CNBC turns Hesai’s lidar partnership with Nvidia into a national-security panic while treating Pentagon blacklist politics as neutral truth. The buried facts show that U.S. law already compels corporate cooperation with surveillance, while U.S. intelligence has long weaponized global digital infrastructure. The real story is technofascism: the merger of state power, Big Tech, surveillance law, and imperial competition. The task is to fight mass surveillance and blacklist politics without joining the new Cold War against China.
Prince Kapone | Weaponized Information | July 8, 2026
When Intelligence Gets Too Cheap for the Empire
On July 7, 2026, CNBC published “Chinese AI models are gaining ground with U.S. companies as OpenAI, Anthropic costs surge” by Kai Nicol-Schwarz, a business story that reads at first like a simple report on enterprise software costs. U.S. companies, CNBC tells us, are increasingly using Chinese-built AI models because those models are cheaper, increasingly competitive, and easier to adapt than the closed systems sold by the great American temples of artificial intelligence. The article names DeepSeek, Z.ai, and Alibaba’s Qwen as examples of Chinese models now entering U.S. business workflows, while OpenAI and Anthropic appear as the expensive incumbents whose token prices are making corporate customers search for alternatives. On the surface, this is a market story. Beneath the surface, it is a confession that the empire is losing its grip on the price of intelligence.
The article’s first device is market anxiety framing. CNBC does not treat Chinese AI adoption as a triumph of cheaper and more efficient technological development. It treats it as a problem to be managed. The facts are difficult to hide: U.S. companies are choosing Chinese models because the cost curve makes sense. But instead of following that fact into the political economy of U.S. AI monopoly power, the article keeps the focus on the discomfort of American incumbents and regulators. Chinese models are allowed to be impressive, but only as a disturbance inside an order where U.S. firms are supposed to sit naturally at the frontier and collect rent from the rest of the world.
The second device is cost mystification. The article repeatedly tells us that Chinese models are cheaper, that companies are saving money, that token costs are rising, that businesses are becoming more cost-conscious, and that firms are routing tasks to cheaper models when the best proprietary model is unnecessary. But it does not ask why U.S. AI is so expensive in the first place. It does not excavate the cloud monopolies, GPU dependency, subscription lock-in, proprietary APIs, venture-capital expectations, data-center buildout, and monopoly pricing that sit behind the invoice. The reader is shown the price tag, but not the class relation printed in invisible ink underneath it.
The third device is soft threat construction. Unlike the CNBC lidar piece, this article does not bang the war drums with talk of backdoors, military nodes, and sabotage. It is more subtle. Chinese AI is presented as useful, competitive, and financially attractive, but its growing adoption is immediately placed beside U.S. state efforts to regulate advanced models and slow the spread of overseas alternatives. That is the velvet glove version of the same imperial hand. When Chinese technology is too weak, it is mocked. When it becomes too strong, it becomes a security problem. When it becomes cheaper than the American product, suddenly the free market needs a border guard.
The fourth device is omission. CNBC mentions open-source and open-weight models, but it does not stay with the contradiction those words contain. Open-weight models threaten the whole rent structure of closed AI because they make intelligence more portable, inspectable, adaptable, and replaceable. If a company can route workloads across models, host systems locally, substitute providers, and avoid total dependence on one proprietary API, then the monopoly wall begins to crack. That is not simply a technical preference. It is a struggle over whether artificial intelligence becomes a toll road owned by a handful of U.S. firms or a productive force that can be reproduced, distributed, and made cheaper.
The ideological function of the article is therefore double. It reports a real market shift, but it prevents the reader from seeing the full historical meaning of that shift. The crisis is not that China is “catching up” in some narrow nationalist scoreboard. The crisis is that capital is trying to privatize intelligence as a monopoly commodity at the very moment technological development is pushing intelligence toward wider reproduction and lower cost. CNBC gives us the symptom: U.S. companies are turning to Chinese models. The real diagnosis is more dangerous: AI enclosure is beginning to fail.
The Price War Beneath the Model War
The first fact is the one CNBC cannot avoid: Chinese AI models are gaining ground inside U.S. companies because they are cheaper, increasingly capable, and easier to substitute into production workflows. The share of tokens used by U.S. companies on Chinese-origin models through OpenRouter has stayed above 30% every week since February 8, 2026, and peaked as high as 46%, after averaging only 11% across the previous twelve months and 4.5% in the first half of 2025. OpenRouter’s Justin Summerville put the cost difference plainly: open Chinese models can run 60% to 90% cheaper than leading Anthropic and OpenAI models. That is not a slogan. That is a price signal cutting through the fog of imperial confidence.
The same pattern appears in production routing. Vercel’s June 2026 AI Gateway index shows that DeepSeek V4 jumped from less than 1% of AI Gateway tokens in April to 17% in May, becoming the third-largest provider by volume while accounting for only about 1% of spend. DeepSeek V4 Flash launched at prices roughly 20 to 50 times lower than comparable Anthropic models, and Vercel’s own conclusion was that teams found the output good enough for production, not merely cheap enough for experimentation. Z.ai’s GLM 5.2 entered Vercel’s AI Gateway in June with a one-million-token context window, no markup, and bring-your-own-key support. Braintrust’s comparison of GLM 5.2 against Anthropic’s Opus 4.8 found GLM slightly behind on retrieval, while average provider cost per trace was roughly 76% to 78% lower. The technical story and the business story converge: Chinese models are not merely cheaper; they are cheap enough and capable enough to alter actual deployment behavior.
This is the article’s buried political economy. U.S. AI firms do not simply sell intelligence. They sell dependency. Closed APIs, subscription access, proprietary weights, cloud integration, vendor lock-in, model routing fees, and high token prices turn artificial intelligence into a rent-bearing utility. The user does not own the model. The developer does not control the stack. The small firm does not command the infrastructure. It pays the gatekeeper. Open-weight competition threatens that arrangement because it makes models more portable, inspectable, adaptable, and replaceable. The Linux Foundation’s Model Openness Framework identifies model completeness through components such as data, code, weights, evaluation, documentation, and other release materials, all aimed at reproducibility, transparency, and usability in AI. This is why “open weight” is not a technical footnote. It is a crack in the enclosure wall.
The chip war created the conditions for this model war. The United States has spent years trying to restrict China’s access to advanced computing, semiconductor equipment, high-bandwidth memory, and the industrial base needed to train and deploy frontier systems. The Biden-era AI Diffusion framework tied advanced computing controls to AI systems by restricting advanced computing integrated circuits, high-bandwidth memory, and model weights for the most advanced AI systems. The Trump Commerce Department later moved to rescind that framework while still strengthening other measures and warning that using U.S. AI chips for Chinese AI model development was contrary to U.S. national security and foreign-policy interests. The form changes; the objective remains. Washington wants to police the hardware layer, the software layer, and the model layer because it understands that AI power is not floating in the clouds. It rests on chips, electricity, data centers, memory, cloud platforms, software ecosystems, and industrial supply chains.
China’s AI rise is therefore not occurring outside pressure. It is developing under siege conditions created by U.S. export controls and technological containment. Reuters reports that China is weighing limits on overseas access to advanced AI models at the same time that Chinese open-source models are widely adopted around the world because of technical strength and cost efficiency. Reuters also reports that China is considering allowing top domestic AI firms such as Alibaba, ByteDance, and DeepSeek to buy limited quantities of Nvidia H200 chips, reflecting the continuing pressure on Chinese computing capacity under U.S. restrictions and the ongoing struggle between foreign dependence and domestic substitution. The model price war cannot be separated from this hardware struggle. Cheap inference is not magic. It is a result of engineering, optimization, infrastructure, industrial policy, and pressure from a hostile imperial environment.
The CNBC article also hides the Big Tech accumulation bloc behind the names of a few model companies. U.S. AI is organized through OpenAI, Anthropic, Google, Microsoft, Amazon, Meta, Oracle, Nvidia, cloud providers, venture capital, sovereign capital, and military procurement. The user sees a chatbot. Behind the chatbot stand GPU clusters, hyperscale data centers, electricity contracts, water use, platform dependency, and finance capital demanding returns. Weaponized Information has already established this terrain. “Empire on Extension Cord” showed that AI rests on energy bottlenecks, data centers, public subsidy, fossil-fuel revival, and military-cloud integration. “Empire in a Lab Coat” showed that Nvidia, chips, and reindustrialization are part of a larger imperial attempt to rebuild U.S. technological command. “Strategic Stability or Strategic Pause” placed chips, AI, Taiwan, rare earths, and military encirclement inside the struggle over the infrastructure of the twenty-first century. This CNBC article belongs inside that same line, but at the model-pricing layer.
The labor question is buried deepest. CNBC writes about companies seeking efficiencies, routing workloads, reducing costs, and choosing cheaper stacks. That language is bloodless because capital prefers bloodless words when it is preparing to reorganize labor. AI adoption means work speedups, task automation, monitoring, deskilling, job cuts, intensified productivity demands, and new forms of management over workers whose activity becomes data for the system. Cheaper models do not automatically democratize AI. They can just as easily make automation cheaper for bosses. The decisive question is not only which model performs best or costs least. The decisive question is who controls deployment, who owns the infrastructure, who captures the savings, and who bears the consequences when “efficiency” walks into the workplace wearing a manager’s smile.
The Enclosure of Intelligence Is Beginning to Crack
The real story is not that Chinese AI models are gaining popularity among U.S. companies. The real story is that artificial intelligence is becoming harder to keep enclosed. Capital wants to turn intelligence into a toll road. It wants every prompt, every workflow, every task, every document, every customer-service exchange, every programming assist, and every act of machine reasoning to pass through a privately owned gate where rent can be collected. That is the dream behind closed AI: not simply better models, but permanent dependency.
But technological development does not obey the fantasy life of monopolists. Once a productive force becomes reproducible, substitutable, and cheap enough to circulate beyond the original gatekeepers, the old rent structure begins to tremble. This is what CNBC accidentally documents. U.S. firms are not moving traffic to Chinese models because they suddenly discovered international solidarity. They are doing it because the arithmetic of production is forcing their hand. The American model is expensive. The Chinese alternative is cheaper. The task does not always require the most costly frontier system. So workloads move. The market, that sacred idol before which the empire usually kneels, has begun to betray the empire.
This is the contradiction of AI enclosure. The great U.S. AI firms want intelligence to remain scarce, proprietary, metered, and leased. They want model access to look like a utility bill and function like a tribute system. The user pays. The developer pays. The small business pays. The public institution pays. The cloud platform collects. The model owner collects. The chip supplier collects. The whole imperial stack feeds at the trough. But open-weight and low-cost models disturb this arrangement because they make intelligence more portable, more inspectable, more adaptable, and more replaceable. They give users and firms a way to route around the tollbooth.
That is why this is bigger than OpenAI versus DeepSeek, Anthropic versus Z.ai, or one benchmark against another. The struggle is over whether artificial intelligence becomes a monopoly utility controlled by a few U.S.-centered firms or a broader productive force that can be reproduced and used across different infrastructures. In capitalist society, even cheap tools can become weapons against labor. But the tendency itself matters. When the cost of a productive force falls, monopoly power must either adapt, crush the competition, or call the state for protection. The AI giants are already arriving at that point.
The chip war is now becoming a model war because Washington understands that control over hardware alone is insufficient. The United States tried to police the chips, the memory, the data centers, and the compute flows. But Chinese firms responded through engineering, optimization, substitution, and open-weight releases that made useful AI available despite the siege. If the hardware gate cannot fully stop the advance, then the model itself becomes the next border. That is why regulation, export controls, national-security warnings, and geopolitical anxiety now move from silicon to software, from GPU clusters to model weights, from the factory floor to the inference endpoint.
The empire’s problem is that it cannot openly say what it wants. It cannot say: we want intelligence priced high enough to preserve monopoly rents. It cannot say: we want Chinese models excluded because they make U.S. models look bloated and expensive. It cannot say: we believe the world should pay American firms for access to the productive force of machine intelligence. So it speaks in softer language. Security. Frontier risk. Responsible AI. Trusted providers. Strategic competition. The words change, but the class content remains. The state steps forward when the market stops obeying the empire’s desired hierarchy.
Here the Big Tech accumulation bloc comes into view. AI is not a lonely model sitting in the ether. It is a whole industrial formation: chips, cloud, energy, data centers, APIs, subscriptions, platform dependency, finance capital, and military procurement. The chatbot is only the friendly mask on the machine. Behind it stand power plants, water systems, server farms, semiconductor supply chains, hyperscale cloud contracts, and the expectation that every layer must generate returns for capital. When U.S. AI becomes expensive, it is not an accident. It reflects the cost of the whole enclosure: the infrastructure, the monopoly position, the speculative investment, the cloud dependency, and the rent expectations of the firms that own the stack.
Chinese models puncture that enclosure because they expose how much of the U.S. AI price is not simply intelligence, but tribute. They show that comparable usefulness can be delivered at far lower cost. They force developers and firms to ask why they should remain chained to the most expensive provider when cheaper systems can do the work. This is why the matter becomes geopolitical so quickly. A cheaper model is not merely a cheaper model when the expensive model belongs to the empire. It is a breach in the payment system of digital supremacy.
But the working class must be clear-eyed. Cheaper AI is not socialism. Open weights are not workers’ power. A lower token price does not mean liberation. Capital can use cheaper models to automate more jobs, intensify more labor, surveil more workers, and cut deeper into living standards. The falling cost of intelligence can democratize access only if the people organize over the terms of deployment. Otherwise, the same productive advance that weakens monopoly rent at one level becomes a sharper weapon of exploitation at another.
That is the dialectic hidden inside CNBC’s business story. Chinese AI is helping break the enclosure of intelligence, but capital will try to capture every break in the wall. U.S. monopoly AI wants to preserve scarcity. Chinese competition pushes toward cheaper reproduction. Employers see lower costs and dream of labor discipline. Workers see the possibility of tools that could reduce toil, widen access, and expand collective capacity. The same technology contains both possibilities because it is born inside class society. The question is not whether AI is good or bad in the abstract. The question is which class commands it, which state protects it, which infrastructure carries it, and whose future it is made to serve.
From the standpoint of oppressed nations, this contradiction has global meaning. For decades, imperialism has tried to monopolize the commanding technologies of development and sell them back to the world as dependence. Now Chinese firms are helping make advanced AI cheaper and harder to contain. That does not abolish contradiction inside China, nor does it make every company a revolutionary actor. But it does alter the world balance. It weakens the claim that the future must be rented from Silicon Valley. It opens space for technological sovereignty, for digital sovereignty, for countries and peoples to ask whether intelligence must remain locked behind U.S. corporate gates.
CNBC gives us a market trend. The real story is a historical fracture. The empire built an AI enclosure and called it innovation. It wrapped monopoly rent in the language of frontier science. It placed the price meter beside the model and told the world this was the natural cost of progress. Then the cost began to fall, the alternatives began to spread, and the very companies inside the empire began reaching for the cheaper tools built by the rival it was told to fear. That is not just competition. That is the beginning of a crisis in the ownership of intelligence itself.
Break the AI Toll Road
The answer is not to cheer for one corporation against another, nor to hand the people’s future to whichever model is cheapest this quarter. The answer is to fight the enclosure itself. That begins by defending real openness against corporate open-washing. The Open Source Initiative’s work on the Open Source AI Definition is useful terrain because it forces the question of what must actually be released for an AI system to be meaningfully open. The Linux Foundation’s Model Openness Framework is also useful because it evaluates model releases around openness, completeness, and transparency, even as we must remain clear that corporate participation in open-source ecosystems can turn openness into another branding exercise.
But openness by itself is not enough. The struggle must also confront Big Tech power, labor discipline, and the state-backed monopoly structure of U.S. AI. DAIR’s work is valuable because it challenges the corporate capture of AI ethics and centers the communities and workers harmed by automated systems, while Public Citizen’s Big Tech accountability work exposes how corporate power shapes digital policy, lobbying, and regulation. These are not substitutes for revolutionary organization. They are terrains of intervention. We use them to sharpen the fight over procurement, transparency, labor protections, public compute, open standards, and democratic control.
The concrete tasks are already in front of us. Workers and unions should demand bargaining rights over AI deployment: no secret automation, no productivity quotas imposed by algorithm, no deskilling without compensation, no surveillance disguised as efficiency, and no job cuts hidden behind the language of innovation. Schools, universities, libraries, journalists, small businesses, and public agencies should demand AI procurement rules that require portability, open standards, auditability, local deployment options, and anti-lock-in clauses. Developers should resist being chained to closed APIs when open-weight, inspectable, and replaceable systems can reduce dependency on monopoly platforms.
At the same time, the anti-monopoly fight must not be captured by the new Cold War. U.S. AI giants will use “China risk” to discipline public institutions and private firms into paying tribute to closed American platforms. That must be rejected. Oppose anti-China model bans, export-control escalation, and blacklist politics that consolidate OpenAI, Anthropic, Microsoft, Google, Amazon, Oracle, Nvidia, and the rest of the AI accumulation bloc under the cover of security. Build teach-ins that connect token rents, cloud lock-in, chip controls, model controls, labor displacement, surveillance, data centers, and military AI into one map of power.
The slogan should be clear: no AI enclosure, no new Cold War, no monopoly ownership of intelligence. The people need tools that reduce toil, expand capacity, and serve social need. Capital wants tools that cheapen labor, raise rents, discipline workers, and police the world. That contradiction will not be solved by choosing the least expensive API. It will be solved only when workers, communities, and oppressed nations fight for power over the infrastructure of intelligence itself.
Weaponized Information 
Leave a comment