The US is preparing an ultimatum for dozens of countries, demanding they choose sides in the technological standoff with China. However, behind this political pressure lies not confidence in its own superiority, but a deep structural crisis in the American artificial intelligence model. Let's examine why administrative barriers are a symptom of weakness, not a strategy of strength.
In mid-August, information emerged that Washington would send a warning to partners: joining the rival Chinese structure would result in exclusion from the US-led AI coalition. Formally, this looks like another round of superpower rivalry, but behind the political gesture lies a much deeper problem.
The ultimatum and the balance of power
The State Department has prepared a letter for 35 countries that previously signed the American "Declaration on AI Capabilities." Those who choose the Chinese side 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. Kazakhstan deserves special attention, as it entered both coalitions simultaneously, causing concern in Washington over its dual position.
Beijing is acting symmetrically. In the summer, Chinese President Xi Jinping announced the creation of a World AI Cooperation Organization, promoting Chinese technology as an alternative to American influence in the industry. In my assessment, the balance of power is already ambiguous: the US retains leadership in fundamental research, development of advanced models, and investment volumes, but China is rapidly closing the gap—primarily in efficiency, applied technology use, and robotics.
An economy that doesn't add up
I see the key vulnerability of the American approach in the gap between infrastructure investments and real returns. Since 2023, US tech giants have poured hundreds of billions of dollars into AI, but monetization remains minimal. The numbers speak for themselves: Microsoft's capital expenditures reached $116 billion over the year, while direct revenue from its AI business is estimated at $40–45 billion. The investment-to-revenue ratio 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 situation for competitors is no better. Google showed negative operating cash flow for the first time in history, halted share buybacks, and grew its debt from $30 billion to $120 billion over fifteen months. Amazon went negative by $25 billion in cash flow 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. Meta's advertising, thanks to AI, adds only about 1% in conversions.
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 on par with 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 around $2.5 per million tokens, reached a level between GPT-5.5 and GPT-5.6 versions and is additionally optimized for Chinese Huawei 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 don't need expensive "genius." What's required is reliability, speed, and low price—and all of this China offers.
Here I formulate an important thesis. By releasing powerful models almost for free, China devalues American investments: each such release pushes the price bar down and deprives the US of any chance to ever recoup its giant investments.
Technology parity and the cost 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 the most complex tasks such as scientific reasoning, but models like Kimi K3, GLM-5.3, and Qwen-3.8 have come close to them while working noticeably faster and cheaper. I admit that at this pace, China could take the lead as early as 2027.
I consider Google's position especially telling: possessing data, chips, and engineers, the company still lost its leadership and is now compensating 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 pushes Washington toward political measures. I interpret the demand to choose sides as a sign of vulnerability: the US 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 might demonstratively leave the American coalition due to coercion, and restrictions will only accelerate China's development of its own chips and algorithms.
The final picture is a paradox. In investments, the US is far ahead, but that money isn't paying off and is piling up debt; in real usage, China already dominates the mass segment; in technology, parity has been reached 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 as an analyst: Washington's ultimatum is not strength, but an admission of its own failure. When the market cannot provide an answer to a competitor's challenge, politics becomes the last argument. But in a technological race where price and efficiency decide everything, administrative barriers are merely a postponement of the inevitable, not a solution to the problem.