Snowflake, Datadog, Cloudflare, anyone?

I feel we didn’t get a lot of discussions about these three companies recently, so below is a quick summary of my current thinking about Snowflake, Datadog and Cloudflare, which combined make up almost 50% of my portfolio. Detailed thoughts and discussion of their numbers can be found, among others, in my last monthly recap, here: Ben’s Portfolio update end of August 2023


Given the current economy and generally speaking, I view consumption based businesses as currently having an edge over subscription based business. That’s because we have been going through a deep decline in revenue growth rates in the last 12+ months and I believe that when they turn, revenue growth rates will turn around quicker in consumption based businesses. So, if a turn-around is imminent - and I believe the data we saw in Q2 shows strong hints of that - consumption based businesses will re-accelerate faster than subscription based businesses. Combine that with the prospect of AI revenues trickling through from GPU sales (Nvidia) to software & compute consumption (e.g. Snowflake) in the near future. This sets up a nice narrative for Snowflake’s near-term outlook, which builds on-top of the original long-lasting growth story of Snowflake.

Coming to the numbers, I think the short term guidance suggests that Snowflake will stay close to 8% QoQ revenue growth in the next two quarters which still compounds to 36% YoY growth (note: I had a mistake in my guide number for Q3 in my last recap - goes to show you should always do you own DD!). My main take-away from the secondary metrics is that, surprise!, RPO growth accelerated again after it saw a largely criticized drop in Q1 - not the first time this happened in SNOWs history and it’ll probably not be the last time. So that, and their great progress on the data-sharing front (fly-wheel!), as well as their continued, although wobbly, progress on gaining new customers, nicely supports my hypothesis of future growth re-acceleration or at least growth durability. Combine that with their outstanding profitability developments and we have what I believe is a great investment - still what I feel under-appreciated by the market and even by many growth investors, with great narrative outlook and healthy secondary metrics supporting that narrative.


Datadog’s narrative feels like it is in a similar situation like Snowflake’s. Their last earnings report, largely misunderstood by the market Datadog: No, Its High Growth Era Isn't Over, It's Just Starting (NASDAQ:DDOG) | Seeking Alpha, didn’t look very impressive at first sight. But looking a level deeper reveals some great things happening with this company. The short story in my mind is that revenue growth and net new sequential revenue growth already showed a big turn-around in comparison to Q1 and while still hampered by seasonal headwinds in Q3, I believe they are bound to continue revenue re-acceleration in Q4. What makes me believe so? We need not look further than RPO and billings, which are great leading indicators for revenue growth: Both jumped up by a lot in Q2. So while we can scratch our heads, trying to interpret their current NRR and customer growth declines, the latter of which, as Bert Hochfeld says in above referenced article, might just be due to a focus on their largest customers, RPO and billings tell a clear story. And Datadog’s multi product expansion tells us as much: Their customers love their products and continue to get more of them. So again, just with Snowflake, we have what I feel is currently an under-appreciated company by the market with great mid-term narrative outlook and secondary metrics supporting that.


Cloudflare’s revenue guidance for Q3 and implied guidance for Q4 suggest revenue growth will accelerate QoQ not only from Q2 to Q3, but also from Q3 to Q4. What is there not to like? Especially if combined with large customer growth AND RPO growth re-accelerating in Q2, after it had dropped sequentially in every of their previous 3 quarters. All that while they continue to expand their profitability margins. Another great setup in my mind.

Other thoughts?



well articulated…
I have exactly the same opinion… their q/q $value growth already rebounded in June / July end reports. Next report will also have easier comps… fast forward to early next year and liky there will be see mild but sustained acceleration in growth rates…

Also MDB, CFLT, ESTC, BRZE and TTD likely to see growth rate improving within next two quarters… benefitting from easier comps, end of “optimization”, push to adapt AI.
(and for TTD with political ad spend.)

NET and ESTC look to be best value today among these on TTM basis
for the rest, they look expensive but we do not know the upcoming level of acceleration… high level of acceleration may make these look cheap in the hindsight


Snowflake is overweight, @ 20%, in my portfolio.

I often go back and read what Muji wrote in June and what Bert wrote a month ago where they both seemed to ‘bang the table’ for buying Snowflake.

My thesis for investing in Snowflake is mostly around the success of their announced partnership with Nvidia, in June (I own ~5% position in Nvidia).

I continually rewrite what Muji and Bert have written, into my own words, and then compare for semantics. I do this so that I can make sure I under stand things. Muji and Bert have been great in their responding to my emails asking for clarification.

The following is what Muji has called, “a Big deal!”.

NVIDIA’s Enterprise AI platform can run anywhere there is an NVIDIA GPU. On a workstation, in a cloud env, or on a DGX supercomputer.

What Snowflake is now allowing is running it within their data clouds, so you don’t have to move data when combining your data with others in your industry (The Nvidia AI Enterprise runs over the GPUs Snowflake has, making it available in Snowpark Container Svcs).

The Software, for Building Enterprise specific “Accelerated” (pre-built & pre trained for industry specific insights, that then only need fine tuning) Production AI models, is Nvidia Enterprise AI

I believe this (the combining of Safe Data Sharing, within Snowflake, with the Capabilities of Nvidia Enterprise AI) is what Snowflake CEO, at the Snowflake Summit Keynote, with Jensen Huang, CEO/ Founder of Nvidia was taking about when he said, “This is The holy grail (Enterprises) have been waiting for over 40years.”!

Will the numbers follow the narritive for Snowflake, soon? I submit that this is “the why” for Enterprises and Hyperscalers to have been buying all those GPUs.




I am also hopeful for a re-acceleration of snowflake growth but am waiting until the numbers start to show this is actually happening. I believe a basic pillar of Saul style investing is that the short-term prospects looks good as well as the long-term so i try to follow that. Hardware is obviously benefiting from AI, the next phase will be software. I’m hoping next year that companies like snowflake and data dog will start to see a large benefit, but again, I’m waiting for the actual numbers to prove this trend.


Same, especially with Slootman’s quote along the lines of AI still being mostly in hype mode with effects likely not showing up for SNOW until sometime next year.


“Moving the benefits of AI from ‘hype-mode’ to providing real competitive advantage for an Enterprise requires a Data Strategy.”, Slootman.

Christian Klinerman, SVP or Product in this recent hour long interview, at about 15 minutes states, “Data is 90% of the value in gathering insights vs 10% for algorithmic models.”

This is said to be due to model commoditization and the price of compute falling precipitously.

How long it’ll take the leaders of Enterprises to figure this out is not something I’m going to try and guess the timing of.




You used the term “under-appreciated” both with Snowflake and with Datadog, but they are among the top 4 most expensive cloud stocks on the market, out of dozens ( Yet others like Samsara are likely to grow faster over the next year. So in what way are Snowflake and Datadog under-appreciated? If anything, they seem extra-appreciated. The market seems to be paying up for some reacceleration.



True, they aren’t exactly the cheapest stocks. On the other hand, you have ADBE in the top 10, a quite large company trading at 11x NTM revenue. So I’m not sure what to make of it.

In any case, just a reminder that Cloudflare’s birthday week starts tomorrow. In order not to spoil anything, judging from recent activity in their (public) code base, I do not expect anything interesting anytime soon. :wink:

Specifically, there’s absolutely no indication that they’re working on something called Vectorize, Cloudflare’s new vector database.

Likewise, it does not look like they’re working on something called Workers AI.

But in a purely hypothetical world:

Workers AI looks intriguing. Seems related to, but (currently) distinct from Constellation, and geared towards easily running select AI models. Notably, there are a lot of references to a 7 billion parameter LLaMa-2 model, which arguably isn’t a tiny model.

Where they’re going with this remains to be seen, but a potential upcoming “Build and deploy a Llama 2 Worker” (i.e. LLM) template in the dashboard feels a bit more ambitious than low-latency inference for an intelligent doorbell.


Hi Bear,

It’s a fair question. Normally I don’t talk about valuation, but I realize by saying what I said, I just did so. With the statement you quoted, I really didn’t want to get into a discussion of classical valuation metrics, although the board is very much welcome to have this discussion; I will just stay out of it. My comment about “under-appreciated” was more to say that I believe the stock price of both Datadog and Snowflake will be significantly higher in the future - but such believe is implicit in anyone investing in anything - So the part of my post you quoted didn’t really add anything valuable to it. So let me try to circle back to the fundamentals: Do I understand the market’s reaction to Datadog’s earnings? Well, I can certainly rationalize it to be a result of their lowered FY guidance. Do I believe Datadog’s stock price should have dropped almost 20% after their last report? No, because their QoQ increase in net new revenue, billings and RPO tell me that there is an increased likelihood that they are getting out of the macro-weeds and picking up their growth story again. But that’s obviously just me. And as outlined in my original post I have similar thoughts about Snowflake.

Hope that clears it up a bit!



Raylight, please correct me if I’m wrong. What you’re not saying here is that the fine- tuning of ‘select’ models, Llama-2 for example, could never be accomplished in a Cloudflare POP and then a Worker’s created App may never have it’s weights updated, all at the edge?

Please say ‘never’:joy:

What am I saying, of course it’s possible. Cloudflare has been putting Nvidia GPUs into Cloudflare Points of Presence for a long time. And Muji has made it clear that, “ NVIDIA’s Enterprise AI platform, which can run anywhere there is an NVIDIA GPU. On a workstation (for example), … “
I’m remembering from a few posts ago, ‘The Software, for Building Enterprise specific “Accelerated” (pre-built & pre trained for industry specific insights, that then only need fine tuning) Production AI models, is Nvidia Enterprise AI.’. Nvidia Enterprise AI having smaller models, I’m guessing, nonetheless similar to LLaMa-2.

Is it much of a stretch to see, if Birthday Week goes as expected, with Workers AI and Cloudflare’s proposed Vector Database being announced- perhaps CEO of Cloudflare, Mathew Prince, wasn’t really making a mistake when he said Cloudflare would be the Fourth Public Cloud (I say mistake because he very quickly changed the messaging to ‘Cloudflare being the glue that stitches together all the public clouds.’)… No, I think he meant that Cloudflare will eventually be the fourth public cloud and the glue that stitches together all the public clouds.

But, since I and at least Raylight are here talking hypotheticals, let’s add to them that Snowflake will never be layered onto Cloudflare’s POPs. Snowflake does however make many of their services available to individual private (on Premises) clouds, presently.

How useful is any of this for present investment, perhaps very little to many. But, things are changing, faster and faster.




Presently, Workers AI looks like something that offers a catalog of AI inference models that can be accessed from a worker or via API. Essentially, Constellation with a greatly simplified interface, and without the ability to upload your own models. Looks like the two will co-exist “for now”. So, inference only.

We’ll know soon enough.

It might not sound like much, but keep in mind:

  1. Currently, Constellation is CPU only with a model size limit of 50 MB.

  2. As far as I know there’s no evidence that Cloudflare has been doing GPU stuffing (on the edge) for a long time.

  3. A 7 billion parameter LLaMa-2 model is two orders of magnitude larger than 50 MB.

What I’m looking for is any kind of update on what’s going on at the hardware/infrastructure level. It’s a non-trivial architectural problem, and they’ll want to be clever about it.

It’s important to note that GPUs and/or custom ASICs goes beyond a single use case of bringing “inference at the edge” to developers. Like the rest of their hardware, they’re taking internal workloads and use cases across all their services into consideration.


This was addressed to S&F… but from my perspective, what is under-appreciated is the potential for this companies to see re-acceleration of growth…

“optimization” took away anywhere between 5% to 15% of growth in last four quarters… just that “end of optimization” should result into re-accel… and ofcourse AI apps taking further from there…


Snowflake mentioned that most users who were optimizing 2-3 quarters back are done with their optimizations. I don’t know why you would think there should be re-acceleration from the stopping of optimizations though. What they optimized were queries and they won’t go back and de-optimize them again. Maybe any new queries/processes won’t be optimized but I don’t see any re-acceleration gained from what was already optimized.


An example, when a family feels their finances are not good during the economic recession they will only spend on basic necessary expenses and cut back on all spending, but once they feel comfortable with finance they will increase spending(restaurant, travel…).

I think that is why Bert said SNOW will reaccelerate. When you want to see the evidence and make a decision that will be too late, at that time the valuation may be too high.


agree, this is not what I meant.

Think of it this way: Last Q SNOW grew 37%… this was a result of higher growth (due to new customers and expansion) subtracted by some of those who were still optimizing…

Say for example: growth was a result of 52% growth due to new customers and expansion subrtacted by 15% reduction by those who were optimizing…

that 15% drag would not occur again while increase in new customers and other customers expanding continues…

So ofcourse, in this example, I am not suggesting that there will be acceleration of 15%… due to law of large numbers AND slowly saturating expansion, it will be lower… but for sure as 15% headwind goes away will produce 37% at the base and may be somewhere in 40s% more likely for next few quarters.

this is just illustration… I do not think any of these companies quantified the optimization headwinds… I wish they did… but at-least for new few quarters, chances of small to more than small acceleration exists for consumption model businesses IMO

hope this makes sense.


Some great discussion here. A few thoughts.

I do believe the AI workloads will be a big growth driver, however, do not overlook Containers. Once that goes GA, that by itself has potential to be a big growth driver and positions Snowflake to be a hyperscaler built upon hyperscalers. Sure, many companies will utilize containers for AI workloads (using Nvidia NeMo models stood up on those), but there will also be many many many companies opting to use Snowflake containers instead of an Azure or AWS VM (or container) to host their internal applications. If a company needs an application running 24x7 and they stand it up in a Snowflake container, that is 24x7 Snowpark compute going directly to Snowflake instead of AWS or Azure. Sure, they’ll pay AWS & Azure for the compute they are consuming as well so its not a huge loss for AWS/Azure either.

On optimizations and why when a company finishes their optimization cycle, their growth will accelerate (not as fast as previously but faster than during their optimization cycle):

Once companies complete their optimization processes, we should see the revenue of those companies growing. One easy & recent example is the recent IPO, Instacart. They grew very quickly during the COVID timeframe. So fast that the development and such was done in a way that was a lot more costly than needed. Once they got to a spot where they were able to focus on optimizing, they were able to cut their spend by 50% (see video of Instacart presentation from Snowflake Summit). However, if you look at their Snowflake spend, it didn’t drop 50%, iirc, it was just slightly lower from the previous year. That 50% is based on what the spend was before optimizing and calculated savings as the optimizations went on. During the entire time, they accumulate more data and add more workloads.
So moving forward, Instacart should continue to increase their spend due to additional data (causing current, optimized queries, to take a bit longer over time) and due to additional workloads.

Finally Foolin