在七月二十四日，超過二十家以 Nvidia, Microsoft, Meta, Hugging Face 和 Andreessen Horowitz 為首的公司，向華盛頓提交了一封長達一頁半的聯名信，警告監管機構不應對開源 AI 模型實施「過早」的限制。值得注意的是，致力於開發封閉專有模型的 Anthropic 與 OpenAI 並未參與連署，這反映出該產業對於中國推出的 DeepSeek, Alibaba Qwen, Moonshot Kimi 及 Zhipu 等免費開源模型，正以超乎預期的速度縮小與美國頂尖實驗室差距的競爭焦慮。（關鍵數字：24, 1.5）
中國正將其熟悉的工業策略應用於 AI 領域：透過積極建設與降低成本，試圖建立全球市佔率。牛津經濟研究院 (Oxford Economics) 上週的報告指出，這波攻勢的競爭影響力可能超越中國先前在太陽能板和電動車等領域的產能過剩，因為 AI 模型能以近乎零的邊際成本輕易跨越國界。目前，美國在模型層面的領先優勢正迅速縮小，儘管在先進運算方面仍具優勢，卻受限於國內電力短缺及資料中心建設的延宕。
報告總結，全球正走向兩極化的發展趨勢：美國主導建立在卓越運算能力上的 AI，而中國則透過價格優勢與無所不在的散佈進行競爭。作為回應，美國企業正採取「以毒攻毒」的策略，就在聯名信發布後數日內，Nvidia 與 Meta 皆發布了自家的開源模型，這等同默認美國的技術護城河已難以守成，並認為若中國打算免費釋出技術，部分美國企業將不得不加入這場開源競爭。
On July 24, over two dozen firms, including Nvidia, Microsoft, Meta, Hugging Face, and Andreessen Horowitz, submitted a 1.5-page letter to Washington regulators, warning against "premature" restrictions on open-weight AI models. Conspicuously, Anthropic and OpenAI, labs developing proprietary models, did not sign. This letter highlights industry anxiety as free Chinese open models like DeepSeek, Alibaba Qwen, Moonshot Kimi, and Zhipu close the performance gap with US frontier labs much faster than anticipated.
China is deploying a familiar industrial strategy to AI: building aggressively and lowering costs to capture global market share. An Oxford Economics report highlighted that this could have a greater competitive impact than China's previous overcapacity in solar panels and electric vehicles, as AI models cross borders digitally at near-zero marginal cost. While the US lead in the model layer is shrinking, its strength in advanced computing faces domestic constraints from electricity shortages and data center building delays.
The Oxford Economics report described a bifurcating global order: US AI driven by superior compute versus Chinese AI competing on price and ubiquitous distribution. In response, the US is adopting a similar strategy. Within days of the letter, both Nvidia and Meta released their own open-weight models, a tacit admission that the US protective moat is failing. With China distributing the technology for free, some US companies feel compelled to adopt the same open approach.