The critical variable in AI security going forward is context—not the raw model—because organizations that build proprietary training context can stick any model on top of it, making model distinctions and even frontier-model dependence less relevant.
Nikesh Arora argues that the competitive moat in AI shifts from model intelligence to proprietary context and training data, enabling model-agnostic infrastructure, echoing but differing from Satya Nadella's architectural point. ✦ AI generated
Nikesh Arora · 20VC · 2026-08-06 · original ↗
starts at this moment · 83:35
“I've seen a lot of your my my interests in the future of models, but but do you care which model you're going to build that context on, or is it agnostic from your perspective?”
the part which we will be build, we are starting to build and we will be building for the next 3 to 5 years is context... All that knowledge, all that learning is being captured by me in effectively vector DBs and in context learning systems... then I can stick any model I want on it. And the model distinction will not matter because the context will become as important or perhaps more important.
verbatim transcript · starts at 83:35
(00:49:56) but we may not care who wins. We may not care who wins this battle. We may this may all blow over and it all may be about compute and we may not whoever wins wins. Whoever wins will plug in. Jason, I think the models will get better and better and the distinction between models may not be enough for you to decide to rip one out because I think
(00:50:15) the part which we will be build, we are starting to build and we will be building for the next 3 to 5 years is context. So think about it for a second. Like you know I run a simple firewall company or a simple complicated firewall company. You can stick any model you want. The model doesn't know why my customer's infrastructure is down. It does not know because my model doesn't
(00:50:34) know what product my customer is using. My model does not know what operating system it's using. My model does not know what the configuration of the customer is. My problem model does not know why this happened the last five times as a customer. All that knowledge, all that learning is being captured by me in effectively vector DBs and in context learning systems. And that's what my team is doing. I have more
(00:50:55) people collecting context than I've ever had. It's kind of like the Whimo thing. I got people planted saying this is a tree. This is why it goes down. So as I build that organizational instead of context, then I can stick any model I want on it. And the model distinction will not matter because the context will become as important or perhaps more important. And you're clearly 100%