Models Eating the World: A Wardley Map Response
How the AI value chain is inverting before our eyes
Someone—Vintage Data (a blog by Alexander Doria), to be specific—published a piece called "The Model is the Product" that hit like a depth charge in my thinking about AI. Their argument boils down to this: AI models aren't just becoming products, they're becoming the product, swallowing the applications above them whole.
I wondered: what would this power shift look like on a Wardley Map?
Three Things This Map Reveals
1. Models Are Climbing the Food Chain
See those red circles moving upward? That's OpenAI and Anthropic (Base LLMs) getting tired of being the lumber that others build houses with. They're becoming architects, constructing DeepResearch-style specialized models and agent systems that perform complex tasks themselves.
The post calls out Databricks' AI VP, Naveen Rao with brutal clarity: "all closed AI model providers will stop selling APIs in the next 2-3 years." Not subtle, that.
2. The Wrapper's Dilemma
The blue circles—your Perplexitys and Cursors (“App wrappers”)—built businesses taking raw model output and making it pretty. Now they face an existential choice that the post frames perfectly: "training or being trained on."
Either build your own models (expensive) or become training data for the models that replace you (fatal). Some, like Perplexity, already see the writing on the wall and are building their own models (and browser–but perhaps that’s a different post.)
3. Token Economics in Free Fall
Bottom right shows inference costs becoming a commodity. When DeepSeek's optimizations mean "all available GPUs could cover a demand of 10k tokens per day from a frontier model for... the entire earth population," the token-selling business is dead. Full stop.
The Shift No One Priced In
Reinforcement learning—this map helps visualize the blindspot that Vintage Data identifies. It's not just that models are getting better; they're learning to do complex tasks internally without the byzantine workflows we've built around them.
Models aren't just predicting text anymore; they're learning tasks. It's the difference between handing someone ingredients and watching them follow your recipe versus hiring a chef who just knows how to cook.
So What?
If this map accurately captures reality, perhaps here is what different players should consider:
For an AI application: Your wrapper advantage is temporary. Either move into specialized training, focus on industry verticals the big labs won't touch, or prepare to be planted. The days of marking up simple API calls is numbered.
If you're investing in AI: Look harder at companies developing training innovations, not just application interfaces. The value is shifting upstream faster than most portfolios are adapting.
For the large enterprise: Be wary of building critical infrastructure on wrapper applications that might disappear. Consider working directly with model providers for specialized solutions or invest in open-source alternatives.
The map suggests we're not just witnessing routine industry evolution but a fundamental restructuring of where and how value is created in AI. The winners will be those who recognize this shift earliest and position themselves accordingly.
As the post concludes, many in the industry are "fighting the next wars with last war generals." This map doesn't predict the future, but it might help you navigate it.