Ask HN: What's so special about Snowflake?

19 points by brightball ↗ HN
I have followed the technology industry pretty closely (or so I thought) for the last 20 years. When Snowflake had their IPO and the stock when through the roof, I was shocked.

1. Because of the stock performance

2. Because I'd never even heard of them

Following the threads on HN, it seemed like a lot of other people were just as confused.

https://news.ycombinator.com/item?id=24265041

https://news.ycombinator.com/item?id=24494633

https://news.ycombinator.com/item?id=24528742

Can anybody explain it to me?

12 comments

[ 2.9 ms ] story [ 40.8 ms ] thread
Good question, but I’m often puzzled about how many things Snowflake doesn’t do well, like DR…
speculation.

their balance sheet is painful to look at (479M sales and marketing costs for 592M revenue, ouch)

Important to keep in mind sales expense often buys multi-year cashflows.

In this case, directly comparing the expense of building a customer base to the revenue from the existing customer base over the same period can lead you to wrong conclusions about the viability of the business.

I would focus instead on the cost of customer acquisition, versus the PV of the average customer's gross cash flows.

------------------------------------------------- ----------------- Lazy Analysis ----------------- -------------------------------------------------

Here is SNOW's latest Q: https://www.sec.gov/ix?doc=/Archives/edgar/data/1640147/0001...

If you look on page 32 we see that the weighted average life of large customer contracts is 1.8 years, and that net revenue retention rate is 169%!

So customers sign up for multiple years, and renew with big purchases.

Turning to the K for some annual figures: https://www.sec.gov/ix?doc=/Archives/edgar/data/1640147/0001...

----------------- Assumptions -----------------

Let's model SNOW's revenue as: The sum of prior year's revenue (existing contracts) plus new revenue cause by the prior year's sales efforts.

On page 42 of the K: We see that each year's gross income is consistently larger than the prior year's sales and marketing expense.

----------------- Example -----------------

In 2020, if we say that all 164mm of new revenue is due to the 125mm of sales expense in 2019, and contracts average 2 years, then we have ~320mm of revenue on 125mm of expense.

By this math in 2021 and forward, it looks like we can expect ~840mm of new revenue from the 293mm of sales expense in 2020, and in 2022 we might expect 1,100mm of total revenue.

----------------- Projections -----------------

In reality, SNOW has projected $1,020 in revenue for 2022. - https://www.cnbc.com/2021/05/26/snowflake-snow-earnings-q1-2...

Note my number is a little high as I did not know what portion of 2019's contracts expired in 2019, so assumed none.

They have very good enterprise sales teams .....

They are also positioning themselves as a firm to firm data exchange hub (similar to how Zoom has positioned themselves as a universal firm to firm video conferencing solution) which allows them to capture a tax every time firms exchange data either externally or even internally (between various divisions).

Technically, I have heard despite their claims that they have a unique "cloud native architecture" [1], they are basically built on a custom distributed parquet system (Can't vouch for the accuracy of this).

[1] https://docs.snowflake.com/en/user-guide/intro-key-concepts....

The uniqueness of Snowflake comes from decoupled compute and storage with data lying at rest on object storage. This was very innovative at the time. The approach is being copied by more or less every competitor in the analytic space so over time it's harder to describe it as unique.
It's a very easy to use and high performance with basically minimal management requirements.

You can also share your data not only across your company but with other snowflake customers with specific granular permissions.

It's amazing.

Snowflake separates compute and data storage which allows you to add more compute during busy times and switch to a lower compute during off peak times. It is also considered to be a data warehouse that is not from a cloud provider(even thought it uses AWS underneath and managed by snowflake).

The company i work for use snowflake. Just like any cloud platform, you can run up the tab if wrongly configured and not optimized.

I really appreciate how they have separated the computational layer from data storage. Want to do a massive pre-calculation without disturbing others? Run it in a separate warehouse.

Also the underlying copy-on-write datastructure is a timesaver, you can clone (like taking a backup & restoring it but with a new name) a 100 GB database in a few tens of seconds or minutes. Also if you run an UPDATE and forgot to include a WHERE clause you can retrive the old data via time travel. You can execute queries directly to an older version of the table (or database) without having to restore it from a backup!

There is an about 500 ms overhead for each query, so it isn't suited for all use cases though.