Key Takeaways

  • Enterprise software portfolios face severe reduction: Krishna notes the average enterprise buys from 500 to 5,000 distinct vendors, but predicts CIOs will replace hundreds of point solutions with basic AI tools and centralize spend into 5 to 10 platforms.
  • Vertical integration creates multi-decade moats: Krishna pointed to the IBM mainframe surviving 65 years because IBM owned the entire stack down to the sand, building its own fabs and chips in the 1960s.
  • Full-stack ownership allows companies to move profit pools: Arora highlighted Nvidia, where owning the stack lets Jensen Huang commoditize targeted layers and shift margins when rivals attack.
  • Middleware and open model routing will fail: Arora argues that tools like OpenRouter cannot survive because context, memory, and reasoning cannot be separated cleanly across independent layers.
  • Point-solution product market fit is expiring: Arora warns that software terminal values are collapsing, predicting few isolated SaaS tools will survive the next 5 to 8 years.

The Mainframe Lesson and the Vertical Stack

Arvind Krishna points straight to mainframe history to explain software survival: "the IBM mainframe which by the way there's a few people in the room who do use it quite well and at scale the IBM mainframe has lasted 65 years. Why? It was vertically integrated. We're a little less right now in the sense we don't go all the way to dirt. In the 60s, we went all the way to dirt. Meaning, we found sand. We made our own wafers. We built our own fabs. We built our own chips."

That durability applies directly to modern tech. Startups spent a decade chaining third-party APIs together, mistaking fragmentation for speed. Nikesh Arora sees Nvidia repeating IBM's playbook: "If you look at what Jensen is doing, he's the modern day version of the vertically integrated stack... And the good news is if you control the vertical stack, you can decide which part of the stack to commoditize. So you can decide to move profit between layers and if you see that there are other people attacking a certain part of your stack, you they destroy the profit pool in that stack, you move the profit to the other part of the stack."

Point-solution software has no other layer. When a foundation model copies its feature, its profit pool disappears completely.

Why Model Routers and Point Solutions Break

The early AI wave bet on horizontal middleware: build a router, swap models dynamically, and treat intelligence as a utility. Arora argues that this architecture cannot hold. “I think at some point in time this notion of running something like an open router become impossible because models are getting more and more each model is requiring more memory more context more understanding you can't separate context and memory from the reasoning,” Arora stated.

Decoupling reasoning from context introduces unacceptable latency, memory loss, and security risks. As a result, enterprises want fewer vendors, not more. Krishna observed that the average enterprise buys from “500 to 5,000 distinct software vendors,” creating an unmanageable attack surface. His conclusion is stark: “could I replace 490 of them with minor AI capabilities and do I need to go back and leverage 5 to 10 main platforms? That implies massive consolidation because that's where the spend goes.”

Arora echoed this warning on SaaS lifespan: “it's very hard to understand the terminal value of software. It's not clear what software is going to survive the next 5 to 8 years with where we are from an AI perspective and the notion that product market fit is getting disrupted for many of these companies.”

What to Do With This

Audit your roadmap this week and flag every feature acting as a standalone point solution. If a core platform like Salesforce or Microsoft can replicate your product with a native prompt, drop that feature and rebuild around proprietary customer data that platforms cannot access.