Databricks in talks to acquire startup Neon for about $1B

The "datalake" is becoming a bit of a commodity. It's getting pretty easy to spin one up yourself[0] using completely open source components.

Databricks and Microsoft (thru Fabric) are trying to build a complete data platform, i.e. ELT + datalake + BI

My bet with Definite (https://www.definite.app/) has been this is too hairy for a large company to do well and we can do it better.

0 - https://www.definite.app/blog/cloud-iceberg-duckdb-aws

> Folks I know in the industry are not very happy with databricks

Yeah, big companies globing up everything does not lead to a healthy ecosystem. Congrats on the founders for their the acquisition but everyone else loses with movements like this.

I'm still sour after their Redash purchase that instantly "killed" the open source version. Tabular acquisition was also a bit controversial since one of the founders is the PMC Chair for Iceberg which "competes" directly with Databricks own delta lake. The mere presence of these giants (mostly databricks and snowflake) makes the whole data ecosystem (both closed and open source) really hostile.

starbucks drinkers is certainly a new way to describe people, though i'm not sure what image that's supposed to invoke

>Should have sobered up before writing this though. But who cares.

In vino veritas, and all that; we appreciate your honesty!

Very unlikely that Databricks would be acquired by Azure. So much of their business is on AWS, and they are invested in by AWS/Azure/GCP.

Starbucks drinkers? What do you mean?

An OLTP solution fixes a lot of the headaches about the traditional extract-load-transform steps.

Mostly a lot of OLAP starts when the data loads in Kafka logs or a disk of some sort.

Then you schedule a task or keep a task polling this constantly, which is always prone to small failures & delays or big failures when schema changes up.

The "data pipeline" team exists because the data doesn't move by itself from where it is first stored to where it is ready for deep analysis.

If you can directly push 1-row updates transactionally to a system and feed off the backend to write a more OLAP friendly structure, then you can hookup things like a car rental service's operational logs into a system which can compute more complex things like forecasting of availability or apply discounts to give a customer an upgrade for cheap.

Neon looks a lot better than YugaByte in tech (which also talks postgres protocols) and a lot nicer in protocol compatibility than something like FoundationDB.

Alloy from Google feels somewhat similar, Spanner has a postgres interface too.

The postgres API is a great abstraction common point, even if the actual details of the implementations vary a lot.

Actual revenue is irrelevant. This is a business decision to corner the market.

I'm sorry but what is "the idea"? Managed postgres?

It seems like execution >>> idea in this case

Whatever. I was able to set up Neon Postgres in 5 mins. It’s still crazy fast with my Fly services, has replication out of the box and backups. Much easier than AWS and from what I can tell, getting something going with Railway. And I don’t have to worry about operating it. My time is valuable.

supabase lured me in with built-in oauth, real-time, and some nice client side features in their JS lib, but I do worry about the latency sometimes.

It'd be a lot of work to run an apples to apples test with a Google Cloud Postgres db vs. Supabase and see what the difference is.

Isn't this an apples-to-orange comparison?

Neon's multi-region support isn't directly comparable to a single Postgres database in a single data center. You can set up Neon in a single data center, too, and I would expect the same performance in that case.

Meanwhile, if you tried to scale your single-Postgres to a multi-region setup, you'd expect higher latencies relative to the location of your data.

Even managed databases are a scam. You can easily get 10x cheaper pricing for the same workload, by, wait for it, installing Postgres yourself on a baremetal machine. Plus you get much better performance, no noisy neighbors, and ability to actually control and measure low level performance. I never got the hype for serverless. Why are people so allergic to setting up a server? It takes a few hours a year of investment, and the performance benefits are huge.

Capture enterprise AI enthusiasm by providing a 1-stop shop for data and AI, optionally hosted on your own cloud tenant. Keep deploying functionality so clients never need another supplier. Partner with SAP, OpenAI, anyone who holds market share. Buy anyone that either helps growth or might help a competitor grow.

Enterprise view: delegate AI environment to Databricks unless you’re a real player. Market is too chaotic, so rely on them to keep your innovation pipeline fed. Focus on building your own core data and AI within their environment. Nobody got fired for choosing Databricks.

Delta Lake is not catching on, but no worries, they bought Iceberg[0] (the competing standard).

I'm joking, but only a bit. Iceberg is open source (Apache), but a lot of the core team and the creator worked at Tabular and Databricks bought them for $1B.

0 - https://www.definite.app/blog/databricks-tabular-acquisition

It provides central place to store and query data. A big org might have a few hundred databases for various purposes - databricks lets data engineers set up pipelines to ETL that data into databricks and when the data is there it can be queried (using spark, so there's some downsides - namely a more restrictive SQL variant - but some advantages - better performance across very large datasets).

Personally, I hated databricks, it caused endless pain. Our org has less than 10TB of data and so it's overkill. Good ol' Postgres or SQL Server does just fine on tables of a few hundred GB, and bigquery chomps up 1TB+ without breaking a sweat.

Everything in databricks - everything - is clunky and slow. Booting up clusters can take 15 minutes whereas something like bigquery is essentially on-demand and instant. Data ETL'd into databricks usually differs slightly from its original source in subtle but annoying ways. Your IDE (which looks like jupyter notebook, but is not) absolutely suck (limited/unfamiliar keyboard shortcuts, flakey, can only be edited in browser), and you're out of luck if you want to use your favorite IDE, vim etc.

Almost every databricks feature makes huge concessions on the functionality you'd get if you just used that feature outside of databricks. For example databricks has it's own git-like functionality (which is the 5% of git that gets most used, but no way to do the less common git operations).

My personal take is databricks is fine for users who'd otherwise use their laptop's computer/memory - this gets them an environment where they can access much more, at about 10x the cost of what you'd pay for the underlying infra if you just set it up yourself. Ironically, all the databricks-specific cruft (config files, click ops) that's required to get going will probably be difficult for that kind of user anyway, so it negates its value.

For more advanced users (i.e. those that know how to start an ec2 or anything more advanced), databricks will slow you down and be endlessly frustrating. It will basically 2-10x the time it takes to do anything, and sap the joy out of it. I almost quit my job of 12 years because the org moved to databricks. I got permission to use better, faster, cheaper, less clunky, open-source tooling, so I stayed.

when you got out of Spark, what did you go to?

I don't think being cheaper is the main value sell of serverless. When I hear "serverless" I think "ease of deployment and automatic scaling".

I had an interview with a senior data engineering candidate and we were talking about how expensive Databricks can get. :D I set up specific budget alerts in Azure just for Databricks resources in DEV and PROD environments.

There are so many gotchas. I'm getting so tired of working around it, but my company is all in on serverless so the pain will continue. A lot of it is tied up with Unity Catalog shortcomings, but Serverless and UC are basically joined at the hip.

A few just off the top of my head:

* You can't .persist() DataFrames in serverless. Some of my work involves long pipelines that wind up with relatively small DFs at the end of them, but need to do several things with that DF. Nowhere near as easy as just caching it. * Handling object storage mounted to Unity Catalog can be a nightmare. If you want to support multiple types of Databricks platforms (AWS, Azure, Google, etc.), then you will have to deal with the fact that you can't mount one type's object storage with another. If you're on Azure Databricks, you can't access S3 via Unity Catalog. * There's no API to get metrics like how much memory or CPU was consumed for a given job. If you want to handle monitoring and alerting on it yourself, you're out of luck. * For some types of Serverless compute, startup times from cold can be 1 minute or more.

They're getting better, but Databricks is an endless progression of unpleasant surprises and being told "oh no you can't do it that way", especially compared to Snowflake, whose business Databricks has been working to chew away at for a while. Their Variant type is a great example. It's so much more limited than Snowflake's that I'm still learning new and arbitrary ways in which it's incompatible with Snowflake's implementation.

hmm, what is a serverless Pg? I don't quite understand. I thought you needed a database server if you wanted to run Pg.

Okay now I am concerned. We're using Neon. We can move easily at this point, but I'm sure they have huge customers storing many terabytes of data where this may be genuinely hard to do.

I went to Archive.org and figured out that in 2023, they announced they were shutting down on May 30th, all databases shutdown on June 30th, only available for downloads after that, and deleted on July 30th.

They aren’t exactly hiding it. I kept my eye on bit.io because they looked very promising. Next day, gone. Shut down immediately. Something is fucky with the investment pipeline because it’s not ”worth” that much on its own, it’s a market dominance play, bad for innovation..

> Or it's just a business decision to corner the market, as someone else said

Given how lax antitrust enforcement is, probably this

Lucky you, you still can as it's Apache 2 https://github.com/neondatabase/neon/blob/release-8516/LICEN...

I haven't studied the CLA situation in order to know if a rug pull is on the table but Tofu and Valkey have shown that where there's a will there's a way

Why would this give you pause? You just don't want the data to be where Databricks is?

Either way, there are plenty of other serverless Postgres options out there, Supabase being one of the most popular.

Try Supabase!

Later stage things are , the potential IPO is a benefit not deterrent. Recruiters and hiring managers will hint at potential IPO being not far off as an incentive to join. It minimizes risk, they do same for potential target’s founders like Neon here .

This is better than earlier stage startups , while you get far better multiples , it is also quite possible that you are let go somewhere into the cycle without the money to vest the options for tax reasons and there is short vesting period on exit.

For this reason companies these days offer 5/10 yr post leaving as a more favorable offer

——

For founders it is gives them a shorter window to a exit than on their own, and in revenue light and tech heavy startup like neon (compared to databricks) the value risk is reduced because stock they get in acquisition is based on real revenue and growth not early stage product traction as neon would be today .

They also have some cash component which is usually enough to buy core things in most founders look at like buying a house in few million range or closing mortgages or invest in few early stage projects directly or through funds

If they are making money, there is no pressure to raise money from IPO.

Why does it matter if you get liquidity events 2-4x per year

they can do employee liquidity event

Building backends is easy. It is sort of weird. In 2003 no one would bat an eyelid at building an entire app and chucking it on a server. I guess front-end complexity had made that a specialism so with all that dev energy drained they have no time for the backend. The backend is substantial easier though!

These high value startups timed well to capture the vibe coding (was known as builidng an MVP before), front end culture and sheer volume of internet use and developers.

"Easy to provision" is mostly a strategic feature for acquiring new users/customers. The more difficult parts of building a database platform are reliability and performance, and it can take a long time to establish a reputation for having these qualities. There's a reason why most large enterprises stick to the hyperscalers for their mission-critical workloads.

> Someone at AWS probably thought about this, easy to provision serverless Postgres, and they just didn’t build it.

AWS is working on this as well: https://aws.amazon.com/blogs/database/introducing-amazon-aur...

Supabase is not just a hosted Postgres, it’s a full(ish) backend stack built on open source components comparable with something like firebase. But being Postgres, encourages same data modeling (and an escape hatch). Their type generation and SDK is quite good, too. It’s one of my favorite services and powers to projects of mine, soon to be 3.

> Maybe a NoSql DB that “understands” the Postgres API?

I believe there are several of these already, like Cockroach DB.

Supabase is particularly valuable for its users. Or right now “vibecoders”

Who pays $1b for an acquihire?

I'd be more worried that they'd shut it down...

cold starts are 500ms on average, and that's only for the first call that wakes up the db from hibernation. people still seem to think that this latency happens for every call (see other threads here) but once the service has woken up (cold start over) you're back to regular (sub 10ms) latency timings and the service continues to run that way. you'll only hit a cold start again if (you have this option turned on) your service goes idle for > 5 min. You can turn scale-to-zero off and you'll run 24/7, have zero cold starts.

$19 plan is going away, will launch a better $5 plan soon.

A non-trivial portion of my consulting work over the past 10 years has been working on data pipelines at various big corporations that move absurdly small amounts of data around using big data tools like spark. I would not worry about purchasing services from Databricks, but I would definitely try to poach their sales people if you can.

I'd be most concerned with Neon being shut down. That's what Databricks did to bit.io (another serverless Postgres provider they bought).

I'm really not looking forward to a migration.

If you can serve all your traffic by a single instance running Sqlite in same process as your application, have at it.

If you need to serve your dats across a network to many clients, managing that with SQLite is much trickier.

I thought SQLite's use case was for a single-user local database.

Enterprises have lots of data. They store it somewhere, and there are multiple vendors that provide such "credible" infrastructure for this type of storage. Think of it like, your dad says he's willing to get a dog, but only trusts these-five-animal-shelters and nothing else. That doesn't mean that's correct (that those are the only places to get a dog), it just means that's what he trusts. Databricks is most likely a unicorn because they have successfully sold the idea that they are one of those trusted vendors, like Snowflake.

The truth of the 2010s up until now is that every startup was a massive sales con job. The wealth of this industry is not truly built on incredible tech, but on the audacity of salesmanship. It's a billion-dollar con job. That's one of the reasons I take every ridiculous startup that launches quite seriously, because you have no idea just how audacious their sales people are. They can sell anything.

Your question is very fundamental, and the answer is just as raw and fundamental too. I would love it if some of these sales people actually reform and write tell-alls about how they conned so many large companies in their years of working. This content has got to be out there somewhere.