Jump to content

SNOW - Snowflake


Seth Lowry

Recommended Posts

This reminds me of the Intrepid Potash deal when I was in banking. Clients calling in and threatening to pull their entire prime relationship if they didn't get an allocation. Potash prices only go up! $659/sh ->$8 over 12 years. Arguably a much better business here but the implied competitive advantage period is astronomical/record/insert verb

Link to comment
Share on other sites

Did some work on this last week - it seems like a great business that I'd like to own. The link below does a comprehensive deep dive. I actually planned on participating at $120 but with where this is pricing ($240 range), it is a good portion of the TAM that mgmt outlines at $80bln. I'm hoping this will follow a path similar to the FB IPO, beautiful business but pricing at IPO got aggressive.

 

https://hhhypergrowth.com/a-snowflake-deep-dive/

Link to comment
Share on other sites

What's interesting about the infographic from the infographic towards the end of this write-up is that Snowflake is kind of sandwiched between a number of different tools.

 

I don't know that their offering is something that exists well on its own. Based on who they compete with (Azure, Google Cloud, AWS) what Snowflake does is just one part of the process. It's important in a way by helping to aggregate date from a variety of sources, but I am skeptical of their solution as it stands (and as I understand it) because it seems like more of a point solution than solving a number of different use cases.

 

In short, I guess I would rather "own the customer" in a sense by owning their data (AWS, Salesforce, Oracle, SAP, etc.) than to be what seems to be a somewhat interchangeable functionality.

 

When I think about which businesses really seem to have staying power, its the companies in the top left part of that infrographic that seem to have durable businesses. The further to the right seems less and less "sticky" unless bundled with other solutions. 

 

https://blog.publiccomps.com/snowflake-s1-ipo-teardown/

 

Not investing in this (or most Saas businesses) but its interesting to think about and try to understand.

 

Honestly MongoDB seems interesting as I understand the pieces more...

Link to comment
Share on other sites

Tweeted this before it opened based on the indicated trading:

 

SNOW raised money at ˜$12bn earlier this year. Either they basically gave away a chunk of their company to private investors, or there's a massive amount of turbo-FOMO going on... Not saying $12bn is the correct valuation today, but 5x that may be a little... extreme?

 

Peaked even higher than that briefly.

Link to comment
Share on other sites

For a 5 year hold at 15% IRR at $75B market cap. currently, market cap. needs to be $150B by year 5. At 20 times forward rev. it is $7.5B or 15 times the ~$500M currently or grow rev. at ~70% annually. At 40 times forward rev. at year 5, it is around $4B rev. or 7-8 times the ~$500M they need to grow rev. ~45-50%. They should be doing probably well in terms of rev. growth right out of the gate that might suppress the need for such high growth rates in subsequent years. One can do similar simplistic analysis for 10 year out hold and/or change the rev. multiple to whatever seems appropriate to each.

Link to comment
Share on other sites

I have absolutely no expertise in the space, so I ask those of you that do. Does Snowflake have an offering that much better than for example a competing AWS product?

 

AWS does have a competing product called RedShift, but the key is that it's not tied to AWS, and allows vendors to move data around to GCP / AWS / Azure. The hypergrowth link from above goes into the technical but might be a little dense to those not in the industry.

Link to comment
Share on other sites

Did some work on this last week - it seems like a great business that I'd like to own. The link below does a comprehensive deep dive. I actually planned on participating at $120 but with where this is pricing ($240 range), it is a good portion of the TAM that mgmt outlines at $80bln. I'm hoping this will follow a path similar to the FB IPO, beautiful business but pricing at IPO got aggressive.

 

https://hhhypergrowth.com/a-snowflake-deep-dive/

 

Agree, it got out of reach when it started getting priced above $80. At this point, they will have to show some "disappointing" revenue figures, either due to consumption habits, or growth pains.

 

To those who complain about its loss-making, you haven't dived deep enough - it's a completely different business than a couple years ago.

 

 

Link to comment
Share on other sites

What's interesting about the infographic from the infographic towards the end of this write-up is that Snowflake is kind of sandwiched between a number of different tools.

 

https://blog.publiccomps.com/snowflake-s1-ipo-teardown/

 

Not investing in this (or most Saas businesses) but its interesting to think about and try to understand.

 

Honestly MongoDB seems interesting as I understand the pieces more...

 

Yes, Snowflakes product indeed is sandwiched between other tools and runs on top of the cloud layer, that’s one reason why their gross margin is relatively low (62%), which makes it even pricier , if you use EV/Gross Margin as a relevant valuation metric. Not that this matters at these multiples.

 

Bill Gurley of course goes bonkers over the first day pop and somewhat rightly so. But then again, there is scarcity in play here - $3-4B in available product is nothing in today’s market. What always surprises me that with an IPO, institutions are not supposed to flip their allocated shares, but if this is indeed true, how would one be able to buy any shares then and why do more shares trade during the first day then where issued? I understand that there are day traders at work here, flipping the same shares many times over.

 

I think the fairer “IPO” would be direct listing with no lockup in place for any pre IPO shareholder.

Link to comment
Share on other sites

I think the fairer “IPO” would be direct listing with no lockup in place for any pre IPO shareholder.

 

That's the argument made by people promoting direct listing or SPAC route.

The counterargument is that both of these are not trivial. And pretty much all three routes are good in bull market (or just market with appetite for IPOs). But all three are not that good if your IPO is not super sexy or market is blah.

 

In this particular case, yes, the insiders/early investors may have captured additional 2x if going direct or SPAC. But it is also possible that by the time they setup direct listing or SPAC, market turns and they are left with less. Overall since even recent private investors got ~2x in IPO, it's tough to feel sad for them. Some people will just have to live with smaller mansions and Porches instead of Ferraris...  ::)

 

Edit: I agree that direct listings should be simpler. I think there's some movement in that direction, but I don't remember details.

Link to comment
Share on other sites

 

I think the fairer “IPO” would be direct listing with no lockup in place for any pre IPO shareholder.

 

The more I see how IPOs are priced and trade the more I agree. It's a big tax Wall Street and the major financial institutions levy against IPO companies, many of which don't need to capital being raised at IPO anymore.

 

Link to comment
Share on other sites

I have absolutely no expertise in the space, so I ask those of you that do. Does Snowflake have an offering that much better than for example a competing AWS product?

 

The best way to think of Snowflake is that they do everything a tad (very small tad) bit better than AWS/GCP/Azure because they are very narrowly focused on their product. They are a bit cheaper than all those 3 when storage + compute are taken into account. They are also cloud-agnostic which is starting to be an interesting play. One very clear differentiator, when compared to AWS's Redshift, is JSON support.

 

Might not seem like a big deal but I contend it is. Today, Snowflake (and others) are focused on SQL (structured) data. There is a whole new world when it comes to unstructured. I think someone mentioned MongoDB above. MongoDB is great but the problem is that the algorithms that are needed to compute over unstructured data (e.g., video, images, text) are still very expensive and haven't really been optimized to embarrassingly parallel levels. Just look at GPT-3, 140+ million features...that's A LOT even by AWS/GCP/Azure standards to do anything with even if you employ transfer learning.

 

There is another data domain that appears to be somewhat untapped - graph. This is where Palantir plays. There are number of vendors here - Neo4j (biggest and probably most mature), TigerGraph, etc. All of these rely heavily on JSON format in some way or shape. (Graph is really a specialized care of SQL that eliminate recursive joins by using pointers [e.g., C++ pointers if you care to Google]). I can certainly see Snowflake move into graph space via parntership/alliance/aquisition and be lightyears ahead of AWS/GCP/Azure.

 

Long way of saying, I'd definitely be a SNOW buyer, maybe once we get a tech sell off.

Link to comment
Share on other sites

I have absolutely no expertise in the space, so I ask those of you that do. Does Snowflake have an offering that much better than for example a competing AWS product?

 

 

Might not seem like a big deal but I contend it is. Today, Snowflake (and others) are focused on SQL (structured) data. There is a whole new world when it comes to unstructured. I think someone mentioned MongoDB above. MongoDB is great but the problem is that the algorithms that are needed to compute over unstructured data (e.g., video, images, text) are still very expensive and haven't really been optimized to embarrassingly parallel levels. Just look at GPT-3, 140+ million features...that's A LOT even by AWS/GCP/Azure standards to do anything with even if you employ transfer learning.

 

 

From what i read in the prospectus, SNOW seems to be suggesting that are able to store and compute over unstructured data, but you are suggesting thats not the case. Who are the key players in this space of handling and computing over unstructured data?

 

Another qst if you have insight. It seems that SNOW charges clients on use basis, both for storage and compute. Since compute over unstructured and graphs seems to require more processing and intelligence, do you think the real money and margins is to be made in this space compared to structured data.

Link to comment
Share on other sites

With the handicap of a very rudimentary knowledge of Cloud based DB technologies, I am trying to get a layman's understanding of what makes SNOW so special. From what I can tell, they provide customers with two distinct advantages:

 

1. Their customer can be cloud provider agnostic since SNOW works with all the three big players. SNOW buys cloud storage and compute capacity wholesale from AWS, Azure, Google Cloud and sells retail to individual customers in a highly customizable format for storage and compute nodes separately with their own software tools which allows their customer to process data efficiently and securely.

 

2. I read on one of the blogs this but i am not sure this is correct: SNOW also provides the advantage of co-location of storage and compute within a datacenter which makes the processing of data much faster.

 

TIA to all the DB/Cloud SW experts on this board.

Link to comment
Share on other sites

Thanks for that color (which I probably 50% understood).

 

Can you expand on MongoDB and its limitations and why more people aren't moving to that standard?

 

tldr; SQL db is still one of the better ways to keep data if it's organized and structure of data doesn't change a lot. Mongo is designed for less than structured data where you may not know what the structure is. Because you get the flexibility of handling data without knowing what data structure will look like, you lose the ability to do a lot of stuff that can be done in SQL-based dbs.

 

This is based on my work with Mongo few years back so things may have changed:

Basically, the way Mongo is setup it allows indexing but not really joins. It allows for faster queirs, but doesn't work if you need joins - basically connecting data across more than one table in the database. That makes Mongo not particularly useful for transactional (think logging of Visa transactions or something like that) systems. There is another concept in databases called normalization - basically, reduce data redundancy which improves a lot of things when it comes to data quality. Because MongoDB is a schema-less db, you can't do normalization so data can be total garbage if there aren't solid rules around ETL processing prior t pushing data into the database. 

Link to comment
Share on other sites

I have absolutely no expertise in the space, so I ask those of you that do. Does Snowflake have an offering that much better than for example a competing AWS product?

 

 

Might not seem like a big deal but I contend it is. Today, Snowflake (and others) are focused on SQL (structured) data. There is a whole new world when it comes to unstructured. I think someone mentioned MongoDB above. MongoDB is great but the problem is that the algorithms that are needed to compute over unstructured data (e.g., video, images, text) are still very expensive and haven't really been optimized to embarrassingly parallel levels. Just look at GPT-3, 140+ million features...that's A LOT even by AWS/GCP/Azure standards to do anything with even if you employ transfer learning.

 

 

From what i read in the prospectus, SNOW seems to be suggesting that are able to store and compute over unstructured data, but you are suggesting thats not the case. Who are the key players in this space of handling and computing over unstructured data?

 

Another qst if you have insight. It seems that SNOW charges clients on use basis, both for storage and compute. Since compute over unstructured and graphs seems to require more processing and intelligence, do you think the real money and margins is to be made in this space compared to structured data.

 

It's a little of an industry jargon but Snowflake processes structured and semi-structured data. Structured - everyday SQL (looks like excel). Semi-structured - XML, JSON, etc. Unstructured - images, voice, video (which is really like images but a lot of them), text. Snowflake lives in Structured and Semi-structured data. They charge for compute and storage but I wonder if compute is where they get most of their money because they would pay for storage (likely through long term arrangement with AWS/GCP/Azure to get the discounts). Unstructured is a totally different animal. Designing systems and algorithms for unstructured data is extremely expensive (infrastructure and talent-wise) and very application-specific.

 

I don't think it would be trivial to replicate their technology. Defiitely doable but probably expensive (though less than $60B). The reason why I say this is because of the architecture they use. It's a hybrid of shared-disk and shared-nothing (computationally). Basically they use a central repository for persisted data that is accessible from all compute nodes in the data warehouse but at the same time they use shared-nothing architecture where they process queries using massively parallel processing (MPP) compute clusters where each node in the cluster stores a portion of the entire data set locally. Basically, this is EMC's Greenplum on steroids without dedicated appliance (device) and on cloud. Snowflake probably has a lot of lessons learned along the way to get their system to work as well as it does with all the features that seamlessly just work.

 

 

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
×
×
  • Create New...