Web 4.0

The internet is (or can be) an amazing place. Tomorrow’s will be even more magical.

Today, though the entirety of human knowledge may be at your fingertips, you still have to do all the work: You search, compare, click, scroll, book, and buy. If you go to a restaurant page today, it’s filled with things that help humans make a decision:

• Photos of the food

• Menus in PDF format

• Potentially a reservation button

• Reviews

All of that is optimized for us — our eyes, our brains, and our clicks.

But in a world of agents and autonomous systems, these representations and design paradigms simply won’t cut it, because the optimization functions for an agent are different from those for an eyeball.

Simply put, we need a new, purpose-built generation of infrastructure companies to support this new world order.

---

Let’s rewind a year or two. ChatGPT is becoming the fastest-growing product in history, everyone is saying AI is a true platform shift, and many VCs are posting some variation of “in a gold rush, sell picks and shovels.”

Every time that meme was posted, we were confused by the over-rotation towards AI infra investments (vs app-layer investments) when history suggested it was still early.

The following is obviously a simplification, but the cycle with the web played out something like this:

1. Technical breakthroughs at the network/storage/compute layers that laid new technical foundations

2. Next, app developers built new, magical experiences on top of those foundations and pushed the boundaries of what’s possible. And as they did, it exposed unforeseen cracks in the foundation.

3. Another layer emerged in between to fill those gaps and optimize the developer experience, which helped those apps scale to greater heights.

When the cloud took off with Web 2.0, apps got more powerful — but also more complex to build, run, and secure. That created massive demand for a new generation of infrastructure companies.

IVP backed several of them, like Crowdstrike, Datadog, Rubrik, Hashi, AppDynamics, and PureStorage. Today, these companies are worth more than $200B today (realized market cap, not private valuations). And by backing them, we learned a few things:

1. The most significant outcomes often come from problems no one notices — until something breaks.

2. Shifts in building and delivering software always create room for new platform winners.

3. When adoption starts, the best tools scale faster than any company before, and become embedded everywhere.

Right now, that cycle has begun for Web 4.0. The only difference is that it’s happening much faster than ever.

We’re already starting to see some magical applications built on those foundations, like Perplexity, Abridge, Glean, DeepL, Cradle, and Laurel. It used to take these kinds of companies three or four years to really scale. With AI, we’re seeing this happen in six months. And just like in the past, these apps expose problems that will be solved by the emerging generation of infra companies.¹

---

So what’s broken today, and what new products do we need to fix it? This list constantly changes, but we’re focused on six key areas for now:

Fast, scalable model deployment

AI models are powerful, but getting them into real-world products is still surprisingly hard.

We’re fortunate to be investors in Baseten, which makes it easy for developers to quickly, reliably, and securely roll out AI models to their customers and build these magical application experiences.

Observability for the thinking web

If machines are making decisions, we need to know why. You can’t just monitor uptime anymore — you need to monitor reasoning. Why did the system take that action? What data did it see?

Agent authentication and authorization

Trust is the new frontier. If software is acting on your behalf, how do you prove it's doing what you asked for—and not being misused? We need secure delegation, agent identity, permissions, and audit trails—an entirely new layer of digital trust.

Protocols for agent collaboration

All these systems need to talk to each other. In a human-run world, we built simple APIs. In a machine-run world, we need inter-agent protocols that extend and interact with APIs. MCP adoption is exploding, but we likely haven’t seen the final form here.

Storage to support the unique characteristics of AI/agents

Databricks’ acquisition of Neon highlights the different requirements for agent backends vs. traditional data stores. And Clickhouse is expanding beyond its observability roots to help power AI workloads. But we think it’s still early here, and we’re excited for new data stores (HTAP? Caching?) to emerge.

Interfaces for machines, not just people

What’s the supercharged version of robots.txt that doesn’t just provide instructions on what to crawl, but also optimizes agent navigation to ensure successful task completion? A beautifully and thoughtfully designed UI, while pleasant for you and me, is largely irrelevant to a bot.

---

Dawn of a New Web

Now that the internet has a brain, it’s time to build a whole nervous system, one that translates thoughts and ideas into actions and helps create even more magical experiences. No doubt more categories will emerge, and this post doesn’t cover all the other great infra products that take advantage of AI.

So yes, it’s finally time for the picks and shovels. It’s an incredibly exciting time to be an infra investor.

1. There are obviously tons of other amazing apps and infra products that we aren’t investors in. Since this was a presentation for our LPs, we only focused on our portfolio companies.