Finally, companies that started using AI found their business was taken by large model companies

Author: Yu Hang Yuan; Source: Geek Park

On July 1, Palantir CEO Alex Karp walked into CNBC’s studio and threw out a bomb with almost out-of-control energy.

He said the AI industry is “effing insane” (crazy), he said U.S. corporate CEOs are “livid” (furious) about OpenAI and Anthropic, and he said companies are doing something absurd—paying for tokens like crazy while handing over their most core operational data to model vendors. And the resulting business value is almost incalculable.

The host asked if this was “shifting blame.” Karp replied, “No, I’m just stating facts.”

That day, Palantir’s stock price rose 9%. The number itself is a kind of vote—the market believes he said a lot of things many people wanted to say but couldn’t.

This isn’t one person venting emotion. When the head of a company with a market cap of more than one trillion dollars blasts the entire large-model industry in a live national broadcast—and the market responds with real money—that means a collective sentiment has reached a breaking point.

Over the past two years, everyone has been talking about how to embrace large models. But now, a new question is coming to the surface—if companies get too close to large models, will they end up torn apart?

01From “carrying on the hype” to “not naive”

Looking back to the start of 2024, the enterprise attitude toward large models could be summarized in four words: “use it first.”

ROI or no ROI—no matter where the data stream goes—just don’t fall behind. At that time, the mainstream narrative was: “The AI revolution is here; if you don’t embrace it, you’ll be eliminated.” Under huge pressure, CIOs and CTOs across industries shoved AI into every business component they could. This was a classic decision driven by technology panic.

By 2025, “full rollout” became the keyword. Companies began seriously embedding large models into core business processes, no longer just doing demos or internal hackathons. From customer service to code generation, from market analysis to product design, the depth and breadth of AI infiltration expanded exponentially.

But entering 2026, a subtle shift in emotion is underway.

Salesforce’s research shows that only half of IT leaders feel confident that their company’s data infrastructure can support successful AI deployment. A research report released by NTT DATA in May this year directly used the phrase “hit a wall”—enterprise AI is encountering architectural bottlenecks caused by data privacy and sovereignty requirements. Gartner predicts that by 2027, 35% of nations will rely on regionalized AI platforms, while the figure today is only 5%.

Karp put this shift more bluntly. He said companies are moving away from mindlessly consuming tokens—“tokenmaxxing”—and toward truly asking about return on investment. “The basic viewpoint is: stop wasting time on tokens.”

This isn’t denying large models; it’s the entire industry moving from ‘hype’ to ‘not being naive.’ After the frenzy fades, companies start reviewing a fundamental question with a colder, steadier lens: the things I hand over versus the things I get back—does the math actually work out?

02When partners become competitors

Karp’s criticism is still at the business-model level. But what truly sends a chill down people’s spines is another, more direct threat: your AI service provider may be using the data and context you contribute to build a product that replaces yours.

What happened in April 2026 turned this worry from theory into reality.

Back in February, Figma and Anthropic were still teaming up to develop a feature called “Code to Canvas,” integrating code generated by Claude directly into Figma’s design workflow. The two firms looked like close partners.

On April 14, Anthropic’s Chief Product Officer Mike Krieger quietly stepped down from Figma’s board seat.

Three days later, Anthropic released Claude Design—an AI design tool that can generate interactive prototypes, PPTs, and marketing materials directly from natural language, precisely targeting Figma’s core business.

Figma’s stock fell nearly 8% that day.

A later report by Fast Company highlighted a detail that’s hard to ignore—Figma and companies like Adobe and Canva had years of collaboration with Anthropic, but before Claude Design launched, nobody was notified. Everyone suddenly realized, caught off guard, that their AI partner—right under their nose—had turned into a competitor.

This story is worth thinking about because it exposes a structural problem in the large-model era, more dangerous than ever before: when you deeply cooperate with an AI company, you don’t just give up your market entry—you also give away your core understanding of the market context and your users’ needs data.

Anthropic’s ability to build Claude Design is largely because, through its collaboration with design-tool companies, it deeply understands designers’ workflows and pain points.

But if you zoom out, this isn’t a new script in the history of technology.

Amazon built its own brands from an e-commerce platform—using platform data to pinpoint the most profitable categories, then launching its own products to erode third-party sellers. Microsoft started with the operating system, then absorbed browsers, office software, and communication tools one by one—Netscape got killed, and Slack was forced to sell. Google extended from the search engine—using the search results page to answer users’ questions directly—marginalizing Yelp and many vertical information services.

The iron law of the tech industry has never changed: once a platform has enough data and understanding of users, it will move upstream to erode the source.

In the large-model era, this law becomes even more aggressive. Traditional platform erosion needs time to accumulate understanding, but large models are naturally a “knowledge accelerator.” Every API call you make and every input of business data helps the model vendor understand your territory faster and deeper.

03The “Roche limit” of the AI era

Astronomy has a concept called the Roche limit—when a celestial body gets too close to a massive star, tidal forces exceed its own gravitational pull, and the celestial body will be torn apart.

This metaphor describes today’s relationship between enterprises and large models with unsettling accuracy.

Large models are that massive star. Every company wants to borrow its gravitational force to accelerate—boost efficiency, reduce costs, innovate. But the problem is that once you get close enough, your “matter” starts to be stripped away. Your data, know-how, and understanding of user needs all flow toward the gravity center during the collaboration process.

So where is the boundary for “dancing with AI” without being ultimately consumed?

This question has already been put on the table in the U.S. But if you think it’s still far from Chinese enterprises, that may be an illusion.

There are differences between China and the U.S. in the pace of AI adoption. U.S. companies have entered large-scale, deeply embedded AI deployment, while Chinese companies as a whole are still moving from pilots toward scaling. Research released by Lenovo in partnership with IDC in March this year shows that 72% of domestic enterprises have already completed agent pilots and put them into formal use, deploying AI on an average of 3.5 scenarios. But the center of the challenge has also shifted—from “lack of compute power, lack of data” to “application outcomes not meeting expectations” and “ROI not clear.”

In other words, Chinese enterprises are entering a “period of AI clarity” similar to that of U.S. firms.

Geek Park, in recent conversations with many entrepreneurs and enterprises with traditional businesses, has found an interesting phenomenon: people’s thinking about these issues often doesn’t come from a direct crisis like “I’m worried the model company will steal my business,” but rather after AI is genuinely embedded into the business—naturally leading them to redefine: in the AI era, what is my core value?

Ultimately, this redefinition comes down to two key capabilities.

04Who controls the “AI foundation”?

The first—and most practical—capability aligns closely with what Karp emphasized: whose “foundation” do your data and business logic actually run on?

Karp repeatedly stressed on CNBC that the most sensitive operational data of enterprises should not flow into a third-party model vendor’s black box. He positions Palantir as an application layer for “sovereign AI”—models can be someone else’s, but data must stay within your own walls; deployment must run on infrastructure you control.

This isn’t paranoia. Chinese enterprises’ lived experience matches this almost perfectly. The head of product R&D at Kingsoft WPS 365, Huang Weijie, recently said something quite on point: “What enterprises lack today isn’t hardware and models, but a secure AI application layer.”

IDC data also supports this trend. In enterprise AI compute deployment, the share of public cloud is declining, while the combined share of private cloud and on-premises deployment rises from 54% to 69%. “Data staying within domain” is shifting from a compliance slogan to the first screening condition when CTOs select vendors.

Karp calls this “commodity cognition” being productized. His view is that the quality of the models themselves is converging; the real differentiating value isn’t in the model layer, but in the application layer that binds model capabilities to enterprise-specific scenarios. Palantir’s “Sovereign AI Engine,” developed in partnership with NVIDIA, is the productized version of this logic—using open-source models plus Palantir’s own ontology layer and governance framework, enabling enterprises to run AI in fully controllable environments without sending a single byte of data out. Palantir’s first-quarter 2026 revenue of $163 million, up 85% year over year, to some extent reflects the market’s vote for this path.

There’s a signal worth paying attention to here: in the future, companies and solutions that help enterprises run AI on their “own foundation” will be even more in demand. Domestically, “AI privatization brains” has become a real track, and many startups are building products around this direction. This isn’t technical cleanliness obsession; it’s a rational choice enterprises make after thinking it through.

05Don’t turn your organization into a “repeater”

The second capability is harder to quantify, but Geek Park has been feeling it more and more through conversations with companies: when AI can replace more and more execution steps, what kind of “people” does an organization actually need?

Some faster-moving enterprises have already fallen into this trap.

When AI is clearly more efficient than humans in certain steps, a natural thought is “cut the people.” But once the organization gets thinner, a hidden problem appears: what AI runs is, in essence, the “best practices” condensed by those people in the old environment. When the environment changes, the market changes, and users change, AI still faithfully executes that old logic, while the organization no longer has enough people to detect those changes and push the business forward to evolve.

Put bluntly, an organization filled by AI but hollowed out by people may just be repeating the past efficiently.

This isn’t saying you shouldn’t use AI to replace execution. It’s saying that as AI takes over more and more execution layers, enterprises actually need another kind of person—people who can “command” AI. This role requires understanding the business holistically, judging whether what AI produces still applies to a changing reality, and seeing new possibilities beyond the “optimal solutions” AI proposes.

Some leading enterprises are already seriously thinking about this. They’ve found that after AI arrives, real competitiveness isn’t “how many people you replaced with AI,” but “whether your people can master AI to do things they couldn’t do before.” If you simply keep automating and endlessly looping within historical data, you’re essentially locked into a snapshot of the past.

The importance of this cognitive flip may be no less than data sovereignty. When AI levels the technical barrier, “human judgment” and “organizational evolution capability” become the hardest things to replicate. Some companies have realized this; some haven’t. But this dividing line may become very clear in the next one or two years.

06The industry needs “new AI companies”

Over the past two years, an implicit assumption has dominated the whole industry: in the AI era, value will ultimately concentrate in model companies. The closer you are to the model, the higher the value.

That assumption is being shaken.

Karp actually points this out on CNBC: model capability itself is turning into “commodity cognition.” As the performance gap among large models keeps shrinking, real differentiation won’t be at the model layer. An industry structure where one can only be dominated by model companies isn’t just unhealthy for enterprises—it also constrains the pace of development for the entire AI industry.

Enterprises never needed a stronger model. What they need is an entire ecosystem: something that can respond to the anxiety around data sovereignty, protect competitive moats from being “siphoned,” and embed AI into business without losing control. This demand is creating a market far more complex than one where you just “sell tokens.”

Clear signals have emerged in several directions.

“Sovereign AI infrastructure” is becoming a real track where big money is being made. This isn’t just a concept. In just the first half of 2026, Europe has three companies building sovereign AI infrastructure (Nebius, nScale, AtlasEsge) that together raised more than $11.8 billion. Just a few days ago, London-based Valarian secured a $50 million Series A. The work is specific: add a “sovereign control layer” between AI systems and sensitive data, deciding which AI can access which data, and under what conditions. Two years ago, there was basically no such demand—now governments and large enterprises are lining up.

“AI gateways” and orchestration layers are becoming an indispensable part of enterprise AI architecture. When a company uses OpenAI, Anthropic, open-source models, and also its own fine-tuned specialized models, who handles unified routing, cost control, permission governance, and auditing? In traditional software, this position is called middleware; in the AI era, it’s called a gateway or orchestration layer. It’s not flashy, but it’s a critical foundational infrastructure for enterprises to move from “using AI” to “managing AI.” Palantir essentially does this layer—only in the heaviest version. More lightweight solutions for enterprises of different sizes have huge room.

At the application layer, vertical industry AI solutions are also shifting from “wrapping” to “going deep.” In the past, many so-called AI applications were, when you strip it down, just wrapping a GPT shell. But now, what can truly stand on its own are products that deeply understand industry-specific know-how and tightly bind AI capabilities to industry logic. The value anchor of these companies isn’t in the models; it’s in industry cognition—precisely the kind of thing large model companies can’t easily obtain through training.

Even at the “people” layer, a new service market is emerging. As more enterprises realize they don’t need more AI tools, but rather people who can “command AI” and the accompanying organizational methodologies, demand for consulting on organizational transformation in the AI era, talent development, and process redesign is also quickly taking shape.

At the end of the day, an industry that only has a “model layer” is fragile. What truly helps the AI industry run faster and healthier is a more three-dimensional ecosystem. In this ecosystem, some build models, some build sovereign infrastructure, some do gateways and governance, some build deeply specialized vertical applications, and some help enterprises reshape their organizational capabilities. Each layer addresses the real needs as enterprises move from “embracing” to “mastering” AI.

Over the past year, these needs have gone from vague to increasingly clear. Next, the new generation of solutions, service providers, and products born around these needs may be entering a period of clear breakout.

Returning to the Roche limit metaphor. Finding that safe orbit is never something one enterprise can do alone. When the entire ecosystem starts growing forces beyond models, enterprises will truly have the confidence to not be torn apart.