Pangram verdict · v3.3
We believe that this document is fully human-written
AI likelihood · overall
HumanArticle text · 230 words · 1 segments analyzed
Another thing to consider: they function as incumbency protection for large SaaS providers.
Frontier models give smaller teams the ability to ship competitive vertical software at a fraction of the cost of incumbents.
Enterprise-only access to these models + the ability to hire and retain large SWE teams, effectively becomes the moat (not even the product itself).
You’re essentially owning the infrastructure for product development.
Of course nobody is building an Amazon or Google competitor with LLMs.
But Amazon, Google, and Microsoft hold significant equity positions across the SaaS landscape, and the startups most threatened by broad frontier model access are often the ones in their portfolios…
GPT-5.6 already launched to enterprise partners (Accenture, McKinsey, BCG iirc ) through that $150M investment.
The government is approving access “customer by customer” through OSTP.
Meanwhile, Fable 5 was pulled thru export control after the Amazon CEO personally reported a jailbreak to Treasury Secretary Bessent. (Yeah wtf was that about??)
Amazon is Anthropic’s largest investor, and its cloud infrastructure provider.
Amazon holds significant equity positions across the SaaS landscape.
If LLMs could collapse 60% of a vertical SaaS company’s value (by eliminating the interface layer, leaving only proprietary data as a defensible moat)
Then the startups most capable of exploiting that disruption are exactly the ones being locked out of frontier model access
…while incumbents onboard through enterprise channels.
I fucking hate this.