Palantir’s latest earnings were dazzling. U.S. commercial revenue surged 93% year over year. By the “Rule of 40”, that simple measure of company health where revenue growth plus profit margin should add up to more than 40, they scored 94. Nearly perfect.
I’ve been following Palantir since 2021. Back then, I was a believer in its long-term potential. Today, I still like the company, but only for the next two or three years. Beyond that, I’m not so sure.
Palantir doesn’t sell a single piece of software. It sells a system. Three layers: Gotham for governments, Foundry for businesses, Apollo for continuous delivery. Together, they function as an enterprise-wide operating system.
The logic is straightforward: no matter how messy your data, no matter how tangled your business processes, plug them into Palantir and the system will integrate it all from the top down. One centralized solution.
But this ambition came with a price. Palantir’s business model has always been heavy.
In the early stages, they’d send engineers directly to customer sites, working side by side to fix the ugliest problems. Often at low cost, sometimes even free. The idea was to prove value first, then sign customers to multi-year, high-dollar contracts. Once hooked, customers rarely left.
This “land–develop–lock-in” model brought in huge revenue from just a few hundred clients. But it also slowed Palantir’s ability to scale.
Then came the wave of large language models. Palantir responded quickly with its AIP (Artificial Intelligence Platform). It lets customers weave AI directly into their core data and workflows. Even non-technical employees could suddenly call on AI as if it were second nature.
That changed everything. On one side, Palantir leveraged the surge of AI-driven demand to solve its customer acquisition challenge. On the other, large AI models greatly boosted programming and development efficiency, further accelerating product deployment.
The results show up in the earnings.
But we shouldn’t confuse a sugar rush with a new metabolism.
Because something bigger is happening. The rise of AI Agents isn’t just an incremental improvement. It’s a shift in the software paradigm itself.
If Palantir is a walled garden, integrated, self-contained, then the Agent future looks more like Lego blocks. A modular ecosystem where businesses mix and match what they need. Supply-driven models give way to demand-driven ones.
We could divide software companies into three generations:
The first: traditional giants like SAP, Oracle, Adobe. They have scale, distribution, and integration, but struggle under the weight of technical debt. The second: cloud and data platforms born in the big data era. Salesforce, ServiceNow, Snowflake, Palantir. They’re modern, faster, and better positioned for AI, but also the first to feel its disruption. The third: AI-native companies. Built from scratch around generative AI, with new interaction paradigms and business models. Think OpenAI, Anthropic, Harvey, Perplexity.
Within that second generation, we already see two paths emerging:
Path 1: Snowflake digs downward, becoming fertile AI-ready soil. It doesn’t try to be a bigger walled garden. Instead, it prepares data so any Agent can grow on top of it. Safe, standardized, accessible.
Path 2: ServiceNow grows upward, aiming to be the Agent Control Tower. As companies fill with hundreds of Agents, someone needs to manage, coordinate, and direct them. That’s the niche ServiceNow wants: not the Lego blocks, but the general who commands them.
Both are still early moves. The real experiments are happening with third-generation companies, the AI natives, who are already breaking down old business boundaries.
Which brings me back to Palantir.
Its current boom in commercial revenue is real. Customers who never touched AI before suddenly see, “Wow, this is exactly what I need.” ARPU soars. Financials shine.
But that’s the “sinking into sand” problem.
Success reinforces the existing model. Palantir’s moat is built on foundations that may not last.
Because once the Agent paradigm matures, today’s business will look like yesterday’s internet BBS boards, valuable once, but quickly outpaced.
So yes, Palantir is winning today. The question is: when tomorrow arrives, will the walled garden still matter in a Lego world?