 years or more, which has spooked the market