Martin Casado and Peter Bailis on Next $100B in Data

Videos · Startups · Source

You need to get the data pipeline right. That’s where the art of the company creation is right now

There are two markets in data right now. There is the analytics market (tends to be characterized by SQL and more traditional use-cases, like BI, regression tools, reporting etc). That’s a massive market

The other part of the market is the operational AI/ML world. This has fancier models that you run in production as part of the app

Both markets are converging now. Analytics and operational AI/ML is definitely converging. ELT is another big trend (away from ETL)

The next generation of companies are data companies, and their economic structure is different from that of software

There’s always been a belief that if you do data you have data network effects and therefore you have a data moat. Ironically, it’s almost the opposite. Data moat doesn’t seem to exist, and the more data you have, the more diseconomies of scale you have

💡 The more data I have, the more work I may have. Therefore it’s not just about collecting the data. It’s about setting up efficient processes/tooling and data as a product so that it’s more manageable.

If you don’t see this as a first class problem, you’ll end up as a services company with bad margins, just trying to throw more bodies at the problem

Can you ever make a data company with software company margins? Probably yes, but must figure out how

More data = better performance is only true in companies like Google and Bytedance, where data management is excellent and a first class primitive. In the absence of good data management, more data might not mean better performance

Two layers:

  • Storage layer (good schema etc )
  • Analysis/app layer (smart data manipulation, data as a product etc)

Used to think that we would delaminate the data warehouse. We would just peel off layers and fund companies at each layer of the data warehouse

But now, not so sure. Snowflake, for example, has done a great job at creating a verticalised solution

The clouds have a massive advantage. Everything is a service now. But 2 reasons why they won’t streamroll the entire industry

  • Market Construction: as you move up the stack, your surface area with the market gets a lot larger. If you don’t have a dedicate sales force around niches, you can’t change customer behaviour

It’s not about PLG. When new technologies arise, you need to talk to people about how to use them. Large customers need to have conversations before they use your product

AWS has never put a startup out of business. This is the primary reason

  • In the past, cloud replaced internal IT. But now, it’s become part of COGS. If you’re building Uber, part of your cost model is the cloud. That puts a lot of pressure on the SaaS providers that are using the cloud to cost optimize

Because of this, we’ll see a lot of disruption