Between Two Quarters: Inside Cisco’s AI Investment Strategy with Noah Yago

His view from that seat is unusually clear.

And what he sees often is founders misunderstanding what it actually takes to win inside an enterprise - not just Cisco, but any large company buying software.

The Enterprise Is Not One Buyer. It's a Matrix.

Cisco has 80,000 to 90,000 employees and a dozen different business units. Noah's team of about 40 people operates across all of them - covering both corporate development (acquisitions) and the corporate VC practice simultaneously.

That structure is intentional.

"We see ourselves as driving inorganic strategy. If we aren't building technology organically, we look at options to partner, invest, or ultimately acquire."

But it creates a real challenge for startup founders: there is no single "Cisco buyer." Getting traction means understanding which business unit owns the problem, who has budget authority, how procurement works, and what the internal approval chain actually looks like.

Most founders skip that work.

They focus on UI, when they should be thinking about incumbency, governance, and policy.

The value of beating an incumbent on features is smaller than founders think. The switching cost, compliance overhead, and political risk of bringing in a new vendor are often larger. The startup that wins isn't necessarily the one with the best demo - it's the one that makes the internal champion look good and makes the enterprise risk feel manageable.

The Proof-of-Concept Trap

The AI era has scaled pilot budgets significantly. Enterprises are spending real money to run proofs-of-concept. That feels like traction.

It isn't always.

Noah draws a hard line between POC and repeatable revenue. Most of what looks like enterprise momentum is still in the experimental phase - controlled environments that hide the data fragmentation, governance gaps, and workflow disruption that surface at scale.

Getting from a POC to a production contract requires something specific: an internal champion with a line-item budget who can work the technology into their operating workflow.

Titles matter in large companies. So does access. The executive sponsor who loves your pitch may not be the person who can actually cut the check - or absorb the implementation cost into their team's operating reality.

Understanding the difference between a champion who wants to buy and a champion who can buy is one of the most leveraged questions in enterprise sales.

When Is It Too Early to Sell Into Large Enterprises?

Yes, it can be too early.

Cisco has over 100 active portfolio companies. What they've learned from watching that cohort: for seed or Series A companies, a direct enterprise partnership with Cisco is often premature without a dedicated account team of three to five people specifically assigned to run the program.

"Best engagement happens when companies are a bit more mature and have dedicated teams to run a program with a large company."

That's not a rejection of early-stage companies. It's a recognition that large enterprise partnerships have real operational overhead - and companies that aren't resourced to manage them often get derailed. Product timelines slip because of custom engineering requirements. Revenue doesn't materialize because the POC never converts.

The mid-market is often the right starting point for early stage founders. Big enough problems, small enough politics, faster feedback loops.

Shadow IT Is Back. AI Is the New Dropbox.

There's a familiar pattern playing out right now.

Dropbox and Box created "shadow IT" because individual users adopted them before IT had a policy. Mobile devices did the same thing. Now AI tools - including Claude - are following the same arc.

Leaders are telling their teams to "use more AI." Practitioners are using it for things like deep research, writing, and synthesis. IT and security have almost no visibility into what's actually happening.

Enterprises will eventually solve this the way they always do: with governance, responsible use policies, and an approved vendor list. It will be messy. It will take time. But it will happen.The question for founders building AI products is: are you building for the practitioner buying moment, or the enterprise governance moment? They require different product decisions. And only one of them sustains at scale inside large organizations.

Cisco's AI Thesis: Four Buckets

Two years ago, Cisco launched a focused program to invest exclusively in AI. Noah described it as one of the most active corporate AI investment programs in the market today.

They organize the landscape into four categories:

Foundation models - positions in Anthropic, Mistral, Cohere, and xAI, among others. The thesis here isn't just financial. Being inside Anthropic's Project Glasswing gave Cisco early access to preview models and the ability to address security considerations before deployment at scale.

Infrastructure - companies like CoreWeave and Groq, focused on compute and inference.

Applications - primarily security-oriented. This is where Cisco's core strategic interest is most concentrated.

Tooling - companies like Scale AI and Lightning that make models more useful and deployable.

They also take a deliberately global approach. AI development is moving at different speeds in different regions - China is ahead in manufacturing applications, North America leads in research - and Cisco wants exposure across that spectrum.

Why Do World Models for Enterprises

This was the most forward-looking part of the conversation.

AI development, as Noah frames it, has moved in phases: text → images → video → physical world. A world model is what makes the physical-world phase possible. It allows you to generate a three-dimensional model of an environment from a prompt.

The implications are significant:

• It makes compute less expensive for video and image generation.
• It improves reasoning and reduces hallucinations because the model understands the laws of physics.
• It allows robots to learn from their environment through interaction rather than hard-coded "if-then" programming.

And perhaps most importantly: it makes all other AI better.

This is where Cisco's most cutting-edge investment bets - including World Labs - are focused. It is early. It is speculative. And by Noah's account, it may be one of the most important infrastructure layers in the next wave of AI.

It's also why we recently invested in Human Archive - a company building the data layer that makes physical AI possible. The insight is simple and sharp: language models had the internet. Physical AI needs its own equivalent. Human Archive captures the full sensorimotor stack of human interaction - egocentric video, depth, tactile force, and body movement - all from the human's perspective, not a robot's. They deploy hardware rigs into real work environments through labor partnerships in India, generating ground-truth data at scale that no lab can replicate.

World models need to understand physics. Human Archive is building the data that teaches them how humans actually move through it.

Check out the full conversation below and for more of nvp capital's Between Two Quarters, follow us on linkedin at https://www.linkedin.com/company/nvpcap