-
Constellation Software Inc. (CSU $2,441.95)
-
Descartes Systems Group Inc. (The) (DSG $99.45)
-
Kinaxis Inc. (KXS $144.74)
-
Thomson Reuters Corporation (TRI $128.18)
-
Shopify Inc. Class A Subordinate Voting Shares (SHOP $173.48)
Q: Hi 5i team,
I’m reading about a thesis and would love your view on it.
The idea is that today’s AI compute scarcity is temporary, kind of like the telecom/bandwidth bubble in the late 1990s. As inference costs keep falling over the next few years, I think the real value will shift from the infrastructure layer (GPUs, hyperscalers, data centers) to the application layer.
However, not every application-layer company will benefit. The ones whose moat is the AI itself could actually get hurt as models become cheap and everywhere. The real winners should be companies with moats that don’t depend on model quality — things like regulated workflows, payment rails, proprietary data, deep system integrations, or network effects. For these companies, cheaper compute should expand margins instead of creating new competition.
Two questions:
1. Does this thesis make sense to you, or do you see any flaws in the logic?
2. In your coverage universe, which names best fit the idea of “moat is the workflow / data / distribution, not the AI itself” and which names’ AI itself ‘might get hurt?
Thanks,
Matt
I’m reading about a thesis and would love your view on it.
The idea is that today’s AI compute scarcity is temporary, kind of like the telecom/bandwidth bubble in the late 1990s. As inference costs keep falling over the next few years, I think the real value will shift from the infrastructure layer (GPUs, hyperscalers, data centers) to the application layer.
However, not every application-layer company will benefit. The ones whose moat is the AI itself could actually get hurt as models become cheap and everywhere. The real winners should be companies with moats that don’t depend on model quality — things like regulated workflows, payment rails, proprietary data, deep system integrations, or network effects. For these companies, cheaper compute should expand margins instead of creating new competition.
Two questions:
1. Does this thesis make sense to you, or do you see any flaws in the logic?
2. In your coverage universe, which names best fit the idea of “moat is the workflow / data / distribution, not the AI itself” and which names’ AI itself ‘might get hurt?
Thanks,
Matt