The Palantir S-1 is a long and meaty read, and a pretty fascinating one considering the company was highly secretive, and often controversial, for so many years. It is also written in a very opinionated style: the newly Colorado-based company takes aim at Silicon Valley and is not exactly charitable to its competitors.
Particularly compared to a Snowflake that has had a meteoric rise since inception in 2012 (see our Snowflake teardown here), the Palantir S-1 also presents the picture of a company that, while unique, has had a long road since it was founded in 2003.
It seems that the company went through an important transition in the last couple of years on the product and go to market front – evolving away from a services company into more of a software one – perhaps in anticipation of an IPO.
Ironically, in some ways, this evolution has made Palantir look more like the Silicon Valley companies it feels so different from.
Here are some quick thoughts and notes (from Avery Klemmer and myself)
HIGH LEVEL THOUGHTS
- First direct listing of an enterprise software company.
- So far, only companies with easily understood products and broad user bases have gone public via direct listing – Spotify and Slack in particular (Asana coming up)
- While it has generated a lot of buzz over the years, Palantir has a very different profile: an enterprise software company with few customers (125 in total) and very large contracts (their average revenue per customer in 2019 was $5.6m).
- Its financial profile is also a bit different from the software startups Wall Street tends to love. Revenues are certainly high ($742.6m in 2019), in particular through a very strong story around land and expand (which Palantir calls “Acquire, Expand, Scale”). Its very large annual loss ($579.6 million in 2019, or a net loss of $337.7 million when excluding stock-based compensation), may give pause to some investors, but others may pay less attention to it, considering that many other IPO software candidates exhibit the same characteristic (and Palantir’s loss seems to be coming down in 2020 so far). What’s more problematic is that Palantir’s growth rate is nowhere near the explosiveness of companies like Datadog, Zoom or Snowflake. 2019 revenue represented a growth rate of 25% over 2018 (H1 2020 showed an acceleration, with a 49% growth rate over H1 2019, impressive at that scale).
- Given that, on top of the above, the company has been polarizing, it will be very interesting to see how this plays out in the markets.
- Not your usual cap table:
- S-1s of enterprise software companies often show a capitalization table where various venture capital firms own large chunks of the company, with management owning the rest
- Much less so in this case: Palantir’s ownership has stayed very much “in the family”, meaning essentially Peter Thiel’s galaxy. Founders Fund (Peter Thiel’s fund) owns a chunk, as does 8VC, the venture firm run by Joe Lonsdale, who was a co-founder of Palantir. Joe Lonsdale is also affiliated with another significant shareholder, Disruptive Technology Solutions (related to Disruptive Technology Advisors which acted as investment bankers for Palantir over the years). Worth noting that management has strong ownership (Alex Karp was also part of Peter Thiel’s galaxy early on, after they met at Stanford Law School)
- Only real outsiders seem to be Japanese insurer Sompo (which recently invested, here) and UBS (as well as, reportdedly, a handful of smaller investors like Tiger Global).
- The company will remain tightly controlled after the IPO, due to a multi-class ownership structure where holders of Class B, and even more so Class F (held by the founders of the company, Peter Thiel, Alex Karp, and Stephen Cohen), will call the shots, possibly in perpetuity.
- Impressive product:
- Perhaps our favorite part of the S-1 was to get a glimpse into what Palantir has actually built (particularly considering its reputation as a services company, see discussion below)
- At least based on the description, the company has built *a lot* of product – basically sounds like an entire integrated system that covers infrastructure and analytics in a box, with a core platform and various modules sitting on top of it (our words, not theirs).
- The core platform provides strong ingestion capabilities (lots of connectors to data sources like SAP, AWS S3, and Azure Data Lake), versioning, orchestration, data lineage, security, compliance, ontology, search, etc.
- On top of the platform sits a long list of product (see below) for analytics and visualization
- For any customer to build a system like this in-house, they would need to stitch together many third party solutions which would take a massive effort over a long period of time, and a sophisticated IT and data engineering team
- Therefore, it’s not hard to see why the integrated Palantir platform would be attractive to big governmental institutions and big companies that may not have access to top developer talent to build their in-house data infrastructure. In turn, only large companies or government institutions may be able to afford the high Palantir price tag. Strong product market fit!
- Given the explosiveness of innovation in the data and AI/ML world over the last 10 years in particular, one can’t help wondering how much platform rebuilding the Palantir team had to do over the years, and/or how the platform compares to the various best in class solutions across all those different modules (or whether the key value of Palantir results in all parts working together in an integrated fashion).
- Is Palantir a services company?
- For the longest time, at least in data circles, Palantir was described as an army of consultants that would install “some” software at the customer and would spend months and months customizing it – more of a services company.
- The S-1 does make several references to lengthy implementation processes, for example: “Implementing our platforms can be a complex and lengthy process since we often configure our existing platforms for a customer’s unique environment”
- However, it seems that Palantir made a considerable push to morph into a more standard services to software mix over the last couple of years, perhaps in anticipation of a potential public listing.
- The company’s gross margin now looks much more like a software company’s gross margin: 67% in 2019 (71% when excluding stock-based compensation) and 72% in H1 2020 (or 78% when excluding stock-based compensation).
- “The improvements in our operating results have principally been driven by a significant decrease in the time and number of software engineers required to install, deploy, and manage our software platforms.”
- Death of the forward-deployed engineer?
- For years, a part of the Palantir mystique is that the company had “no sales people, only forward-deployed engineers”. For anyone working with enterprise software startups, the example of Palantir would frequently come up in conversations with entrepreneurs, especially technical founders, as evidence that one could build a company without hiring annoying salespeople (!).
- Palantir has now officially changed its stance, and embraced a more traditional distribution model: “We are investing in an account-based sales force to identify new customers and opportunities. We believe that our decision to grow our sales force in recent years has resulted in multiple new customers in 2019 that are in the Fortune 100 and include leading government agencies around the world. We will continue to expand headcount in our direct sales force.”
- A cloud company?
- For anyone following the data industry, it’s well known that government agencies and big companies in regulated industries (financial services for example) have been slow to embrace the cloud, particularly for sensitive data
- So it’s interesting to see that Palantir, in addition to on prem solutions, has built the ability to deploy in a number of different environments: “a public cloud, a private cloud, on-premises data centers, air-gapped networks in classified environments, edge computing environments, on laptops, and on specialized hardware.”
- Palantir seems to anticipate a lot of cloud deployment in its future: “In December 2019, the Company entered into a new minimum annual commitment to purchase cloud hosting services of at least $1.49 billion over six contract years”
- Founded 2003 to power counterterrorism operations
- Released first platform “Gotham” 2008 for US defense agencies
- Released “Foundry” 2016 to address industry challenges
- Currently powering a large aviation use case (100+ airlines, 9K aircraft)
- Revenue (FY 2017, 2018, 2019): $515M, $595M, $743M
- H1 2020 49% growth over H1 2019
- Gross Margin (FY 2018, 2019): 72%, 67%
- Increased support & cloud costs
- S&M as a % of Revenue (FY 2018, 2019): 78%, 61%
- Started investing in direct salesforce in 2018, only 3% of total headcount
- R&D as a % of Revenue (FY 2018, 2019): 48%, 41%
- G&A as a % of Revenue (FY 2018, 2019): 51%, 43%
- *Note: each government agency counts as a separate customer
- Revenue per customer = $5.6M
- Revenue per customer (top 20) = $24.8M
- Top 20 customers = 67% of revenue
- 53% of revenue was commercial (2019)
- 60% international (2019)
- Top 20 customers have lifetime of 6.6 years
- Contribution margin (Q3 2019, Q4 2019, Q1 2020, Q2 2020): 15%, 33%, 41%, 55%
- Improvements driven by investment in onboarding: customer start up time decreased 5X to avg. 14 days in Q2 2020
- CM = loss from operations + R&D + G&A + stock-based compensation
- Acquire / Expand / Scale
- Acquire = short term pilots, operated at a loss
- 2019 generated $65.4M contribution loss on $0.6M revenue; (109)% contribution margin
- Expand = $100k+ ACV accounts with negative contribution margin
- 2019 generated $75.8M contribution loss on $176.3M revenue; (43)% contribution margin
- Scale= $100k+ ACV accounts with positive contribution margin
- 2019 generated $311M contribution margin on $565.7M revenue; 55% contribution margin
- Acquire = short term pilots, operated at a loss
- Accumulated deficit as of June 30 2019 = $4B
- Ontology management — let customers create a domain-specific taxonomy for their world from “objects” “properties” and relationships that tie them together
- Dat correctness & freshness — users must be able to see full context around a given decision
- Time series analysis on real time sensor data — developed a compression format that improves read performance 2-5X and uses 60% of disk space relative to open source alternatives
- Serving data for 1.3B time series, avg. of 8.8M new points written every second
- Model deployment — enable teams to plug in their own models, see clear metadata
- Gotham enables agencies to identify patterns deep within data sets
- Graph: WYSIWYG drag & drop interface to explore & interact with entities
- Gaia: plan, execute, and report on live map
- Dossier: live, collaborative document editor
- Stencil: structured form-entry tool
- Video: for viewing streaming and historical video
- Table: top down query & analysis tool
- Ava: AI system to scan billions of data points & send out alerts
- Forward: built to make Gotham reliable in unreliable network environments
- Mobile: mobile app for field
- Foundry: data integration and analysis tool
- Monocle: manage data lineage through GUI
- Contour: top down exploration of big data (billions of records)
- Object Explorer: search objects rather than rows
- Fusion: spreadsheet interface
- Workshop: low code application builder on top of data sets
- Vertex: virtualization engine for “What if” analyses
- Code authoring: data engineering tool for transforms
- Quiver: multidimensional charting
- AI/ML: create or apply models to data sets
- Code workbooks: advanced analytics & data science for pipeline building
- Reports: publish dynamic work