Why does the CEO of PLTR criticize the Frontier Big Model Laboratory for emphasizing corporate autonomy? Karp appears to be advocating for the autonomy of enterprises in the AI era, but in reality, it is promoting sovereign AI solutions. At a deeper level, it is a counterattack launched by the direct competitive pressure of large model manufacturers in defense and commercial customers. The business war in the AI era is so simple and straightforward. As someone who has achieved some results on pltr and has written many articles about pltr before https://(x.com)/qinbafrank/status/1887674164867440965? The investors of s=46&t=k6rimWs Ebo2D2TXolYcM-A have some say in pltr. Let's talk about personal opinions: 1. The issue of 'loss of autonomy by enterprises' does exist For clients such as defense, energy, finance, manufacturing, healthcare, and intelligence agencies, AI is not just a simple chat tool, but an "operating system" embedded in production, approval, command, compliance, supply chain, and combat processes. What they are really worried about is not a vague statement about whether data will be stolen, but several deeper things: where the model call chain is, where the logs are, whether prompt words and intermediate inference products are retained, whether the model version is controllable, whether it can run in a disconnected/isolated environment, whether it can be audited, whether it can switch models, and whether manufacturers will abstract their business processes into generic products and sell them to competitors. Karp captured a real contradiction, but he made the statement that 'cutting-edge laboratories steal enterprise data' too absolute. OpenAI and Auroc have both explicitly stated that the API and input/output data are owned by the client and are not used for training by default. 'Data must be taken for training' is not a rigorous accusation; The more rigorous version is that companies are concerned about external model vendors mastering key business contexts, forming dependencies, and moving up to the application layer in the future. 2. Why is Karp firing now? Because Palantir is promoting a 'sovereign AI' solution This is not a casual bombardment. On June 29th, Palantir announced a partnership with Nvidia to launch an engine for deploying Nvidia Nemotron open models in a "sovereign environment," with a focus on US government agencies and critical infrastructure; The announcement clearly states that the Palantir+Nvidia solution aims to enable organizations to retain control over data, IP, and AI systems, and support sensitive environments such as classified and air gapped. This plan states that customers can have a 'self-improvement model tailored to their own tasks'. In other words, Karp is not simply criticizing OpenAI/Anthropic, he is drawing a boundary for the market: 1) OpenAI/Anthropic=Model Capability+Token Rental+Closed API 2) Palantir/Nvidia=Model in Customer Environment+Compute+Ontology+Permissions+Audit+Business Loop This is where 'butt determines head'. The core business narrative of Palantir is that the model will be commodified, and the true value lies in the enterprise AI operating layer, which includes data, permissions, workflows, business objects, auditing, and on-site deployment. So Karp must lower the narrative of the token API and raise the narrative of Palantir's ontology/application layer. 3. Pltr has indeed felt the pressure of cutting-edge model manufacturers entering defense AI Defense industry AI is no longer exclusive to Palantir. Last year to this year, the US Department of Defense signed direct cooperation agreements with several major model manufacturers, and Frontier Labs is no longer just providing "pluggable models" to Palantir. They are becoming direct contractors for government AI budgets. Moreover, both OpenAI and Authentic have established FDE model companies to assist enterprises in directly embedding AI into the entire workflow. This has a dual impact on Palantir: Short term collaboration is possible as Palantir can integrate Claude, OpenAI, or other models into their AIP/Foundry/Ontology; There is a long-term threat as model vendors, cloud vendors, and system integrators are all moving towards Palantir's "application layer/workflow layer/government platform layer". Karp competes while collaborating; Palantir does not train cutting-edge models, but is increasingly competing with OpenAI and Anthropic to become a platform for enterprises and governments to deploy AI. This is not the first time Karp has fired shots. During Pltr's first quarter earnings conference call, Karp harshly criticized large model manufacturers 4. Based on the third point, at the business customer level, Karp is more afraid of "model vendors moving up and eating up the application layer" OpenAI and Anthropic were originally model APIs, but now they are becoming less and less like pure model vendors. They have Enterprise Edition, Government Edition, Code Assistant agent、 Files/connectors workflow、 Industry solutions. These are all sensitive points of Palantir. The valuation and growth story of Palantir is not 'I sell a database/BI tool', but 'I am the operating system for enterprise and government AI'. If the model manufacturer directly owns the user entrance agent workflow、 The value of Palantir's ontology/AIP/forward deployed engineering will be compressed by data connectors, enterprise knowledge bases, code execution, and automated tasks. So what Karp really wants to say is: The model layer should not capture value, value should stay in the deployment layer, business layer, and data layer, and Palantir is that layer. ” For Palantir, this is also a valuation narrative defense battle. Its high valuation comes from scarcity. Scarcity in the field of defense and military AI, in the AI operating systems of governments and enterprises. In general Karp is not simply outraged by the concept of "corporate autonomy", but also due to the pressure of competition to break through defenses. This is a competitive narrative war driven by real pain points. Karp is not talking nonsense, he is speaking for Palantir's position. The butt determines the brain, and of course, there is indeed a very real issue of enterprise AI power transfer sitting under the butt this time. This article is sponsored by @ bitget_zh, titled 'Bitget Buying US Stocks: Instant Entry, Smooth Trading'