Palantir Technologies Inc. – Palantir continues executing mission-critical objectives for government customers, making an immeasurable contribution to Western society amidst global conflicts. Unfortunately, the commercial segment appears likely to underperform expectations baked into the stock given the increasing competitive intensity on their new “AI" product, setting up PLTR for 30%+ downside.

Since its founding in 2003 amidst the restructuring of the US intelligence apparatus in the aftermath of 9/11, Palantir Technologies Inc. has played a crucial role in powering Western governments with modern software on the frontier of extracting signals from data. Over the last two decades, their technology has evolved with new capabilities emerging with each new buzzword, from “Big Data” in the 2000s to “Generative AI” today. From afar, it’s easy to underappreciate the role Palantir has played in protecting the US and its allies given the classified nature of their government activities, including their reported inclusion in the technology stack intelligence operatives used to locate UBL in Operation Neptune Spear in 2011. More recent projects include working with Health and Human Services on the spread of COVID-19, Immigration and Customs Enforcement on border control, and the Army on aggregating data from disparate sources (drones, satellites, sensors, etc). Palantir’s success in supporting important objectives for government customers deserves sincere appreciation and support for their contribution to society.

Even as far back as the early venture rounds, the investor debate came down to bears asserting that Palantir was far more of a consulting shop than a software business, and thus deserved an ultimate valuation multiple closer to single-digit forward revenue and high-teens/low-20s forward FCF territory for services-heavy comparables. Given the 257% run-up in the stock from closing at $6.00 in December 2022, that strong form of the bearish thesis is no longer necessary for attractive risk-adjusted returns shorting PLTR. The government business demonstrates strong competitive advantages that should support durable growth within a sticky customer base. Unfortunately, the commercial business does not. The aggregate business consequently feels far more likely to disappoint than live up to the lofty expectations baked into the stock, and underperformance against recent hype around “Generative AI” is expected to be the near-term catalyst to that end. In addition to announcing underwhelming guidance on topline growth, 2024 Q1 earnings out yesterday signal a tipping point in the burst of momentum in the commercial segment decelerating, with close today down 15% to $21.40, now valuing the company at $44B EV (15.6x NTM Revenue, 19.2x NTM Gross Profit, 54.5x NTM FCF). That’s why the company is highly overvalued based on its DCF valuation.

Fair warning that shorting PLTR is not for the faint of heart, and may fall into the “too hard” bucket outside of long-term oriented investing styles that can get away with stomaching volatility (eg, long-biased, software-heavy portfolio with short PLTR exposure alongside software longs with thematic overlap). Currently 3.5% short interest, down from a high of 7.8% in October 2023. Imagine this forum embraces that fairly priced equity is a public good in well-functioning financial markets, necessary to support optimal capital allocation across the economy (and from Palantir’s perspective, better to give employees fairly-priced equity than over-priced equity). Palantir’s CEO has a long history of getting emotional about stock market participants, as in March 2024: “I love burning the short sellers. Almost nothing makes a human happier than taking the lines of cocaine away from these short sellers who, like, are going short on a truly great American company, not just ours but just love pulling down great American companies so that they can pay for their coke. The best thing that could happen to them is that we will provide, we will lead their coke dealers to their homes after they can’t pay their bills.” (1) While the CEO seems not to understand that “the stock is not the company, and the company is not the stock” (2), will intentionally refrain here from getting into further detail around uncertain management quality that has been covered elsewhere. Ironically enough, short sellers are more aligned with the CEO’s own stock behaviour than anyone long PLTR… the CEO has also put downward pressure on the stock, selling over $1B of PLTR since its IPO in September 2020 without buying a single share. (3) Of course, why spend your own money going long your own stock if you’ve already got guaranteed RSU + options vesting with no performance-based criteria beyond merely going public?

Plenty of other interesting things going on under the hood at Palantir (eg, Palantir investing in SPACs that then became customers, relatively shareholder-unfriendly governance, granular public disclosure on federal contracts, eroding go-to-market efficiency,  etc), but this idea framing will focus on the fundamentals of the government segment, Palantir’s tenuous competitive position in “Generative AI”, and what that means for intrinsic value.

High-Quality Government Segment

Segment revenue of $1.3B LTM up 13% YoY (54% of total), $335M in 2024 Q1 growing 16% YoY. Publicly disclosed federal spending data demonstrates inherent lumpiness in elephant-hunting large government contracts, and while the last two quarters of the calendar year appear lighter on new federal obligations to “Palantir” entities, the recent announcement of $178M 2Y TCV TITAN program with the US Army demonstrates continued momentum within the government segment. (4) Marketed as “Gotham”, the core innovation behind Palantir is software that integrates data from disparate sources, transforms that data into a consolidated data corpus, powers analytic queries and reporting to inform operational decision-making, and enables customers to deploy applications on top of the data directly in the platform. Palantir pairs their standardized technology with teams of “forward deployed engineers”, embedded inside their customers to get into the weeds on the entire process, from architecting the right data pipelines across an organization all the way to designing applications collaboratively with customers. Government customers often lack the internal technical talent to adopt modern software, and Foundry was architectured as a general-purpose data platform that can then be tailored to use cases across government settings.  Back to the consulting versus software debate, the reality falls somewhere in the middle. The key to Palantir’s success is that this software + services hybrid is the only way to make government customers successful compared to software-only approaches that leave the government empty-handed on deriving value from tools or consulting-only projects that build frail custom software from scratch for each engagement. As well-established by the enduring market capture of “the Big Five” defence primes, building momentum with government customers tends to unlock compounding competitive advantages in the go-to-market motion to support durable growth: not only are the government contracts up for grabs sticky, multi-year engagements, but figuring out the contracting process to even get into the RFP is extremely challenging for emerging competitors to crack, to the degree that by the time a government project reaches the multi-stage evaluation process, Palantir has already been personally engaged with the customer architecting technology requirements that only Palantir can offer. Palantir benefits from a talent arbitrage, bringing cream-of-the-crop engineering talent to government customers in a way that legacy defence vendors simply cannot beat in competition for talent. As long as Palantir continues to maintain the fastest shipping velocity as the first to bring emerging computational capabilities to government customers, it’s unlikely to see the multi-decade track record of sluggish legacy defence software vendors changing anytime soon. Once embedded in a customer, Palantir not only creates extreme stickiness by mapping a customer data corpus to the Palantir “ontology”, but also gets a wedge into customers from which to expand into new use cases across the same organization, with incremental use case expansion becoming much easier once already approved by the government entity and deployed within their system infrastructure. Plenty of runway for growth in multi-$10B market opportunity in government alone (which Palantir sized at $63B back in 2020 including the US and allies).

Mediocre-Quality Commercial Segment

In 2010, Palantir expanded its go-to-market motion to sell to commercial customers as a new source of growth, eventually launching “Foundry” in 2016 as a new offering tailored to commercial use cases. The commercial segment now amounts to segment revenue of $1.1B LTM up 23% YoY (46% of total), $299M in 2024 Q1 growing 27%. Similar hybrid software + services approach as in government, with customer success stories ranging from monitoring PG&E’s power network for wildfire risks all the way to helping Ferrari analyze F1 vehicle performance. While commercial deployments still benefit from some level of stickiness of applications deployed on top of the “ontology” data corpus organized inside the platform, the segment does not benefit from the same degree of customer lock-in nor go-to-market upper hand as in the government market. Further, Palantir faces far more competitive intensity in the enterprise, as the talent arbitrage they can monetize in government customers evaporates in the enterprise market going up against the entire field of software vendors fighting to win data infrastructure + operational analytics + application development budgets. Again, plenty of room for future growth in another multi-$10B market segment (which Palantir optimistically sized as $59B at IPO back in 2020). However, that’s been the story for the last decade… what seems different today is that this new cycle in “AI” unlocked by the transformer architecture resets the board on competitive positioning, leaving Palantir in a substantially weaker place relative to expectations baked into the stock. 

Tenuous Competitive Position in Enterprise “AI”

Palantir was relatively early to take advantage of renewed enterprise interest in “AI”, launching AIP (“Artificial Intelligence Platform”) in June 2023 as an environment for deploying foundation models on private enterprise data corpuses. Launching this model serving layer on top of Foundry reaccelerated the commercial business, inflecting growth from a low of 10% in 2023 Q2 to 32% in 2023 Q4, with management commenting that the US commercial segment inflected to 70% YoY from 20% in June 2023 on the back of “more than 560 boot camps across 465 organizations to date” (collaborative pilots in which Palantir builds applied AI use cases directly with customers over a few days). However, this encouraging momentum in boot camps as a leading indicator of durable acceleration in the commercial business did not continue in 2024 Q1… commercial revenue accelerated from 10% YoY growth in 2023 Q2 to 32% growth in 2023 Q4 but decelerated down to 27% in 2024 Q1. Growth in the US commercial segment in particular that management anchors as the clearest signal of AIP adoption collapsed from 70% YoY last quarter to 40% YoY this quarter. This can be sped up by having your website indexed instantly by Google. Speed to market might have earned them an initial burst of customer activity, but that momentum appears to be receding as competitors ultimately brought their offerings to market. Throughout the first few years of a new technology cycle, emerging technology often changes fast enough for customers to vote with their budgets on best-in-class new technology, during which the platform bundle value proposition of Palantir (ability to pitch customers on a one-stop-shop of relatively mature technology) does not resonate as strongly when not attached to the distribution advantage of cloud platforms at scale. As depicted in the 2024 Q1 MS CIO Survey below, enterprise buyers report a dramatic contraction in their plans to rely on an “AI Application Development vendor (ie, C3.ai, Palantir, etc…)” to operationalize “AI”, falling from 18% in 2023 Q1 down to 4% in 2024 Q1. The cloud platforms reportedly lead the pack at 41%, with Microsoft Azure ahead as their relationship with OpenAI has contributed to unlocking roughly $4B in run-rate revenue from “AI services” over the last year. Then at 15% “Data Management vendor (ie, Databricks, Snowflake, etc…)” maintains an advantage on the frontier of engineering new models with market-leading results at lower cost (eg, Databricks open-sourcing DBRX via MosaicML, Snowflake shipping a foundation model Arctic that only cost $2M to train, etc). Even “Application vendors (ie, Salesforce, ServiceNow, etc…)” are ahead of Palantir at 11% as they maintain a competitive advantage in being closer to domain-specific customer use cases + existing data corpus in functional systems of record. This survey data maps to a sample of customer call anecdotes that suggest Palantir’s competitive position in the commercial business is getting worse, not better. Contextualizing this against commercial segment prospects, emerging applications of the transformer architecture will ultimately older techniques in machine learning, natural language processing, and computer vision; not only is competitive weakness on the frontier of AI a risk to new growth but also to revenue retention as legacy applications from the “Big Data” era are ultimately displaced by new methodologies.

Market Multiples

At consensus multiples of 15.6x NTM Revenue (0.74x G-Adj.), 19.2x NTM Gross Profit (0.91x G-Adj.), 54.5x NTM FCF (2.59x G-Adj.), PLTR screens as one of the most expensive high-growth software stocks around. The essence of this short thesis is that at these levels, the stock no longer needs to play out against the full bearish story towards earning the lower valuation multiples of consulting services… merely valuing the government segment and commercial segment separately given such variation in business quality is enough. Relative to public sector comps Tyler Tech (9.6x NTM revenue, 21.6x NTM GP) and Axon Enterprises (11.2x NTM revenue, 18.6x NTM GP), applying an extremely generous 18x NTM GP multiple to the government segment contributes $21.5B EV. The commercial business certainly may benefit from some thematic momentum in “AI, but ultimately resembles a business model on the lower end of the quality spectrum, thus deserving a lower bucket NTM GP multiple of 10x (relative to 11.2x at C3.ai), contributing $10.9B EV. This sum-of-the-parts against current market multiples ultimately arrives at $32.3B EV and $15.09 PPS, a 30% downside from current trading levels. Again, current multiples in software are by no means cheap, so further downside tilt via sector beta.

Intrinsic Value

However, since the current market for high-growth software still feels far from fairly priced, extrapolating PLTR’s growth prospects against what’s baked in the stock paints an even more concerning picture. Consider the illustrative scenario below, in which even generous assumptions around PLTR ultimately reaching 35% mature FCF margins and earning an aggregate 35x NTM FCF multiple (blend of generous premium multiple on government, weaker multiple on commercial) requires an 18% CAGR to grow into the current valuation. In other words, not only must they maintain top-decile software FCF margins in the most competitive enterprise space over the next decade, but the contracting growth trajectory must inflect upwards and stabilize at a floor of last year’s growth rate… in that optimistic scenario, the intrinsic value of PLTR is roughly flat three years out. Without digging into the debate on where market treatment of SBC will ultimately land as a true economic cost of running the business, the conservative extreme detracting the full burden of SBC from FCF suggests a 46% downside in the same optimistic scenario (and 23%% downside if the theoretical cash replacement cost ultimately amounts to 50% of SBC). Without digging too deep into case building here, this calibration corroborates the multiple framing above that PLTR is substantially overpriced, with more likely execution scenarios yielding even higher downside.

Upside Risks

PLTR has developed a fervent retail investor fanbase, and while it’s encouraging to see the stock react to the fundamentals of this earnings cycle, there is still room for retail interest + ARK-style thematic inventors to support the stock at uneconomic levels for years in the medium-run. On AIP, quarterly customer adoption has certainly accelerated, even if revenue growth has decelerated; in the (unlikely) scenario that Palantir can unlock the same level of customer lock-in with enterprises as in government, the segment could outperform or beyond the flat stock scenario.