“Don’t depend on AI models” \u2014 the head of Microsoft (MSFT), one of the world’s largest AI investors, has started shouting this message. A long-form essay posted by CEO Satya Nadella on X in June racked up over 66 million views, triggering a massive response. Two weeks later, data analytics giant Palantir Technologies (PLTR) entered the fray with its “AI Sovereignty” manifesto. A chorus of “don’t cede control to big AI labs” is now sweeping through the U.S. enterprise AI discourse. The spark that lit this fire: the very real fear among corporate leaders that their AI bills are spiraling out of control while the source of their competitive advantage gets absorbed by the very companies providing the models.<\/p>\n
Nadella’s June 14 essay on X (formerly Twitter), titled “Frontiers Without Ecosystems Are Unstable,” posed a fundamental question about corporate management in the AI era. He identified two forms of capital companies must cultivate: “human capital” and a term he coined, “token capital.” Token capital does not simply mean the ability to rent and use external models. It refers to a company’s own unique AI capability \u2014 one where its workflows and domain knowledge are deeply embedded into AI systems and refined against proprietary evaluation criteria.<\/p>\n
The core of Nadella’s argument is that building a “learning flywheel” that continuously compounds both forms of capital will become the defining intellectual property of future enterprises. “The real opportunity isn’t picking the best model,” he wrote. “It’s building a learning flywheel on top of models where human capital and token capital compound.” He then offered a litmus test for whether a company is maintaining control: if you swap out a general-purpose model for another, does the veteran-employee-like expertise you’ve accumulated get lost?<\/p>\n
The essay’s latter half turns into a warning. “What no one wants is a world where every company in every industry cedes value to a handful of models that consume everything in sight.” Nadella cautioned that if a structure spreads where a few frontier models absorb specialized knowledge from every industry and resell it as a commodity, the same phenomenon that hollowed out manufacturing economies during the first wave of globalization will repeat itself. He went further: “If all value accrues to a handful of models, political economy won’t permit it. Society will not grant a social license to an AI future that hollows out entire industries.”<\/p>\n
Roughly two weeks after this quietly forceful essay rattled the psyche of the executive class, Palantir joined the same battlefront with a completely different tone. The nine-point “AI Sovereignty” manifesto the company published on its official X account on June 30 reads like a sharply worded exhortation directed at enterprises and government agencies. Point one declares flatly: “Your AI sovereignty will determine your organization’s future,” warning that surrendering sovereignty means “ceding your organization’s future right to choose to someone else.” Point two frames data as “your treasure,” asserting that transferring it externally means “giving away your connection to existing winning plays and the means of production for new ones.”<\/p>\n
The most aggressive is point three. Palantir uses the term “token-maxxing” to dismiss the practice of simply having AI generate massive volumes of scripts as “intoxicating you with the illusion of progress.” It then takes direct aim at the usage-based pricing models of major AI labs like OpenAI and Anthropic: “There is a reason those who sell tokens steadfastly refuse value-based pricing.”<\/p>\n
Point four deploys the phrase: “He who controls the weights, controls the destiny.” “Weights” here refers to the parameters of an AI model \u2014 the crystallized knowledge an organization has trained into it. Allowing someone else to control those weights, the manifesto warns, means “the alpha of your business migrates to theirs.”<\/p>\n
This manifesto was a coordinated information campaign tied to CEO Alex Karp’s appearance on CNBC the following day, July 1. On the program, Karp railed that “corporate leaders are furious,” positioning himself squarely as the voice of executives fed up with exaggerated claims about AI lab capabilities and exorbitant token fees.<\/p>\n
To be sure, this chorus is not receiving universal praise. A considerable number of reactions on X were coolly dismissive, essentially asking: “Isn’t this just a sales pitch to use their own platform?” Pointing to Palantir’s track record of providing surveillance systems to government agencies, some delivered biting sarcasm: “A company that built its business peering into people’s data is now lecturing us not to hand over our data?”<\/p>\n
The question of “sound argument or sales pitch” applies equally to Nadella’s essay. The advice not to get locked into a specific model is, read another way, a sales message to “make models interchangeable components and place your sovereignty in the substrate where those components get swapped \u2014 namely, Microsoft’s Azure cloud platform and Microsoft Foundry AI development platform.” Moreover, in June, Microsoft announced MAI-Thinking-1, its own in-house reasoning model trained from scratch without relying on distillation from other companies’ models, steadily advancing its “de-OpenAI-dependence” strategy at the model layer itself. The calculation is transparent: declare models a commodity while quietly securing control over both the model and the infrastructure layers.<\/p>\n
Palantir is no different. The manifesto’s prescribed design for “compounding alpha while protecting sovereignty” is, in concrete terms, nothing other than placing AI models as interchangeable components on top of the company’s core product, “Ontology” \u2014 a system that centrally models an enterprise’s data, business logic, and execution authorities. Peel back one layer, and the nine-point exhortation is effectively a specification sheet for the Palantir platform.<\/p>\n
However, the messenger’s motives and the validity of the message are separate questions. The real backdrop to why this chorus has generated such enormous resonance is the urgent reality of ballooning corporate AI bills as AI agents become deeply embedded in business operations. Every time an AI agent cycles through planning, information retrieval, deliverable creation, and self-verification, it consumes tokens at orders of magnitude beyond the era when humans simply asked questions via chat.<\/p>\n
A symbolic case is U.S. ride-hailing giant Uber (UBER). After encouraging the use of AI coding tools to the point of creating internal leaderboards, the company burned through its annual budget in just four months and scrambled to impose a monthly cap of $1,500 per employee. At Meta (META), employees spontaneously created a token consumption leaderboard called “Claudeonomics,” while Amazon (AMZN) is reportedly still pushing employees to “burn through tokens.” Corporate attitudes are beginning to split sharply between viewing token consumption as an engine of growth versus a cost to be controlled.<\/p>\n
Productivity gains are real, but the bills are uncapped, and there is no certainty that the value returned justifies the expense. This frustration had been accumulating in the executive suite when Nadella’s words \u2014 “learning is the one thing you can’t outsource” \u2014 and Palantir’s \u2014 “there is a reason those who sell tokens refuse value-based pricing” \u2014 struck a nerve. The soil in which this chorus took root was not ideology; it was “real invoices.”<\/p>\n
The party most acutely sensing this shift in momentum is the very target of the criticism: the big AI labs themselves. David Sacks, former White House AI czar, pointed to a pattern where Anthropic is directly moving into domains once served by applications built on top of its models \u2014 “Claude Science,” “Claude Security,” “Claude Legal,” “Claude Code.” He criticized the playbook: “Observe where value is being created, then move directly into that space. First dominate the model layer, then use that position to attack the most profitable vertical markets.” This “observe, replicate, expand” dynamic is turning into a real-world nightmare for companies building businesses dependent on large language model APIs \u2014 the fear that the data and usage scenarios they provide could become the very ammunition that helps a competitor enter their market.<\/p>\n
Meanwhile, cooler voices are emerging from inside the AI labs themselves. A source at one AI lab dismissed the uproar: “Reacting to Karp’s performance is foolish. He is simply acting in his own self-interest.” Both OpenAI and Anthropic have explicitly stated they do not use enterprise customer data for model training, and their position is that Karp’s criticism is merely sales rhetoric.<\/p>\n
Yet, it is the model providers’ own pricing strategies that have poured fuel on the fire. In a Bloomberg interview tied to the announcement of a new AI model version this past weekend, Meta CEO Mark Zuckerberg declared his entry into price competition: “Other labs’ pricing is extremely high, and their margins are abnormally fat. We believe we can offer cutting-edge or high-level intelligence at a more affordable price.” This statement indirectly corroborates Karp’s critique of the AI labs’ high-pricing structure.<\/p>\n
According to a report from South Korea’s Naver, the fundamental challenge many companies face is not simply adopting AI technology but transforming organizational culture and ways of working. Multiple chief technology officers and AI executives point out: “Even if you invest enormous sums to deploy the latest AI, if the organization and business processes remain stuck in the past, AI cannot produce real results.” The essence of AI competition is “speed” \u2014 rapid decision-making by a small group of executives and leadership willing to execute even at the risk of failure are essential. In an era where AI produces “answers,” competitive advantage will come not from those who can quickly find the right answer, but from those who can define “what to ask.”<\/p>\n
Nadella’s line in his essay \u2014 “You can outsource the work. You might even outsource the job. But you cannot outsource the learning” \u2014 carries even greater weight in this context. Delegating tasks to AI and surrendering your organization’s learning capability itself to an external party are entirely different acts. The staggering 66 million views reflect the fundamental anxiety of corporate leaders, distilled into that single sentence.<\/p>\n
The spectacle of the heads of Microsoft and Palantir \u2014 companies that themselves run massive AI-related businesses \u2014 speaking in unison to shout “protect your sovereignty” is evidence that the power balance in the AI industry has reached a critical inflection point. The tug-of-war between model providers and model users over the ownership of data, knowledge, and value is transcending mere technical debates or price negotiations. It is evolving into a strategic management issue tied directly to corporate survival.<\/p>\n","protected":false},"excerpt":{"rendered":"“Don’t depend on AI models” \u2014 the head of Microsoft (MSFT), one of the world’s largest AI investors,…\n","protected":false},"author":2,"featured_media":103725,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[11],"tags":[26998,6865,321,53,420,7829,34623,1122,320,7828,20594,157,651,2806,40855,5411],"class_list":["post-103724","post","type-post","status-publish","format-standard","has-post-thumbnail","category-microsoft","tag-ai-sovereignty","tag-alex-karp","tag-amazon","tag-anthropic","tag-azure","tag-azure-ai","tag-mai-thinking-1","tag-meta","tag-microsoft","tag-microsoft-ai","tag-ontology","tag-openai","tag-palantir-technologies","tag-satya-nadella","tag-token-capital","tag-uber"],"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/103724","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/comments?post=103724"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/103724\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media\/103725"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media?parent=103724"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/categories?post=103724"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/tags?post=103724"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}