PLTR represents customized AI systems
Anthropic and ChatGPT represent general-purpose AI
An analysis of the competitive and collaborative outlook for these two types of systems
My assessment is that the market ceiling for general-purpose AI is higher; AI platforms like Palantir’s, which are deeply embedded in business operations, are more likely to build highly sticky businesses with sustainable revenue. But both are expanding into each other’s domains.
Therefore, the key question in the future is not just “general-purpose or customized,” but rather: Who controls a company’s data, workflows, and execution points? And who can turn AI capabilities into outcomes that customers are willing to keep paying for?
The following analysis is based on product information verifiable as of November 10, 2026; the outlook is my own inference.
A little clarification first
The classification above captured the main point, but Palantir is not simply a contracting company that builds a new AI system from scratch for every enterprise.
It is closer to “a standardized platform, deeply deployed to fit each enterprise’s individual workflows.” AIP can connect to GPT, Claude, Gemini, and open models; its core function is to integrate enterprise data, business rules, operational permissions, and real-world actions. That is also why it emphasizes Ontology.
On the other side, OpenAI/ChatGPT and Anthropic/Claude are useful comparisons: they operate models, user products, and enterprise developer tools, and are no longer just providers of chat interfaces.
I. Palantir’s outlook: the more enterprises want AI to “actually get things done,” the greater its opportunity
Take an ordinary company or processing plant, for example:
Having AI organize raw material quotes and analyze formulation documents could create value with general-purpose tools.
But if AI is to adjust production schedules based on orders, raw material inventory, batch quality, equipment constraints, and delivery dates, issue purchase orders, and then route them to managers for approval according to permissions, many systems and rules have to be connected.
Palantir’s value lies mainly in the second kind of problem. This is a conceptual example, not a recommendation that your factory needs to buy Palantir. Besides, a small player like me couldn’t afford it anyway!
These platforms have three long-term advantages:
* Results can be readily linked to economic benefits. If a platform can reduce downtime or inventory, its fees can be compared with the costs saved by the customer.
* Once deeply embedded in workflows, replacement costs are higher. What needs to be changed is not just the model, but also data mappings, permission settings, workflows, and employee habits.
* Model improvements can be an advantage. Cheaper, better models can make an existing platform more useful rather than necessarily eliminating the platform itself.
But it is important to distinguish between customization and business success: more customization does not mean a better business. In other words, getting workflow proposals out the door quickly and actually winning contracts and making money are two different things!
If every project requires a large number of engineers to redevelop the solution, rising revenue will also drive up labor costs. The most valuable model is one where the underlying capabilities can be reused while customers still get results tailored to their needs.
Palantir’s latest reported revenue for the third quarter of 2026 was $2.313 billion, up 96% year over year; U.S. commercial revenue was $984 million, up 148% year over year. This supports the view that enterprise demand is strong, but total company revenue is not the same as pure AIP revenue, and these growth rates should not be assumed to persist over the long term.
II. The outlook for general-purpose AI: a broader market, but distinguish “usage” from “profitability”
The potential markets for OpenAI and Anthropic include individuals, enterprise employees, software developers, and a wide range of applications.
Their key advantage is that an improvement in model capabilities can improve many products and workflows at once. Compared with deep, company-by-company deployments, product distribution is usually faster and can more easily reach small and midsize businesses.
In my view, these companies have two main opportunities:
First, to become the primary gateway for everyday work.
If users start here to search, code, analyze data, and carry out tasks, the platform has an opportunity to own customer relationships and generate revenue from follow-on services.
Second, to become the intelligence provider for other software.
Even when users operate products from other brands, those products may still use its models behind the scenes, generating API revenue.
However, both types of revenue face pressure:
* When model capabilities converge, customers can more easily compare prices and switch providers.
* The more complex the task, the more inference, tool calls, and retries it may require. But so far, fees have not necessarily increased in step.
* The ongoing race for leading-edge models requires substantial investment in research and computing.
Or, to put it simply: these AI giants still have to pour money in—and they may not necessarily make money back!
Therefore, rapid growth in AI task volume alone is not enough to prove that model companies’ profits will grow quickly. The real question is whether the revenue from each valuable task completed can cover computing, service, and research costs.
III. The biggest shift: the two sides are starting to converge head-on
“Palantir handles enterprise integration, while model companies just provide the brains” no longer adequately describes today’s market.
OpenAI launched Frontier in 2026, explicitly covering enterprise use cases, agent execution, permissions, and evaluation. Anthropic’s Claude Managed Agents also offers a managed execution environment, tools, and support for long-running tasks; at the time of review, it was still labeled Beta.
This means general-purpose AI companies are lowering the barriers for enterprises to build their own workflows.
But competition and collaboration can coexist. For example, Palantir and Anthropic announced in 2024 that they would integrate Claude into AIP for U.S. government-related deployments. This at least demonstrates that the two can complement each other architecturally.
My assessment of the different markets is as follows:
Use case Party more likely to benefit Reason
Personal and general office work General-purpose AI platforms Easy to use, low deployment costs
Software development, documents, and research General-purpose AI + specialized applications Many of these tasks are already digital
Standard workflows at small and midsize businesses General-purpose AI + existing enterprise software They may not be able to afford deep integration
Large-scale manufacturing, logistics, and complex supply chains Palantir-style platforms have an advantage Multiple systems, physical constraints, and complex permissions
Highly sensitive, regulated operations Platforms with existing deployment and governance capabilities Traceability and reliable execution matter more than flashy demonstrations
This is not a permanent division of labor. If model companies become good enough at enterprise integration, some of Palantir’s market will come under pressure.
IV. The risk most worth watching for PLTR is “good-enough substitutes”
Palantir does not need to be comprehensively surpassed on the technology front to face competition.
All it takes is for customers to realize that:
Existing enterprise software, combined with general-purpose AI, can meet most needs at a lower total cost.
Palantir’s ability to win new customers and command pricing power could then be affected.
So what it must prove over the long term is not “we can do AI too,” but rather:
The complexity we handle, our reliability, and the real-world benefits we deliver are enough to justify a higher total deployment cost.
Conversely, the more capable general-purpose models become, the more enterprises may need someone to manage data, permissions, and actions centrally. Palantir could therefore expand its market as models improve. Better model capabilities can both help and hurt PLTR; we should not look at just one side.
V. How I would assess it from an investment perspective
Looking at industry developments over the next three to five years:
* Market ceiling: I lean toward general-purpose AI platforms having the larger opportunity.
* Stickiness within specific enterprises: I lean toward platforms deeply embedded in operational workflows being stronger.
* **Visibility into business results:** PLTR publishes financial reports, making it easier to keep validating its performance.
* Where the profits ultimately accrue will still depend on who controls the customer gateway and execution workflows, as well as cost and pricing power.
For PLTR, I would keep an eye on whether existing customers expand adoption, whether deployment becomes more efficient, and whether growth translates into cash flow. For model companies, I would watch paid retention, the cost per task, and whether they can evolve from providing models to owning the gateway to work.
My core view is that demand for customized workflows will persist over the long term, but customized platform companies will not necessarily capture all of it. The more mature general-purpose AI becomes, the more it can handle some customization on its own. Palantir must defend its hard-to-replace capabilities in operational integration and execution.
For PLTR stock, there is another question: even if the business outlook described above holds, has the current share price already priced in too much growth? This analysis supports the industry opportunity, but does not determine whether the current share price is cheap or expensive.