We attended Snowflake’s Data Cloud World Tour where we met Frank Slootman (CEO), Benoit Dageville (co-founder), employees, clients, and developers. Benoit is a passionate innovator that makes many French computer scientists proud and Frank represents the ‘Dutch’ no-nonsense execution.

The physical moats of businesses like Coca-Cola and Union Pacific are clear. Even if a new competitor would have USD 100 billion, it would probably fail to replace their network.

We see the moats of various European family businesses that already for decades have been making specialized machines for manufacturing that require niche technical knowledge. These machines and robots enable others to produce the highest quality products at the lowest cost.

It is also relatively straightforward where to find the competitive edge of businesses that combine cutting-edge hardware with software, e.g., ASML, Apple, AWS (Amazon), Tesla, Azure (Microsoft), and Nvidia. The moat is not just around the technical knowledge of combining hardware and software but also comes from managing the complex network of specialized suppliers.

It becomes more challenging to get conviction on what the lasting moat is for the consumer-facing pure software businesses (B2C). For example, Spotify. How durable is engagement and how well can it be monetized?

Even more abstract is how to define a moat in the intangible (pure software) business-to-business (B2B) software world. We clearly see clear moats around both Visa and Mastercard as the network is dominant and hard to replicate. How about Snowflake? Does it have a moat?

An important aspect of a moat in enterprise software is to be found around sticky client relationships that exist because switching costs for the client to move to another platform are high and because the product delivers a good value proposition where the price is a fraction of the perceived value that the client derives from using it. Pricing power is a derivative of this. Microsoft Office and Revit (Autodesk) are good examples. If the platform is able to get scale, distribution advantages, and pricing power, then it can be possible to get confidence around the moat and the durability of growth. In contrast to the Coca-Cola Company, where repricing is a daily topic, this period is the first time most tech businesses are experiencing an inflationary environment in the United States and Europe. We need to see more evidence of pricing power as well as intelligent capital allocation.

The word ‘cloud’ does not reflect the reality of building and managing large modern data centers. Hardware needs to be built and the racks, power supply, and air conditioning, can break and needs to be fixed. The cost of a new single-story AWS hyperscale data center of about 20,000 sqm is north of USD 1 billion. The entry barriers here have become significant also because more efficient computing power is coming from ever more powerful custom-designed chips.

Moving workloads to the cloud is just the beginning. The real productivity gains come when people start to leverage data to optimize decision-making and create new things. Besides Amazon, Microsoft, and Google, two relevant platforms have emerged that enable better data management: Snowflake and Databricks. Both teams come from different backgrounds, have strong fundamentals, and have different approaches to building modern data architecture. We own shares in both businesses and are following their chess game closely.

In contrast to AWS and Azure, Snowflake does not build its own compute infrastructure and instead runs ‘on top of’ the main public clouds. About 80% of Snowflake’s workloads run on AWS and 18% on Azure; Snowflake is both a large client as well as a competitor for the two dominant cloud players.

Then where is Snowflake’s moat? At what point will AWS be able to offer a superior value proposition because it owns the compute infrastructure? Is Snowflake basically the client acquisition tool for AWS until it decides that it likes to eat Snowflake’s lunch? Would it ever make sense for Snowflake to build its own infrastructure or is a focus on the core capabilities a superior strategy? How significant are switching costs for Snowflake clients that have most data in its data cloud and collaborate using the data-sharing marketplace? How successful can Databricks or Google be in positioning itself in a way that may compete with Snowflake’s cloud-neutral position that reduces data silos across clouds?

Enterprise relationships gradually shift over time. In the 1990s IBM and Oracle dominated the space. The way Microsoft became a B2B leader is one of the greatest turnaround stories in history. In contrast, Google has great technology but lacks the same relationships which makes it more difficult to scale its cloud services. How many pricing incentives does Google have to offer to convince businesses to pick its technology versus that of Snowflake? We never underestimate Google; it is competitive and a leader in machine learning.

Snowflake is adding value to the hyperscalers. This is best reflected in the partnership that both AWS and Azure have with Snowflake. The following recent remarks by Michael Scarpelli, CFO of Snowflake, illustrate this point well:

‘What I would say is AWS of the three clouds is probably the friendliest to Snowflake. And I think last year, we co-sold about $1.2 billion with them. Microsoft is next, and GCP is the one we have a relationship that’s not very strong. And GCP is the most expensive cloud for us to run and their pricing is not very good. AWS gives us great pricing. We’re in the midst of renegotiating a contract with AWS, and I know Microsoft will follow. It was funny, everyone was concerned last quarter about how can you guys be growing when Microsoft was seeing weakness, and that’s why I pointed out that 80% of our business is in AWS. It was funny that this got the Microsoft execs to actually call us to tighten their relationship with us.’

It is remarkable how good Snowflake seems to be at onboarding the world’s largest enterprises. Clients mention the ease of use and cloud-neutral architecture. This growing relevance in being an important part of the modern data stack is one part of the moat we recognize at Snowflake. The data cloud includes a data-sharing marketplace that creates a network effect because once a client is sharing data switching costs become high.

Snowflake is one of the faster scalers we know. Once clients are onboarded, they tend to consolidate on Snowflake which is visible in the high net and gross retention rates. Capital One massively ramped up its spending on Snowflake from USD 1 million a few years ago heading toward USD 70 million. This exponential growth is possible because Snowflake has a consumption pricing model where clients pay for compute power consumed. This model can be more scalable than a SaaS model where clients pay for the right to use the software which is ultimately capped. As a result, the path of Snowflake may look more like that of AWS which is now at a USD 82 billion revenue runway.

Most businesses will not be able and should not build their own physical infrastructure but leverage the cloud infrastructure instead. From an investing perspective, the splintering of the layers but standardization around the pillars is interesting. What company is a pillar and is allowing access to partners to leverage the platform? Does this result in stickier client relationships? Snowflake has already become an ecosystem of partners.

Thinking about moats is interesting and never easy. For example, Amazon Prime has opened its logistics network to outside merchants. How does this impact the moat of Shopify which so far is a pure software business?

We are seeing many businesses arising on top of open-source platforms. For instance, Elastic on Elasticsearch, Databricks on Apache Spark, and Confluent on Kafka. These businesses must make sure their value proposition is good enough to attract organizations to pay for a hosted product rather than pay dedicated FTE that manage the software internally. How sticky are the client relationships once they started paying?

Over the past few weeks, we have seen enormous (USD 100m+) ‘seed’ funding rounds in the field of generative AI. For example, Stability AI and Jasper AI. These teams develop use cases on top of mostly open-source machine learning models. We will see where the lasting moats will come from for these businesses and therefore where the value will accrue.

What seems clear is that these breakthroughs in AI unlock creativity at the tail end of creators and all interactions with these models increase demand for computing power and therefore demand for AWS, Azure, GCP, as well as for Nvidia’s GPUs. Every person and machine is now feeding data into the data clouds. The prompt for Stability AI’s DreamStudio to generate the image based on the text ‘the quest for the moat of Snowflake, by Casper David Friedrich, matte painting’ is just adding incremental demand for compute.

What is not reflected in today’s prices? In the case of Snowflake, we think the valuation today implies that management will hit the target revenue guidance of USD 10 billion in fiscal 2029. Even management seems to think that this is a low hurdle and if this is achieved earlier with a 25% free cash flow margin then the stock is likely to double in the next 3-4 years. Is that good enough versus owning the hyperscalers that have become attractive based on net earnings? Over the past decade, the public has permanently underestimated the addressable market for the leading cloud businesses and as a result, Amazon and Microsoft have been undervalued most of the time. Cloud is one of the biggest and fastest growing industries and it is likely that platforms that are relevant in this ecosystem can achieve significant scale as long as the moat is growing daily.

This week we saw a slowdown in growth at AWS and Azure; in the third quarter of 2022, AWS grew 28% YoY at a USD 82 billion revenue runway, and Azure grew 42% YoY at a USD 50 billion runway. There will be more focus on cost efficiency both at the hyperscalers and at their clients. However, this does not impair the long-term earnings power and moat of the leading cloud businesses. It is possible that there will be more focus on deepening valuable partnerships that are bringing business to the infrastructure clouds.

We have experienced a historic drawdown, especially in tech-driven businesses. This is painful and interesting as well because what are the businesses today that are comparable to what Amazon and Salesforce were in 2008? In other words, what business gives us a high conviction of durable growth and would be able to increase the market value at least tenfold over the next ten years?

Our focus remains on leading tech-driven businesses. What really matters to us is to be right about the quality of these businesses. If we are correct about the fundamentals, then share prices will take care of themselves over time.