Description

Long ESTC

Elastic N.V. is a leading provider of search and analytics software. Founded in 2010, the company’s open-source codebase (f.k.a. the ELK stack) has become a foundational technology across much of the software landscape. Trading at ~4x sales, on de-risked FY24 numbers, the current valuation fails to reflect the company’s position as an AI beneficiary.  

Key points:

1.  Elastic represents a pick and shovel play in the new AI data stack

2. 100s of millions in revenue could migrate from Amazon to Elastic Cloud

3. Underappreciated upside to numbers given return to conservate guidance

Picks and shovels

Elastic’s dominant position in enterprise search has positioned it to be a natural leader in vector search, which is a nascent yet essential building block in artificial intelligence applications. Without getting too technical, vector search basically turns unstructured data into coordinates within a spatial graph, where each axis or vector is equivalent to an individual attribute. The closer in proximity two ‘objects’ are within the graph, the greater their likeness, whether those objects are images, words, videos, text, or code. In short, this technology is foundational to a multitude of different AI applications including natural language processing, image recognition, and recommendation systems.

As overall demand for artificial intelligence application grows, the importance of vector search will only increase. More importantly though, thought leaders in technology expect that vector search will play a pivotal role in the enterprise adoption of AI, by making it both safe (enterprises control the data) and relevant (real-time with in-context learning). This is because LLMs are also based on vector embeddings, so by indexing data in a semantically meaningful format, enterprises can communicate with LLMs in their own language. Rather than sending large data sets back and forth, by using vector embeddings enterprises can also pre-process data and only send what’s relevant to a particular query. Over time, vector databases have the potential to to become the storage and memory layer for artificial intelligence.

There are a few startups focused exclusively on the vector database market fetching sizable valuations given the state of the current funding environment. Meanwhile, we believe that the public markets are largely unaware of the advancements that Elastic has made in this area. Elastic has been investing in vector search for the last two years, around the same time that the majority of the startups in this space were founded. With a ~$200M/yr R&D budget, Elastic’s capabilities have improved tremendously over this time period. As a result, the benefits of switching to a niche vector database vendor have been greatly diminished relative to the costs of migrating data, reintegrating data sources, and missing out on the upcoming ‘copilot’ capabilities Elastic plans to launch in its adjacent solutions.

“The thing that makes us unique is that we won’t only do copilots like other security or observability vendors. We can also run an LLM ourselves because we’ve built a vector database. There’s magic in the ability to combine the two for privacy reasons. That’s where things get really exciting for us in the near future.”

(Shay Bannon, Elastic Founder & CTO). https://youtu.be/fdB9W7RD0L4

It seems that many AI models using private data will leverage vector search and while there are private startups like Pinecone recieving large valuations, our checks indicate that it is ESTC's market to lose.  ESTC has not only been investing as long as any private competitor, but it also already has a strong data residency advantage - If you are an existing ESTC customer you are likely to have a strong preference to use a vector search vendor where your existing data already resides.  The vector search technology seems to be a major hidden asset within the company.  In fact, we have talked to some existing ESTC holders that know very little about it. 

Amazon opportunity

Amazon is a latent, multi-$100M revenue opportunity for Elastic. Historically, the two companies have had an acerbic relationship, which dates back to Elastic’s early decision to open source its code under the Apache 2.0 license. While this allowed the software to proliferate in the hands of free users, many of which have become paying customers, it had the unintended consequence of enabling Amazon to build a managed Elasticsearch offering that was rumored to be earning more revenue than the entire eponymous business at the time of its 2018 IPO (Further Reading). Since then, following MongoDB’s lead, Elastic shifted its code base from Apache to Server-Side Public Licensing, eliminating the opportunity for Amazon to profit from the redistribution of its codebase. Elastic and Amazon also reached an agreement on a trademark dispute regarding Amazon’s use of the term “Elasticsearch.” 

AWS to “fork” the codebase at 7.10 (released November 2020) and to begin developing AWS OpenSearch. The further out from November 2020, the further behind OpenSearch falls. In fact, you can see this in real-time with Github insights, in any given month Elasticsearch sees 5-10x the number commitments to main as OpenSearch—when you compound this over years the differences become obvious. This is especially true when it comes to vector search and vector embeddings, which have been introduced to Elasticsearch in its 8.x updates. 

Amazon Partnership

https://www.businesswire.com/news/home/20220518006202/en/Elastic-Announces-Expanded-Collaboration-With-AWS

In May 2022 AWS and Elastic announced a go-to-market partnership. Since then, we have interviewed numerous OpenSearch customers who are in the process of migrating to Elasticsearch (including one of the project’s original sponsors). This decision was made easier by the go-to-market agreement, which allows customers to allocate committed AWS spend to Elastic. Some estimates have placed the AWS Elasticsearch business at upwards of $500M in revenue. Beyond expansion of its existing base and net new customer growth, we believe this represents a sizable idiosyncratic growth driver for Elastic. 

Evidence from customer interviews:

Upside to Guidance

Elastic has returned to a more conservative guidance philosophy which we believe is underappreciated by the Street. For context: in a measure that runs counter to the growth CFO playbook, entering FY23 (ending in April), Elastic messaged to the Street that it was taking a new approach to guidance–one which removed “excessive conservatism” i.e., expect smaller beats. It also introduced a long-term guide for revenue of $2 billion in revenue by FY25, which some have speculated was an attempt to increase the bid in a takeout. When combined, a lack of conservatism embedded in the guide, a long-term target which demanded that revenues reaccelerate into FY24-25, and the company’s outspoken stance that it was not seeing any of the macro headwinds or elongation of deal cycles highlighted by other software vendors made this an easy short. 

In F3Q23, Elastic found some religion. Management told investors to pencil in mid to high-teens growth for FY24 (vs. consensus largely at 20%+) and removed the overhang of its LT guidance. While the macro environment remains highly volatile, we’ve been encouraged by results from MSFT, CFLT, and DDOG and our channel checks intra-quarter seem to indicate a surprise to the upside.

Catalyst

• Earnings
• AI updates - in the same podcast interview mentioned above, the founder / CTO talks about how the company is planning to significantly improve its marketing around its vector search capabilities.