[By GuanchaNet Columnist MindObservatory]
"We have observed that attempts to smuggle chips have become increasingly sophisticated, including through shell companies and disguised shipments (such as fake pregnancy bellies or lobster packaging). BIS should invest in advanced detection technologies, such as AI-driven supply chain analysis, to identify and intercept such activities... AI is reshaping the global economic and geopolitical landscape. America's lead in computational infrastructure and AI model development is the cornerstone of its national security and economic competitiveness. However, foreign entities—including potential adversaries—are seeking to acquire these technologies through both legitimate and illegitimate means."
On May 3, 2025, NVIDIA issued a fierce critique of the American AI startup Anthropic. Prior to this, Anthropic had publicly supported the U.S. government's "AI Dissemination Export Control Framework," advocating for stronger export restrictions on AI chips to China. The above passage is from Anthropic's public statement.
NVIDIA's response was merciless: "China has half of the world's AI researchers and top experts at every layer of the AI technology stack. America cannot win in the AI field by manipulating regulatory bodies. American companies should focus on innovation and facing challenges rather than fabricating absurd rumors."
This dispute once again exposed Anthropic's naked stance in the Sino-American AI competition and pushed this company founded by former OpenAI researchers to the forefront of public attention. To understand Anthropic's motivations, we need to trace its growth trajectory and analyze the ideas of its founders—especially Dario Amodei—to glimpse the future direction of Sino-American AI competition behind his arrogance and prejudice.
They shared a common belief: the potential of artificial intelligence is enormous, but if not constrained, it may bring catastrophic consequences. Anthropic registered as a Public Benefit Corporation, committing to developing "responsible" AI for the long-term benefit of humanity.
Dario, this prodigy who studied math courses at UC Berkeley in his teenage years, likes to sketch AI intelligence growth curves on the company whiteboard and is also keen on inspiring creativity through "vision exploration." Daniela brings a humanistic perspective with her literary background, attempting to build a low-ego, high-trust team culture. This team deeply influenced by effective altruism philosophy not only develops AI technology but also considers the profound impact of AI on human society as their mission.
In 2023, Anthropic launched the Claude series of large language models, directly challenging OpenAI's ChatGPT. Claude's uniqueness lies in its "Constitutional AI" framework, ensuring output safety and value consistency through built-in ethical guidelines. This innovation allowed Anthropic to quickly stand out.
In the early stages, Anthropic secured $5.8 billion in funding, including $500 million from FTX. Although FTX's collapse was lamentable, Anthropic's financial chain was not affected—Amazon invested $12.5 billion (planned total of $40 billion), and Google added another $5 billion (promising to add another $15 billion). These resources propelled Anthropic's technological leaps.
In 2024, the Claude 3 series (Haiku, Sonnet, Opus) surpassed competitors in multiple benchmark tests; in June, the Claude 3.5 Sonnet was released, adding real-time code preview functionality, further solidifying its market position. Anthropic not only competes technologically with giants but also ensures commercial interests do not overshadow public welfare through its unique corporate governance structure—the Long-Term Benefit Trust. By November 2025, this trust will control the majority of board seats, demonstrating Anthropic's commitment to its mission.
However, Anthropic's ambitions extend beyond technical breakthroughs. Its founders view AI as a geopolitical battleground, involving a clash between democratic and authoritarian values. This belief was vividly demonstrated when Anthropic supported U.S. chip export controls.
In his article "On DeepSeek and Export Controls," Dario wrote: "Export controls are not intended to avoid competition but to ensure that democratic nations maintain leadership in AI development to defend values of freedom and transparency."
He specifically mentioned DeepSeek Company in China, whose R1 and V3 models challenge Western AI with low cost and high performance. Despite DeepSeek's slightly inferior performance in code generation and conversational subtlety compared to Western models, its open-source model and price advantage made Dario uneasy.
He pointed out: "China's ability to expand AI on low-cost infrastructure is awe-inspiring, but export controls effectively restrict their access to cutting-edge chips, preserving key advantages for the U.S." In Dario's view, restricting China's access to advanced chips and model weights not only slows down their technological progress but also prevents "AI being used to generate information about biological weapons or undermine democratic values."
This stance infuriated NVIDIA. As the global leader in AI chips, NVIDIA relies on a vast market including China. Anthropic's description of chip smuggling as being hidden in "fake pregnancy bellies or lobsters" was clearly veiled, and NVIDIA criticized it as "absurd."
NVIDIA believed that China's AI talent (accounting for 47% globally) and technical capabilities mean that regulation cannot completely stifle its development, and excessive restrictions may weaken the global competitiveness of American enterprises.
This dispute also revealed divisions within the U.S. AI industry chain: chip suppliers like NVIDIA value market openness, while model developers like Anthropic worry more about the risks of technology diffusion. It is worth noting that Anthropic's deep cooperation with Amazon makes it rely more on Amazon's Trainium chips than NVIDIA's GPUs, which may give it more confidence in facing NVIDIA's criticism.
Why is Anthropic so determined to support export controls? From a strategic perspective, this young AI company faces immense pressure from OpenAI and Google. If Chinese companies like DeepSeek sweep the market with low-cost models, Anthropic's survival space will be compressed. By supporting restrictions, Anthropic can delay the pace of Chinese competitors and buy itself time to develop.
However, strategic considerations are just the tip of the iceberg. Dario's articles reveal deeper ideological drivers. He frames the Sino-American AI competition as a "democratic AI" versus "authoritarian AI" showdown, warning that Chinese models may generate "illegal or harmful content" due to different ethical standards. He writes: "If we lose our lead, AI could be used to amplify control rather than freedom." This belief aligns with Anthropic's self-perception of focusing on ethics and public interest, and may also be a strategy to seek policy support after the Trump administration took office to enhance its influence in Washington.
Currently, the Sino-American AI competition presents a complex picture. The U.S. currently holds an advantage—its enterprises dominate advanced chips, massive data centers, and cutting-edge models—but China is catching up. Models like DeepSeek R1, Qwen2.5, Doubao-1.5-Pro, GLM-4-Plus, and Yi-1.5 rank high on open-source rankings, and Huawei's Ascend chips also show initial self-reliance capabilities. Even if export controls prove effective in the short term, they will accelerate China's independent research in the long run.
NVIDIA expects to lose $5.5 billion in H20 chip sales due to export controls, which is the economic cost of these controls for American enterprises. Dario also realized this, stating: "Controls must be precise to avoid provoking China's determination to circumvent them." But he still insists that export controls are necessary to prevent AI misuse, particularly in risks like bio-weapons or mass surveillance.
Anthropic's attempt to frame the AI competition as a "democratic versus authoritarian" showdown actually masks its underlying strategic anxiety and technological protectionism. This stance lacks factual basis and underestimates the resilience of China's AI ecosystem and the irreversible trend of global technological cooperation.
Anthropic's technical assessment of Chinese AI companies like DeepSeek is equally biased. In his article, Dario admitted that DeepSeek V3 models perform "close to the most advanced models in the U.S." but tried to downplay its innovative significance by emphasizing its costs align with industry trends. This argument ignores China's breakthroughs in computational optimization and engineering efficiency. For example, DeepSeek achieved high performance at lower costs through hybrid expert models (MoE) and efficient key-value caching technology, with training costs being one-tenth of Claude 3.5 Sonnet. This efficiency advantage is not "industry normalcy" but the result of Chinese AI companies' long-term focus on algorithms and infrastructure.
Dario's accusation that China "relies on unofficial channels to obtain chips" is baseless, attempting to attribute China's technological progress to "smuggling" rather than self-innovation. China's self-sufficiency rate in computing power and the application ratio of domestic chips in AI training have significantly increased. Anthropic's rhetoric essentially conceals its concerns about the high costs and slow technological iteration of American AI enterprises.
Anthropic's supported export control strategy will ultimately backfire. Dario claims that controls are necessary to "ensure democratic nations stay ahead," but this unilateral mindset underestimates the global influence of China's AI ecosystem. China's open-source models like DeepSeek V3 not only rival Western models in performance but also gain favor in developing countries due to their low thresholds and high accessibility. In contrast, Anthropic's Claude models, due to their high subscription fees and closed ecosystems, struggle to gain widespread popularity in the global market.
Moreover, Anthropic framing the AI competition as a "democratic versus authoritarian" ideological showdown is a misleading bias. In 2024, a U.S. mainstream model was exposed for generating large amounts of racially discriminatory content due to insufficient filtering, sparking widespread controversy.
Most importantly, Anthropic framing the AI competition as a "democratic versus authoritarian" ideological showdown is a misleading bias. Dario warned that Chinese AI models like DeepSeek might generate "harmful content" due to open-source designs, yet ignored issues of bias and ethics in American AI models. Research shows that certain Western models may generate anti-human content during fine-tuning, including racist views, clearly showing that limitations in safety mechanisms are unrelated to the developer's ideology. Anthropic's "Constitutional AI," although emphasizing safety, is deeply influenced by Western values and has been criticized for lacking adaptability to non-Western cultures.
Dr. Leslie Teo, Senior Director of AI Products at Singapore, said when launching the large language model (LLM) SEA-LION for Southeast Asian languages: "Western large language models carry very obvious West Coast American biases—they are extremely 'woke'." In contrast, DeepSeek's multi-language support and low-cost APIs have gained popularity in Southeast Asia and Africa.
Dario's accusations of "authoritarian AI" is a stigmatization of non-Western technological paths, attempting to mask America's hegemonic intentions in global AI governance with ideological labels. China has always advocated the inclusiveness of AI development, proposing the "Global Initiative for Artificial Intelligence Governance" in 2023 and promoting technology sharing and multilateral cooperation through platforms like the United Nations and BRICS summits in 2024, gaining support from many developing countries. In contrast, Anthropic's supported export control policies will exacerbate the fragmentation of the global AI ecosystem, hinder knowledge sharing, and go against the original intention of technology benefiting humanity.
This article is an exclusive contribution from GuanchaNet. The views expressed in the article are solely those of the author and do not necessarily represent the platform's views. Unauthorized reproduction is prohibited, and legal action will be taken if violated. Follow GuanchaNet on WeChat (guanchacn) for daily interesting articles.
Original source: https://www.toutiao.com/article/7502233692631351842/
Disclaimer: The article represents the author's personal opinions and welcomes your feedback via the up/down vote buttons below.