Snowflake to buy Crunchy Data for $250M | Hacker News Reader

Snowflake to buy Crunchy Data for $250M

wsj.com

157 pointsmfiguiere60 comments

It's interesting that Snowflake went shopping for Crunchy Data over Neon. While Neon focused on bringing compute and storage separation to OLTP, Crunchy Data focused more on bringing OLTP/PostgreSQL closer to OLAP with DuckDB and Iceberg.

In a way, Crunch Data was a competitor to Snowflake as they literally name themselves as "Postgresql Data Warehouse" but correct me if I'm wrong. Neon sounds more complementary to Snowflake as they were struggling with an OLTP backend, namely their Unistore product, which was announced 3 years ago but never went into general availability due to its scalability issues.

Maybe Neon was 4x more expensive, but this acquisition sounds more like an answer to Databricks than a strategic acquisition if I'm being honest. Apparently, Crunchy had $30M ARR, so it's 8x ARR, which is a cheaper answer to Databricks.

Crunchy is the company with more in depth technical knowledge of Postgres.

A couple of core Postgres members work there and iirc also the guy who spearheaded Heroku Postgres.

> literally name themselves as "Postgresql Data Warehouse" but correct me if I'm wrong

That's not their primary product. Crunchy Postgres is their primary offering and they recently announced Crunchy Data Warehouse.

I thought Crunchy Data Warehouse was their main product, looking at most of their marketing posts. What's the advantage of using their managed PostgreSQL offering on the cloud, compared to native offerings such as AWS RDS and GCP Cloud SQL?

1) built using an open source kubernetes operator, as I understand 2) Crunchy provides true superuser access and access to physical backups – that's huge

Why is that huge out of interest?

Business continuity. If you don't have access to your backups, there's nothing you can do to work around a vendor issue.

Sounds like Stackgres?

Neon acquisition was ~$1B.

We don't know that Snowflake didn't miss out on neon - seems plausible (and silly for neon to not run an acquisition by Snowflake)

> Part of the reason Snowflake and Databricks are interested in database companies is because PostgreSQL can serve as the underlying database for customers to create AI agents with data they store in the companies’ respective platforms.

I don't understand this part. What does PostgreSQL offer here that these vendors believe they can't add to their existing platform? Is it the ecosystem?

I read this as "buying potential competitors off the market" right?

Less ability for customers to roll-their-own => more customers for Snowflake?

With neon being bought by databricks, serverless Postgres tech has effectively disappeared from the market.

What do you mean by this? Neon is still operating the same service.

Source: I work there.

Oh my bad! I was under the (evidently mistaken) impression that since they were bought by databricks they would just become a part of that and cease to be.

Evidently, I was very wrong, which I’m glad to hear tbh.

is it open source? (this? https://github.com/neondatabase/neon.git ) and since it's serverless, in terms of being on the internet what are you saying has disappeared, a proprietary version? support/consulting contracts?

There is also this one from Supabase: https://www.orioledb.com/

Prisma Postgres is a nice choice if you're looking for serverless postgres.

Crunchydata is an excellent vendor and a purist in the ecosystem. The Crunchydata Warehouse product was also extremely compelling.

It’s probably worth it just for their people.

Crunchy has some really smart people working there. Maybe most notably Tom Lane. I wonder if he'll stay on after the acquisition.

Low-latency and cheap retrieval for RAG.

But why do they need serverless Postgres for that?

They could achieve the same with normal pg, or SQLite. Or any number of other embedded DB’s. There’s also plenty of disaggregated compute options available…

I imagine they are buying the expertise in managing the transactional system rather than the IP itself. Operationally running a transactional system is a different ballgame for these OLAP players.

(of course what they're not getting is scale readiness.. it's not like these companies have anything resembling RDS level customer workloads)

I think it’s the large ecosystem of BSD licensed stuff they can fork and relicense as proprietary software.m, because the BSD license allows that.

There are a couple different reasons that make this acquisition interesting.

First is their long pursuit of HTAP and the failures around unistore.

Snowflake wanted to get into transactional workloads for a long time and for good reasons.

I wonder what will happen to Unistore after this acquisition.

The other interesting part is ETL/ELT, CDC and the whole business of replicating transactional databases into OLAP.

What crunchy built with duckdb and iceberg is a potential solution to this problem. A problem that has been painful to solve for a long long time.

Being able to replicate your transactional database into your data lake or data warehouse without having to deal with Debezium and all the rest of the stuff, is going to make many data teams happy.

Bummer that all the postgres serverless providers are getting acquired. First Neon, now this. Hope the innovation and competitive pricing continues!

Sounds like time to build up a new postgres serverless company and get acquihired/exited!

There's still Xata. And plenty of other options that support a Postgres compatible API like CockroachDB and Yugabyte.

The problem is there's so much sprawl in this postgres ecosystem that it seems like no one other than the hyperscalers is really able to reach a escape velocity...

Yugabyte is Postgres from my understanding. They recently rebased from 12 to 15, I think.

I thought they were slowly closing the car in terms of capabilities

If everyone in this sector is getting lucrative acquisitions, that should encourage more copycats.

I feel like the history of anti-trust regulation in America has something to say about that...

That only works if there are moats that discourage new entrants (e.g., buying a social media upstart works because of the network effect).

Not obvious serverless postgres is in that category.

there's only so many buyers

I think $250M is fairly low. Must be a good deal for snowflake.

With Neon going to Databricks, the pool of potential buyers dramatically shrank.

That's such a wild way to view this. I see this the opposite way: the pool of incredibly awesome fantastic postgres technology companies who are uniquely top of the game was down to very very very few.

The musical chairs here is who can get such long proven incredible fantastic well knowing talent. Who can snarf it up & convince these incredible doers to fold into the amorphous indistinct corporate giant.

I wonder what accounts for the gap between this and Neon's $1B price tag. Is the deal structured less favorably for Neon? Does Neon have significantly more revenue?

Seems like Neon raised a lot more venture funding, too.

Value is an intriguing concept. They may not have the revenue to justify their value or maybe they have. The price tag could be the result of an amazing negotiation or could be genuine forward-looking features they had.

You will never know.

They'll earn that much back from anyone doing just a few gigs of data egress transfers.

Congrats to the Crunchy Data team! Thanks for making containerized postgres so easy for years and years. Wish you all the best!

Snowflake is becoming the Juicero of data.

As a DE, I have an unpopular disdain of Snowflake because it trivalize a lot of stuffs. I think I'm going to switch to OLTP given the chance.

Right there with you.

Developed my disdain after having to put up with the incredibly shitty behaviour from the sales and account teams a few years ago.

Sure they had some novelty years ago, but everyone and their dog has disaggregated compute these days, and all their other “feature” just feel like enterprise money extraction that they’ve acquihired in.

Expensive, slow, and painful.

If you're using Snowflake as an OLTP you're looking at the wrong technology anyway.

Can you go into more depth on this?

If you can afford it, I have a hard time coming up with reasons to not use Snowflake.

One good reason is that a huge population of companies just don't have enough data to justify Snowflake. We sell a product built on it, and I wish we'd had DuckDB 3-4 years ago; it's perfect for 95%+ of our clients

I think DuckDB is great but I don't think it is necessarily playing the same game as Snowflake. A lot of people want the serverless option and DuckDB is not that.

DuckDB on a lambda/serverless python function wrapper or in WASM (for client-side workloads) is as stateless and serverless as you can get.

In that case BigQuery is more managed than both, PAYG for analytical queries without thinking about compute nor clusters whatsoever.

BigQuery is more expensive than Snowflake though. You might as well just do Motherduck which would be cheaper than BQ but let you pull data from S3 which is cheaper than Snowflake storage.

Just for the data sharing feature alone it's worth using. It's so damn easy to onboard and maintain data sources when they have a Snowflake share. You don't have to worry each day about loading processes randomly failing and you don't have to write any custom logic to hit APIs and properly flatten and merge responses into the database.

I had them mixed up with Citus and was confused since I thought they already belonged to MS!

Was a cross-over of a number of team members with both product. An easy thing to be confused by.

github repo : https://github.com/crunchydata

Snowflake sounds like nominative determinism. I was just looking at this thing today, totally puzzled as to how to update it and postgres itself without rolling dice that it destroys everything on the cluster that uses postgres. Perhaps someone with k8s experience could explain to me why CRDs are not Singleton hell? The LLMs just run me in circles..

I run a two service cluster in the home lab for fun. I use PVC mapped to a NFS share for the actual data so you could always run a local postgres binary against it. In a production environment I would map these to local disk partitions like you would normally do for a db.

The upgrade process is actually quite nice when it works but it is "another" thing to learn and troubleshoot.

I think of CRDs as a troubleshooting flowchart that someone with more experience than me has put together. When it's right it's great and when it's wrong it makes trouble shooting harder. That is until you remember that the whole point of k8s is ephemeral containers. When one breaks just delete it and let pgcluster CRD resync the data.

any guesses as to whether existing products will remain?

I honestly have no idea what these companies do that makes them so valuable.

I guess you have to be pretty close to C level at a big company to even understand.

Reply on news.ycombinator.com