No, AI Winter isn’t coming – but many of today’s AI leaders won’t survive to see its summer
Obituaries for generative AI have started trickling in across the media, with one widely shared Fast Company piece predicting an AI winter for reasons typical of any technology’s early phase.
Those of us who lived through the Internet’s boom-and-bust cycle remember identical predictions about the World Wide Web’s demise, including one from Robert Metcalfe, the inventor of Ethernet, who publicly predicted the Internet’s collapse in 1995!
So, is an AI winter coming? No. Will most of today’s dominant AI players lose their edge or cease to exist? Most likely. And that distinction is exactly what CxOs betting on a single vendor relationship need to plan around.
The Internet took five to six years to go from hype to bust, 1995 to 2001, because the hype was triggered by an enabling product, the browser. Generative AI’s hype was triggered by an end product, ChatGPT, which compresses that same cycle into a fraction of the time. If your AI strategy is really a ChatGPT strategy, you’re planning around the wrong layer of the stack.
Here’s my forecast, and the framework I’d use to pressure-test any vendor roadmap against it. LLMs will commoditize fast, in fact mush faster than we imagine. When every major lab trains on similar public data with a similar transformer architecture, differentiation gets thin. This is precisely why some of them are lobbying hardest for legislation that limits competition. Custom, private, and personalized models will win the next phase, in both enterprise and consumer markets. The early movers here stand to capture outsized value. Domain-specific models and delivery platforms are the next trillion-dollar layer, and it’s still wide open. And underneath all of it, the real paradigm shift nobody’s pricing in yet is the move from deterministic, rule-based computing to prediction-based computing, a genuinely tectonic change in how software gets built.
Call this the Commoditization Clock. The closer your vendor sits to the raw model layer, the faster they commoditize; the closer they sit to your proprietary data and workflows, the more durable your advantage. Use it before your next multi-year vendor commitment, not after.
I’ll go deeper on that predictive-computing shift in a future post. For now, if your procurement team is locking in a long-term contract with a foundation model provider, it’s worth asking where on the Commoditization Clock that vendor actually sits.