Influences for this piece:
- The Consequences of Abundant Intelligence — Michael Bloch
- Future of Entertainment is Interactive — Jordi @ TBPN
If we take the bull case for AGI and abundant intelligence seriously, the cost of inference, the defensibility of information-based moats, and the difficulty of creating new technology will all collapse toward zero, or start to asymptote close enough to it over the long term. So where (on earth - pun intended) does value accrue when those things commoditize?
There are many sides to this argument, but I have a view on both enterprise and consumer and where I think each stands over the short to long term. Before we get into it, it's worth naming where in the stack my argument lives.
In a broad simplification, there are four macro layers to the AI economy: the energy/power layer (grid capacity, data centers, etc.), the compute layer (chips, fabs, data centers), the foundational layer (OAI, Anthropic, Deepmind, Ai2, Safe Intelligence, the frontier models), and the application layer (products built on top of those models, Harvey, Sierra, Artos, etc.). My argument lives on the application layer, because that's where the question of "where does value accrue post-commoditization" gets most interesting in the long tail.
Enterprise
Over the short-medium term (5y horizon), I tend to believe that smaller, heavily verticalized applications will have a strong argument for existence due to workflow lock-in and access to proprietary data. You're only as good as the information and context you're trained on, and if you have high-quality, structured underlying information in a regulated ecosystem, you can make significant strides in efficiency toward a desired use case.
Structure is italicized for two reasons: structure and accuracy go hand in hand. For structure, this particularly leans into regulated industries initially due to specific legal requirements on how information is stored and processed — Pharma, Accounting, etc. For accuracy, think about your SFDC CRM. Plenty of data, but ask yourself how much of it is current and accurate, and I'd argue you'd be afraid to train a model off it.
Think of why Anthropic has started acquiring healthcare businesses. Data moats have a higher advantage in regulated industries because the majority of the information within them is heavily proprietary and not found via open web scraping. Foundational players are now reaching up into the application layer, acquiring vertical-specific businesses to get closer to the proprietary data and workflows that frontier models can't access via public data. If the application layer were going to be flattened by ever-better general models, the foundational labs wouldn't be paying up to own pieces of it. For enterprises today, the right move seems to be grabbing up as much proprietary data as possible per end customer.
You can far-extrapolate this, and I'm always interested in discussing potential wedges into broader industries. I think model selection will also become an interesting angle over the short-medium term — though I'm unsure about its long-term moat. As the capabilities of the foundational models collapse, the race to the bottom on usage costs will always be top of mind for buyers coming to these new usage-based pricing models. The ability to select output and cost constraints will let users derive the best value for the output they want — not everything has to run on Fable or Opus 4.7, and you can get high-efficacy output from lower model selection.
Consumer
On the consumer side, the parallel looks like proprietary data, just in a different form. It's proprietary context and relationship with the end consumer and their community.
Why do you stay in a group chat? Why do you keep coming back to the same Discord? Why do you watch a specific creator? Why do you stay biking with certain friends? None of that is about information being scarce. It's about who, with whom, what you've built together, and the shared context created. That's the moat — just at the social contextual layer instead of the database layer.
I also think the aspect of taste comes into play here. By taste, I mean the ability to discern a quality experience versus a manufactured one. Consumers lean into things that feel inherently authentic, in whatever form.
The shared observation
Both versions point at the same thing: when the underlying capability commoditizes, the thing that holds value is whatever is embedded in a context that can't be cloned. Enterprises call this workflow, compliance, regulation, tribal knowledge. For consumers, it's community, identity, and context. (If interested, see my post on context in a more macro sense.)
In a far-extrapolated world, the durable moat has to live in something commoditization can't reach. I often think about what people organize their lives around outside of friends and family: sports teams, esports orgs, fandoms, religious communities, cities, scenes, group chats, Discords, fitness communities. None of these are about information access. They're about belonging. These are the things that get more valuable as everything else gets cheaper and more abundant, because identity and connection are things you can't generate. You have to be there, over time, with the same community, growing together.
Esports and shows like TBPN are great examples because they're digital-native (forward-looking) versions of this. The thing itself is software. The durable layer is the experience, the connection, and the authenticity. Because of that, I think a lot of proprietary data flows toward them — players on esports teams, founders breaking news on TBPN.
A different framing: value flows toward whatever is irreducibly social. As intelligence starts to look like electricity — necessary and abundant — the margin disappears from intelligence itself. The margin lives in the things intelligence can't substitute for: accountability, embeddedness (relationships and context that compound over time), and connection (the communities and identities).
Each of these is a different outlook with a different business shape, but they share the property that "intelligence-as-commodity" makes them more valuable.
None of this is new. Every time a major input gets commoditized, value relocates to whatever the abundant thing can't substitute for. So what does abundant intelligence create more demand for as a complement, and what is it structurally incapable of producing? For Enterprise, it could break down towards access in owning the IP creating or structuring proprietary data in the form of codifying tribal knowledge and processes. For consumer, Connection becomes the scarcest resource, and community experiences could be where the next decade of value will be built on.
Footnote: as with all things in life, influence and bias are everywhere and unavoidable. In my opinion, our first and last non-influenced action is crying as a newborn. I won't claim true original thought throughout this post — we're all taking in influences across our lives — but I will claim authentic throughlines and parallels drawn from my own experience and thinking. As always, use Pangram to check the things you read.