If I could invest in just one part of the AI stack, which would it be?

Hi, this is Sandro from Vanagon.

I wanted to share a few thoughts that have been on my mind lately — especially around how foundational AI models are evolving, and where I believe the most interesting investment opportunities are starting to emerge.

Let’s dive in.

It seems like 99% of the talk (and usage) right now is centered around foundational models.

And for good reason.

🤖 What Are Foundational Models?

These are large AI models trained on massive datasets, designed to be adapted or fine-tuned for countless downstream tasks.

Think: GPT-4, Claude, Gemini.
They’re the base layer for a wide range of applications—from chatbots to code assistants to creative tools.

🏗️ Foundational Models Are...

• Costly and complex to build and train

• General-purpose and highly adaptable

• Increasingly infrastructure-like—used by others to build more specific tools and services

🧠 A Mental Model: Are Foundational Models the New Telcos?

I recently revisited some thoughts from Richard Socher (🇩🇪 Yes, Germans can build 😜), and one analogy stuck with me:

Why that matters:

Telcos like AT&T, Vodafone, and Deutsche Telekom spent decades laying the infrastructure that powers our digital lives. But the real economic winners?
Not them.

It was the application-layer giants—Google, Apple, Facebook—who built on top of that infrastructure and captured most of the upside.

📉 The Telco Trap

Over time, telcos became:

• Low-margin utilities

• Squeezed by massive infrastructure costs

• Bound by price caps and regulation

Meanwhile, tech companies captured the real value, delivering services that ran on those networks.

🎯 Where I'd Invest Instead: The Application Layer

If I could only invest in one part of the AI stack, I’d go up the stack — to the application layer.
Especially in massively underserved verticals with deep domain integration.

Look for platforms that:

• Are embedded deeply into workflows

• Capture proprietary feedback loops and data

• Replace multiple legacy SaaS tools

🚜 6 Real-World Vertical Plays for AI

1. Agriculture & Food Systems

Why now: Climate stress. Yield volatility.
Why AI: Crop risk prediction. Precision input planning. Autonomous monitoring.

2. Logistics & Supply Chain

Why now: Post-COVID volatility. VUCA pressure.
Why AI: Forecasting. Route optimization. Demand-matching copilots.

3. Manufacturing & Industrial Ops

Why now: Aging workforce. Costly downtime.
Why AI: Predictive maintenance. Quality control. Process tuning copilots.

4. Materials Science & CleanTech

Why now: Climate urgency. Fragile supply chains.
Why AI: Generative chemistry. Simulation engines. Robotic R&D loops.

5. Biotech R&D & Life Sciences

Why now: Data deluge. R&D cost pressure.
Why AI: Literature copilots. Trial optimization. Predictive modeling.

If you're curious how we at Vanagon partner with family offices, operators, and fellow investors to back deep, vertical AI plays— just reply to this email.

Happy to share how we think, how we invest, and where we’re placing our early bets.

Bests
Sandro