This week marks an important step for both 22V and me. We are launching a new research wall on the 22V Research website, designed to give a broader audience access to the institutional-grade research I publish each week.
I am intentionally starting with this Palantir research to help you better understand the “SaaS-pocalypse” panic currently sweeping the markets. As AI shifts from a support tool to a primary driver of the agentic world, the traditional seat-based SaaS model is facing a structural reckoning. In my YouTube video this week, I discuss why companies like Palantir, along with Bitcoin, are being swept up in that fear despite representing the very “orchestration layer” that will emerge as the winner in this transition.
To give you a clear view of the type of work that will live behind the new research wall, I’m publishing this report on Substack today.
How to stay connected:
Full Research Archive: To access my ongoing body of institutional research, visit 22vresearch.com. The site is scheduled to go live later today or tomorrow; if it’s not yet accessible, please use the contact info there and the team will follow up directly.
Weekly Deep Dives: Subscribe to my YouTube channel, Jordi Visser Labs, to see the visual frameworks and data behind these reports.
Executive Summary
In a moment when investors are questioning whether the seat-based SaaS model can endure an agentic AI world, and when many software companies are scrambling to adapt, Palantir is exhibiting a very different set of signals. In the same week that SaaS stocks saw broad compression on fears of structural impairment, Palantir reported 137% U.S. commercial growth, 57% adjusted operating margins, and a Rule of 40 score of 127%. This paper argues that Palantir’s results matter less as a single-company story and more as a window into how large enterprises are choosing to integrate AI in practice.
The core point is simple: most enterprises are not starting AI adoption by buying “more tools.” They are starting by confronting years of accumulated software bloat, fragmented data, and duplicated workflows that make AI unusable at scale. In that environment, the prize is not the next application, it is the orchestration and integration layer that can connect models to operational reality with governance, security, and auditability. Palantir’s growth suggests that enterprises are increasingly directing budget toward this systems layer, because it is the fastest path from experimentation to production.
Traditional SaaS was built on a reductionist premise: break an organization into functional silos and sell a specialized application for each. Agentic AI inverts that premise. Foundation models are inherently polymathic, they reason across domains and generate their greatest value by connecting context across systems. That makes the siloed application stack not just inefficient, but increasingly mismatched to how AI delivers outcomes.
Palantir’s Ontology, integration layer, and governance architecture should be understood in this light. They are not “features.” They encode a systems-thinking approach to enterprise AI, designed to map the enterprise as a connected organism rather than a collection of departments. The deployment velocity, margin structure, and customer expansion dynamics imply that this architecture is resonating because it solves the real enterprise bottleneck: operationalizing AI inside complex, regulated, legacy environments.
Whether or not you own the stock, Palantir’s trajectory is a high-signal indicator of where enterprise AI spending is concentrating and which layers of the stack are most likely to capture value as the AI transition accelerates.
A Personal Note on Why This Moment Matters
In 1999, I was in a leadership role at Morgan Stanley, returning from Brazil to take over a new area. When I received my initial budget, one line item immediately stood out: technology. It was far larger than I expected for a business after coming from Brazil where our tech budget was very small.
I asked to meet with the CTO to understand what was driving the cost. What followed was not a single explanation, but a tour through a web of people, teams, and software. Some systems were built in-house, others sourced from external vendors. Many overlapped in functionality. Each division had its own stack. Each company Morgan Stanley had merged with or taken over had its own. Each stack had its own owners and programmers. Similar problems were being solved repeatedly across the organization, multiplied by the number of desks, geographies, and reporting lines.
What struck me most was not any single piece of software, it was the multiplication effect. Complexity compounded as the firm scaled. Data lived in silos. Business logic was embedded in workflows and tribal knowledge rather than systems. Technology spend grew faster than insight, faster than productivity, and faster than decision quality. At the time, no one called this inefficient. It was simply considered the unavoidable cost of operating a large, sophisticated enterprise.
That experience stayed with me. It shaped how I think about enterprise software, not as a collection of tools, but as an organizational mirror. When systems sprawl, budgets sprawl. When context fragments, decision-making slows. And when software proliferates without integration, value leaks. Over time, scale itself became the source of the bloat, not malice, not inefficiency, but the quiet tax of complexity.
In that sense, AI’s impact on SaaS may prove analogous to Ozempic’s impact on processed foods: a force that does not attack the industry directly, but quietly suppresses the appetite for excess that accumulated over time. Ozempic did not make junk food taste worse. It reduced the craving that made overconsumption feel normal. Similarly, agentic AI is not rendering individual SaaS tools useless; it is eliminating the organizational appetite for buying more of them. When an orchestration layer can coordinate across existing systems, the reflex to solve every new problem by purchasing another application starts to fade. The bloat does not get attacked. It simply stops getting fed.
The Week SaaS Panicked
Which brings us to today, the week SaaS panicked and fears of contagion spread across asset classes. In the same week that Palantir Technologies reported 137% U.S. commercial growth and a Rule of 40 score of 127%, a wave of anxiety swept through enterprise software. Stock prices compressed, terminal value assumptions collapsed, and investors began asking an uncomfortable question: what if the SaaS business model itself is structurally impaired by agentic AI?
This did not feel like a sudden break so much as capitulation in a trend that had been building for nearly a year, ever since “vibe coding” entered the vocabulary and the market began to deal with how quickly AI was compressing software value chains. Private credit concerns resurfaced from last year, Bitcoin sold off sharply, and headlines rushed to label the moment a “SaaS-apocalypse.”
But this was not a typical risk-off move or a simple multiple reset. Even as someone who has been bearish on SaaS for a long time, this felt a little like a week of emotional capitulating SaaS vomit. Markets were not reacting to deteriorating fundamentals as much as to growing worries with every Claude release that the economic assumptions underlying seat-based software, long procurement cycles, and fragmented application stacks may no longer align with how enterprises operate in an increasingly AI-native world.
Against that backdrop, Palantir’s earnings quietly stood out to me as a sharp counterpoint with important signal amidst the noise. I had not spent much time focused on Palantir historically, and like many, I had tended to think of it primarily as a defense and government-oriented software company rather than a core enterprise platform. But in a week defined by SaaS anxiety, its results, and the broader discussion around software economics, including comments from Brad Gerstner on the All-In Podcast, prompted me to work with my AI thinking partners and take a deeper look at what was actually driving its growth.
That deeper look brings us back to Palantir, not as a celebration of one company, but as a signal of how enterprises are thinking about their own complexity as they move to adopt AI. For many organizations, AI adoption is not starting with new tools; it is starting with confronting years of accumulated software bloat, fragmented data, and duplicated workflows. Palantir’s ability to grow enterprise revenue at this scale suggests that companies are prioritizing platforms that help them see, rationalize, and integrate what they already have before layering on new intelligence. In that sense, Palantir’s success offers a window into how enterprises are attempting to resolve past sprawl in order to make AI operational in the present.
In that context, the more important discussion is not whether a particular SaaS category is dead or whether individual vendors can defend their margins. Those debates focus on symptoms, not causes. What matters is that Palantir is offering a fundamentally different solution from a systems perspective, one designed to reconcile complexity rather than add to it, and enterprises are responding accordingly. The scale and speed of its adoption suggest that this approach is not theoretical; it is already working in practice as organizations attempt to make AI usable within the constraints of real-world operations.
From Defense to Offense: Why Palantir Is the Counter-Case
Most of the market’s reaction to this moment has been defensive, asking which SaaS companies can survive, which margins can hold, which moats are deep enough. But that framing accepts the existing architecture as given and simply asks who weathers the storm. The more interesting question is whether any company has been building for a fundamentally different architecture all along, one where the value sits not in the application but in the connective tissue between applications, data, and decisions. That question leads directly to Palantir, and its Q4 2025 earnings suggest the answer is already showing up in the numbers.
Palantir’s Q4 2025 earnings (released February 3, 2026) were not just strong. They validated a thesis the market is still struggling to fully internalize. The company reported $1.41 billion in quarterly revenue, representing 70% year-over-year growth. More importantly, U.S. commercial revenue grew 137% year-over-year in Q4, with full-year FY2025 U.S. commercial growth of 109%.
These are not small numbers for a company approaching $5 billion in annual revenue. They point to a structural shift in how enterprises are buying, deploying, and valuing AI.
CEO Alex Karp was characteristically direct on the earnings call:
“We’re seeing demand that we haven’t seen in the history of the company. The bootcamps are working beyond our expectations. Customers are going from zero to production in weeks, not months or years.”
CFO Dave Glazer underscored why this growth matters financially:
“We delivered 57% adjusted operating margins in Q4, with nine consecutive quarters of GAAP profitability. This is not a trade-off between growth and profitability. We’re achieving both simultaneously because our go-to-market model is fundamentally different.”
From Q3 2025, CTO Shyam Sankar captured the deeper inflection:
“AI is now the central organizing principle for how businesses make decisions. What we’re seeing is enterprises realizing that the model itself is not the constraint, it’s connecting that model to their actual operational reality. That’s what we’ve been building for 20 years.”
Client deployments reinforced this point. AIG disclosed that Palantir’s AIP platform reduced claims processing time by 40% while improving accuracy, noting that no other vendor could connect AI to legacy systems with the required governance and security. Walgreens announced a 4,000-store deployment of Palantir’s operational AI platform. Fannie Mae deployed AIP for mortgage fraud detection, processing billions of dollars in mortgage data with AI-powered analysis that previously required large analyst teams.
What these examples share is not model sophistication. It is context, control, and execution.
The Three-Part Architecture That SaaS Companies Cannot Replicate
Understanding Palantir requires understanding three interconnected layers that together create what the company calls operational AI.
1. The Ontology: The Semantic Foundation
At the core of Palantir’s platform is the Ontology, a proprietary semantic layer that maps raw enterprise data into a digital twin of business operations. This is not a database. It is a structured representation of how objects in the real world (customers, aircraft, purchase orders, patients) relate to each other and how they flow through business processes.
The key insight: large language models are brilliant at reasoning, but they are blind without context. A model looking at a row in a database does not know if that row represents a person, a transaction, or a sensor reading. The Ontology provides that context, turning data into knowledge that AI can act on.
This creates compounding lock-in. Once an enterprise builds its business logic into the Ontology, switching costs become enormous. As one analyst noted, leaving Palantir means losing the entire digital map of how the business functions. Palantir’s 139% net dollar retention rate, meaning existing customers spend 39% more each year, confirms this dynamic.
2. The AIP Bootcamp Model: Rewriting Enterprise Software GTM
Traditional SaaS companies sell software seats. They hire expensive sales teams, navigate 6 to 12 month procurement cycles, and then hand customers documentation. Palantir does the opposite.
The AIP Bootcamp model deploys elite engineers directly to customer sites for intensive one-to-five day sessions. These engineers do not sell. They build. They take a real business problem, connect Palantir’s platform to the customer’s actual data, and demonstrate operational AI in production within days. Prospects convert from skepticism to production deployment in under 25 days, a cycle time that is 10 to 20 times faster than traditional enterprise software.
The financial implications are profound. Palantir’s sales and marketing expenses collapsed from over 60% of revenue five years ago to approximately 23% in late 2025. This is the inverse trajectory of SaaS companies that need increasingly expensive go-to-market motions to maintain growth. Lower customer acquisition costs plus faster time-to-value equals a flywheel that traditional SaaS cannot replicate without fundamentally redesigning their organizations.
3. The Integration Layer: Where the Money Is
Here is what the market initially missed: Palantir is not competing with ChatGPT or Claude or Gemini. It is model-agnostic. The company integrates third-party LLMs and connects them to enterprise data systems, including ERP, CRM, supply chain management, industrial IoT, and legacy databases, with the governance, security, and auditability that regulated industries demand.
This positioning is strategic genius. As foundation models commoditize (which they are, rapidly), Palantir captures increasing value. The company sits between cheap, powerful AI models and messy, valuable enterprise data. That is the bottleneck. That is where enterprises are spending.
Palantir’s gross margins of 80%+ once implementation stabilizes confirm this. This is software economics applied to what was traditionally a services business. The Ontology and integration layer allow Palantir to deliver consulting-grade insights and automation at software margins, a combination that traditional consultancies (who have 40 to 60% margins) and traditional software companies (who lack the integration depth) cannot match.
The Ultron Comparison: Consumer AI vs. Enterprise Reality
On this weekend’s All-In Podcast, Jason Calacanis offered a concrete, real-world illustration of where software value is migrating by walking through how he built an internal system he calls “Ultron.”
Ultron is not a new SaaS application. It is an agentic orchestration layer built on top of existing tools. Calacanis described constructing it using OpenClaw, an open-source agent framework, and wiring it directly into his firm’s operating systems through native APIs, including Slack, Gmail, Notion, and calendar services. Rather than living inside any one application, Ultron sits above them, ingesting messages, documents, emails, meeting context, and task histories into a unified operational memory.
Functionally, Ultron behaves like a “canonical employee.” It lives inside Slack, understands what is happening across the organization in real time, and can answer questions or coordinate actions that previously required stitching together context across half a dozen tools: what meetings occurred yesterday, what decisions were made, what follow-ups are outstanding, and who owns them. The individual SaaS products remain in place, but they are demoted to execution endpoints and data sources, not decision centers.
The economic implication Calacanis highlighted is critical: the profit pool is no longer anchored to the application layer. Slack, Gmail, Notion, and calendars still matter, but the user’s primary interaction shifts to the orchestration layer that spans them. Value accrues to the system that understands context across tools, not within any single one. Applications risk becoming commoditized infrastructure unless they control, or successfully monetize, the orchestration layer above them.
This is the precise transition Palantir is enabling at enterprise scale.
Palantir is effectively Ultron for Fortune 500 organizations, but with two structural advantages that make the comparison instructive rather than superficial.
First, enterprise data complexity is orders of magnitude greater. Ultron works because consumer SaaS tools expose clean, well-documented APIs. Enterprises operate across legacy ERP systems, on-prem databases, industrial control systems, regulatory data stores, and decades of bespoke schema evolution. Palantir’s Foundry platform was purpose-built over twenty years to normalize, integrate, and govern these heterogeneous systems at the data layer itself, without relying on fragile application-level APIs. This integration depth is the hardest part to replicate and the core of Palantir’s defensibility.
Second, enterprise AI carries real liability. If Ultron makes a mistake, the cost is embarrassment or inefficiency. In regulated industries, including healthcare, finance, defense, and critical infrastructure, mistakes carry existential risk. Palantir’s origins in intelligence and defense forced it to embed security, auditability, permissioning, and human-in-the-loop controls from day one. These are not features that can be bolted on later; they are architectural requirements that consumer-grade agent frameworks and open-source tools are structurally unequipped to provide.
Calacanis made an observation that further validates Palantir’s positioning: “If SaaS companies close their APIs, I’m leaving immediately. I need the data to flow.” Consumer agents are fragile because they depend on continued API openness. Palantir sidesteps this risk entirely by integrating directly at the data layer, connecting to raw databases and operational systems rather than application-controlled interfaces. This makes it resilient to the very API-gating risk that threatens agent-based platforms built purely on top of SaaS.
Seen through this lens, Ultron is not a curiosity. It is a micro-proof of the same structural shift Palantir is exploiting at scale. As AI becomes the primary interface for work, applications recede into the background, and the system that reconciles data, context, and execution becomes the operating system. Palantir is not fighting the SaaS collapse narrative; it is monetizing the layer that emerges after it.
Why This Is Different From SaaS
Traditional SaaS companies face an existential dilemma in the agentic era: open their APIs (and lose value to the orchestration layer) or close them (and lose customers to competitors who remain open). This is why SaaS multiples compressed in many examples from 30x to 15x free cash flow. Terminal value certainty collapsed.
Palantir does not face this dilemma because it IS the orchestration layer. The more AI capabilities improve, the more valuable Palantir becomes, not less. Every improvement in foundation models, every new AI capability, makes Palantir’s integration and governance layer more valuable because it is the mechanism that makes those capabilities usable in complex, regulated, mission-critical environments.
The financial profile confirms this categorical difference:
Growth acceleration, not deceleration. U.S. commercial guidance for FY2026 is 115%+ growth, accelerating from an already high base.
Margin expansion, not compression. Adjusted operating margins of 57% and expanding, while SaaS peers face margin pressure.
Customer expansion, not churn. 139% net dollar retention means customers spend 39% more each year, the opposite of the SaaS seat-based model where growth requires new customer acquisition.
Value-based pricing, not per-seat licensing. Palantir prices on business outcomes delivered, positioning it to capture expanding value as AI capabilities improve.
David Friedberg articulated this on the All-In Podcast: “Software is going to transition from selling seats to value-based pricing, and the TAM expands 10 to 100 times because you’re no longer selling software, you’re absorbing the services economy.” Palantir is already there. The company competes not just against other software vendors but against management consultants, business analysts, and decision-making services. If successful, Palantir’s addressable market is not the $200 billion enterprise software market. It is potentially the $2 trillion enterprise services market.
The Investment Debate That Misses the Point
The consensus debate regarding Palantir centers on valuation. Bulls point to exponential growth. Bears point to a forward P/E above 150x. Both are correct and both are largely irrelevant to the structural question.
For the debate around SaaS, the real question is whether enterprise AI is a new computing platform or just a feature upgrade to existing software. If it is a feature upgrade, Palantir is egregiously overvalued. If it is a new platform comparable to the shift from mainframes to client-server or from on-premise to cloud, then Palantir is not expensive; it is early. And much of SaaS is in trouble.
I believe it is a platform shift, and the reason comes back to a fundamental mismatch between how AI thinks and how enterprise software was built. The SaaS model is reductionist by design. It decomposes complex organizations into discrete functional categories, CRM, ERP, HRIS, supply chain, communications, and sells a specialized tool for each. That architecture made sense when software was expensive to build and humans served as the integration layer, synthesizing context across systems through judgment and experience. The human was the systems thinker. The software was the specialist.
AI inverts that. A foundation model is inherently polymathic. It reasons across domains, connects disparate data, and holds context that spans functional boundaries. Its power is connective, not specialized. That capability makes the siloed application stack not just inefficient but architecturally mismatched to how AI generates its deepest value. You cannot unlock polymathic intelligence by feeding it one silo at a time.
Palantir understood this before the market did. The Ontology, the integration layer, the governance architecture: these are not product features. They are a systems-thinking approach to enterprise AI encoded in software, built on the premise that an organization is a system, not a collection of departments. That is why the deployment data looks the way it does. That is why customers expand rather than churn. And that is why traditional SaaS companies cannot replicate it by adding an AI tab to their existing applications.
The risk to this thesis is that I am wrong about the category, that enterprise AI proves to be a feature upgrade absorbed by incumbent vendors rather than a new architectural layer. If that is the case, Palantir’s valuation continues its recent compression and the SaaS model proves more durable than the current panic suggests. That outcome is possible, but the deployment velocity, the margin structure, and the customer expansion data increasingly argue against it.
The actionable takeaway is not whether to own Palantir at this multiple. It is that Palantir’s trajectory reveals where enterprise AI spending is concentrating, which layers of the stack are capturing value, and which competitors face structural margin compression. Whether or not you own the stock, if you are investing in enterprise software or analyzing the AI landscape, Palantir is the signal for decoding what is actually happening beneath the hype, and right now, what is happening favors the systems thinker over the specialist by a widening margin.