In mid-August, information emerged that Washington was preparing an ultimatum for dozens of countries, demanding they decide whose side they would take in the technological confrontation with China. Formally, this looks like another round of superpower rivalry, but behind the political gesture lies a much deeper problem.
The State Department prepared a letter for 35 countries that had previously signed the American "Statement on AI Capabilities." Those who choose the Chinese framework face exclusion from the Pax Silica initiative—a project to control supply chains for models, semiconductors, and critical minerals. About two dozen states have joined the American initiative, but Kazakhstan draws particular attention: the country entered both coalitions simultaneously, and it is precisely its dual position that causes concern in Washington.
Beijing is acting symmetrically. In the summer, Chinese President Xi Jinping announced the creation of a World Organization for AI Cooperation, promoting Chinese technology as an alternative to American influence. However, in my assessment, the balance of power is no longer so clear-cut. The United States retains primacy in fundamental research and the development of advanced models, but China is rapidly closing the gap—primarily in efficiency, applied technology use, and robotics.
The demand to choose a side is not a demonstration of superiority but a symptom. The American AI model struggles to compete with China's bet on openness and low prices. Political pressure has become an attempt to compensate for what cannot be achieved through market methods.
An Economy That Does Not Add Up
The key vulnerability of the American approach is the gap between investments and real returns. Since 2023, U.S. tech giants have poured hundreds of billions of dollars into AI, but monetization remains minimal. Microsoft's capital expenditures reached $116 billion over the year, while direct revenue from its AI business is estimated at $40–45 billion. The ratio of investments to revenue has jumped from the usual 7–10% to 35%, and cash flow over seven years has grown only 1.6 times against a 2.7-fold increase in revenue.
The position of competitors is no better. Google showed negative operating cash flow for the first time, halted share buybacks, and over fifteen months grew its debt from $30 billion to $120 billion. Amazon went into the red by $25 billion over six months, and Meta spends 97% of its operating cash flow on AI. Direct revenues are incomparable to the scale of costs: Microsoft 365 Copilot gathered about 30 million subscriptions against investments of $190 billion, while Google's potential revenue from subscriptions and APIs is estimated at $15–20 billion against investments of $200 billion.
This entire structure rests on hope: after market consolidation, American companies expect to dictate prices and finally recoup their costs. It is precisely this prospect that China is destroying.
China's Bet on Cheapness and Openness
Instead of racing for "intelligence" at any cost, Chinese developers have bet on cost efficiency and open access. A clear example is DeepSeek V4 Flash: the model costs almost nothing, about $0.08 per million input tokens, yet delivers performance at the level of American flagships from spring 2026.
The response was forced price cuts from OpenAI. The GLM-5.3 model from Z.ai, at a price of about $2.5 per million tokens, reached a level between GPT-5.5 and GPT-5.6 versions and is additionally optimized for Huawei's Chinese accelerators, reducing dependence on Nvidia. The result has directly impacted the market: according to data from the largest traffic router OpenRouter, a year ago American models accounted for 75–85% of requests, while now 60–70% of token traffic is provided by Chinese solutions—DeepSeek, GLM, and others.
The reason is simple: for mass tasks like code generation, data processing, and agentic execution, businesses do not need expensive "genius." What is required is reliability, speed, and low price—and all of this is offered by China.
By releasing powerful models almost for free, China devalues American investments: each such release pushes the price bar down and deprives the United States of any chance to ever recoup its enormous investments.
Technology Parity and the Price of Isolation
The quality gap, on which Americans counted to maintain a premium, has almost disappeared in the mass segment. Flagships like GPT-5.6 and Opus-5 still lead in complex tasks such as scientific reasoning, but models like Kimi K3, GLM-5.3, and Qwen-3.8 have come very close to them while working noticeably faster and cheaper. At this pace, China could take the lead as early as 2027.
Google's position is especially telling: possessing data, chips, and engineers, the company still lost its leadership and is now making up ground with cheap models and dumping. This points to a systemic flaw in the American approach—a reliance on computing scale without proper optimization.
It is economic weakness that is pushing Washington toward political measures. I interpret the demand to choose a side as a sign of vulnerability: the United States is trying to administratively isolate China from the global market because it can no longer maintain sales of expensive models through market means. Such a strategy carries serious risks. The world could split into two technological blocs with duplicative standards, developing countries, due to coercion, could demonstratively leave the American coalition, and restrictions would only accelerate China's development of its own chips and algorithms.
The final picture is paradoxical. In terms of investments, the United States is far ahead, but that money is not paying off and is piling up debt; in terms of actual usage, China already dominates the mass segment; in terms of technology, parity has been achieved with a trend not in America's favor. Administrative barriers rarely work when a competitor's product is cheaper and good enough, so artificial isolation risks only accelerating the formation of an independent Chinese AI bloc.
My conclusion: Washington's ultimatum is not strength, but an admission of weakness. The market has already voted for efficiency, and political measures are unlikely to change that choice—they will only accelerate the fragmentation of the global technological ecosystem.