They not only want your money but also want your business. On July 1st, Palantir CEO Alex Karp walked into the CNBC studio and almost uncontrollably threw a bombshell.
He said the AI industry is "effing insane," he said American corporate CEOs are "livid" about OpenAI and Anthropic, he said companies are doing something absurd—madly paying for tokens while handing over their most core operational data to model suppliers. And in return, the business value obtained is almost immeasurable.
When asked by the host if he was "passing the buck," Karp replied, "No, I'm just stating the facts."
Palantir's stock price rose by 9% that day. This number itself is a vote—the market believes he has said what many people have thought but not dared to say.
This is not just one person's emotional outburst. When the CEO of a company with a market value of over a trillion dollars takes aim at the entire big model industry in a nationwide live broadcast, and the market responds with tangible positive feedback, it signifies that a collective mood has reached a tipping point.
Over the past two years, everyone has been talking about embracing big models. But now, a new question is emerging—if a company gets too close to the big model, will it be torn apart by it?
01 From "Enthusiasm" to "No Longer Naive"
Thinking back to early 2024, the corporate attitude toward big models can be summarized in four words—"just use it first."
Regardless of ROI, regardless of where the data goes, just don't fall behind. The mainstream narrative at that time was "the AI revolution is here, and not embracing it will lead to elimination." CIOs and CTOs from various industries were under tremendous pressure, squeezing AI into every possible business process. It was a typical decision driven by tech panic.
By 2025, "full deployment" became the keyword. Companies began seriously embedding big 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 penetration have expanded exponentially.
But as we enter 2026, a subtle emotional shift is taking place.
Research from Salesforce shows that only half of IT leaders are confident in their company's data infrastructure to support the successful implementation of AI. A study released by NTT DATA in May of this year directly used the term "hitting a wall"—enterprise AI is facing an architectural bottleneck due to data privacy and sovereignty requirements. Gartner predicts that by 2027, 35% of nations will rely on regionalized AI platforms, a number that stands at only 5% today.
Karp put this transition more bluntly. He said that companies are moving away from mindlessly stacking tokens in what he called "tokenmaxxing" to genuinely questioning the return on investment. "The basic idea is, stop wasting time on tokens."
This is not a denial of big models, but rather the entire industry is shifting from "hype" to "sobriety." After the frenzy, companies are starting to take a more sober look at a fundamental question—does the trade-off between what I give and what I get back make sense?
02 When Partners Turn into Competitors
Karp's criticism remains at the business model level. But what truly sends shivers down the spine is another more direct threat—your AI service provider may be using the data and contextual understanding you contribute to build a product that competes with yours.
What happened in April 2026 turned this concern from theory into reality.
In February of this year, Figma and Anthropic were still collaborating on a feature called "Code to Canvas," seamlessly integrating Claude-generated code into Figma's design process. The two companies seemed like close partners.
On April 14, Mike Krieger, Chief Product Officer of Anthropic, quietly resigned from Figma's board.
Three days later, Anthropic launched Claude Design—a tool that can directly generate interactive prototypes, PowerPoint presentations, and marketing materials using natural language, directly competing with Figma's core business.
Figma's stock price dropped nearly 8% that day.
In a later report by Fast Company, there was a particularly intriguing detail—Figma and companies like Adobe and Canva had longstanding partnerships with Anthropic, but no one was notified before Claude Design's release. Everyone was caught off guard, realizing that their AI partner had turned into a competitor right under their noses.
The reason this story is worth pondering is that it exposes a structural issue in the era of large models—a more dangerous situation than ever before. When you deeply collaborate with an AI company, you not only give up market access but also your core understanding of use cases and user data.
Anthropic was able to create Claude Design to a large extent because, through its partnerships with design tool companies, it deeply understood the designer's workflow and pain points.
But taking a broader view, this is not a new script in tech history.
Amazon transitioned from an e-commerce platform to its own brands, using platform data to precisely identify the most profitable categories, and then launched its products to encroach on third-party sellers. Starting from an operating system, Microsoft gradually incorporated browsers, office software, and communication tools—Netscape was killed, and Slack was forced to sell. Google extended from a search engine, using search result pages to directly answer user queries, leading to Yelp and many vertical information service providers being marginalized.
The iron rule of the tech industry has never changed—once a platform has enough data and user understanding, it will erode upstream.
In the era of large models, this rule has become even more brutal because traditional platform erosion takes time to accumulate understanding, while large models are a natural "understanding accelerator." Each of your API calls, each input of business data, is helping the model provider understand your domain faster and deeper.
03 The "Roche Limit" of the AI Era
There is a concept in astronomy called the "Roche Limit"—when a celestial body is too close to a massive star, tidal forces exceed its own gravity, and the body is torn apart.
This analogy to describe today's relationship between enterprises and large models is disturbingly accurate.
The large model is that massive star. Every enterprise wants to leverage its gravitational pull to accelerate—efficiency, cost reduction, innovation. But the problem is, when you get close enough, your "matter" begins to be stripped away. Your data, know-how, understanding of user needs—all flow towards the gravitational center during the collaboration.
What are the boundaries for a company to “dance with AI” without being eventually consumed by it?
This question has already been raised in the United States. However, if you think it is still far from Chinese companies, that may be an illusion.
There are differences in the pace of AI application between Chinese and American companies. American companies have already entered a stage of large-scale, deep business AI deployment, while Chinese companies are still moving from pilot projects to scale. A survey released by Lenovo in collaboration with IDC in March this year showed that 72% of domestic enterprises have completed smart body pilots and put them into formal use, deploying AI in an average of 3.5 scenarios. However, the focus of the challenges has shifted from "lack of computing power and data" to "application effects falling short of expectations" and "unclear ROI."
In other words, Chinese companies are entering a "wake-up call" phase similar to American companies.
Geek Park recently discovered an interesting phenomenon when talking to many entrepreneurs and traditional businesses — many people's reflections on these issues are often not directly due to a fear of "model companies taking away my business," but rather, after truly integrating AI into their business, they naturally begin to redefine "In the AI era, what is my core value."
This redefinition will ultimately focus on two key capabilities.
04 Who Controls the "AI Foundation"?
The first, and most practical, is highly consistent with what Karp said — whose foundation is your data and business logic really built on?
This is the core argument that Karp repeatedly emphasized on CNBC. A company's most sensitive operational data should not flow into a third-party model vendor's black box. He positioned Palantir as an application layer that provides "sovereign AI" — while the model can be provided by others, the data must remain within their own walls, and deployment must be on their own controllable infrastructure.
This is not paranoia; Chinese companies' sentiment is actually completely aligned. Wei-Jie Huang, head of product research and development at Kingsoft WPS 365, recently made a very pertinent statement — "Today, what companies lack is not hardware and models, but a secure AI application layer."
IDC's data also confirms this trend. In enterprise AI computing power deployment, the proportion of public cloud is decreasing, while the total proportion of private cloud and on-premises deployment has increased from 54% to 69%. "Data sovereignty" is evolving from a compliance slogan to the first screening criterion for CTOs when selecting vendors.
Karp refers to this as "commodity cognition." His argument is that the quality of the model itself is converging, and the true differentiated value is not at the model layer, but at the application layer that binds the model's capabilities to enterprise-specific scenarios. The "Sovereign AI Engine" launched by Palantir in partnership with NVIDIA is a productization of this logic—using open-source models coupled with Palantir's own ontology layer and governance framework to allow enterprises to run AI in a fully controlled environment without any data leaving. Palantir reported $16.3 billion in revenue for Q1 2026, an 85% year-over-year growth, which to some extent is the market's vote of confidence in this path.
There is a noteworthy signal here—companies and solutions that help enterprises run AI "on their own turf" in the future will be more sought after. In China, the "AI privatized brain" has already become a real track, with many startups building products around this direction. This is not about technological obsession but a rational choice made by enterprises after careful consideration.
05 Don't Turn Your Organization into a "Parrot"
The second capability, harder to quantify but strongly felt by GeekPark in its interactions with enterprises, is when AI can replace more and more execution tasks, what kind of "people" does the organization really need?
Some fast-moving companies have already fallen into this trap.
When AI's efficiency surpasses that of humans in certain aspects, a natural thought is to "cut people off." However, after the organization becomes leaner, a hidden problem begins to emerge—what AI is running is fundamentally the "best practices" consolidated by these people in the old environment. When the environment changes, the market changes, the users change, and AI faithfully executes that old logic, while there are no longer enough people in the organization to perceive these changes and drive the business forward.
In essence, an organization filled with AI but hollowed out by people is likely just efficiently repeating the past.
This is not to say that AI should not replace execution. Instead, it means that as AI takes over more and more execution layers, enterprises actually need a different kind of people—not those traditionally executing specific tasks, but those who can "command" AI. This role requires an understanding of the business's overall picture, the ability to judge whether the output of AI is still relevant to a changing reality, and the vision to see beyond the "optimal solution" provided by AI.
Some forward-thinking companies have already begun to seriously consider this issue. They have found that after having AI, the real competitive advantage is not "how many people AI has replaced," but "whether your people can harness AI to do things that were previously impossible." If AI is only used to continuously automate and iterate on historical data, then you are essentially trapped in a snapshot of the past.
The importance of this cognitive shift may be no less than data sovereignty. As AI flattens technological barriers, "human judgment" and "organizational adaptive capability" have become the most difficult things to replicate. Some companies have already realized this, while others have not. But this turning point is likely to become very clear in the next year or two.
06 Industries Need "New AI Companies"
Over the past two years, an implicit assumption has dominated the entire industry—the value of the AI era will ultimately be concentrated in the hands of model companies. The closer one is to the model, the higher the value.
This assumption is being shaken.
Karp actually pointed out something on CNBC—the model itself is becoming commoditized cognition. As the capabilities of various large models become more similar, the real differentiation is no longer at the model layer. An industry structure dominated only by model companies is not only unhealthy for enterprises, but also a constraint on the development speed of the entire AI industry.
What enterprises need is never a stronger model. What they need is a whole ecosystem—capable of addressing the anxiety of data sovereignty, protecting competitive barriers from being "siphoned," and enabling AI to be truly embedded in business without losing control. This demand is giving rise to a market much more complex than "selling tokens."
Several directions have already shown clear signals.
"Sovereign AI Infrastructure" is becoming a real and well-funded track. This is not just a concept. In the first half of 2026 alone, three companies in Europe (Nebius, nScale, AtlasEsge) working on sovereign AI infrastructure raised over $11.8 billion in total. Just a few days ago, London-based Valarian raised a $50 million Series A round, doing something very specific—adding a "sovereignty control layer" between AI systems and sensitive data, deciding which AIs can access which data and under what conditions. This kind of thing was not needed at all two years ago, but now governments and large enterprises are lining up for it.
The "AI Gateway" and Orchestration Middle Layer are becoming an indispensable part of enterprise AI architecture. When a company is using OpenAI, Anthropic, open-source models, and its own fine-tuned proprietary models simultaneously, who will handle unified routing, cost control, permission governance, and auditing? In the traditional software era, this role was known as middleware; in the AI era, it is called a gateway or orchestration layer. It may not be glamorous, but it is a key infrastructure for enterprises to transition from "using AI" to "managing AI." Palantir essentially operates in this space, albeit in the most heavyweight version. Lighter-weight solutions tailored to enterprises of different scales have significant potential.
At the application layer, AI solutions for vertical industries are also moving from "shell" to "depth." Many so-called AI applications in the past were essentially wrapped in a layer of GPT. However, now, the ones that truly stand out are those that deeply understand industry-specific know-how and tightly integrate AI capabilities with industry logic. The value anchor of these companies lies not in the model but in industry knowledge—a quality that large-model companies find challenging to acquire through training.
Even at the "human" level, new service markets are emerging. As more and more companies realize that what they need is not more AI tools but people who can "command AI" and the accompanying organizational methodologies, the demand for organizational transformation consultancy, talent development, and process redesign around the AI era is rapidly surfacing.
Ultimately, an industry with only a "model layer" is fragile. What truly accelerates and sustains the AI industry is a more comprehensive ecosystem. In this ecosystem, some work on models, some on sovereign infrastructure, some on gateways and governance, some on deep vertical industry applications, and some help reshape organizational capabilities for enterprises. Each layer is responding to the real needs of enterprises as they transition from "embracing" to "controlling."
These needs have become increasingly clear over the past year. Next, a new generation of solutions, service providers, and products born around these needs may experience a distinct period of rapid growth.
Returning to the metaphor of the Loshin Limit. Finding that safe track is never the job of a single enterprise. When the entire ecosystem starts to develop strengths beyond just models, enterprises truly gain the confidence not to be torn apart.
Welcome to join the official BlockBeats community:
Telegram Subscription Group: https://t.me/theblockbeats
Telegram Discussion Group: https://t.me/BlockBeats_App
Official Twitter Account: https://twitter.com/BlockBeatsAsia
