October 11, 2025 — The artificial intelligence (AI) industry is undergoing a fierce clash over ecosystem choices. From ChatGPT's text generation to Sora 2's video revolution, the technological frontier keeps advancing, but competition between open and closed ecosystems has intensified. The closed ecosystem, dominated by giants such as OpenAI, emphasizes proprietary control yet has been troubled by the copyright and ethical storm around Sora 2; the open ecosystem, rooted in community collaboration, sees Chinese LLMs topping open‑source rankings and driving more democratic global AI innovation. Meanwhile, the US‑China tech conflict has escalated — the United States tightening chip export controls and China accelerating self‑reliance — further widening ecosystem divergence. This article is adapted from a The Verge podcast and analyzes these hotspots and the geopolitical and ethical contests shaping AI's future.
Closed Ecosystem: The Sora 2 Fallout Highlights Control's Double‑Edged Sword
The closed ecosystem centers on proprietary models from giants like OpenAI, Google and Microsoft, systems that rely on secret algorithms and massive datasets to monopolize high‑performance AI. Since ChatGPT's launch in 2022, monthly active users have exceeded 800 million (OpenAI, Developer Day 2025), and OpenAI's valuation reached $500 billion in 2025. Yet the cost crisis of the closed model is increasingly apparent: training ChatGPT‑4 cost $200 million, daily inference costs run $700,000, and Nvidia H100 GPUs have surged to prices around $40,000. Compute shortages not only raise operational burdens but also increase energy consumption; global AI data center carbon emissions account for roughly 20% of total emissions.
The release of Sora 2 further exposed the closed ecosystem's fragility. On October 4, 2025, OpenAI launched the Sora 2 video generator and an accompanying social app, claiming it could generate 120‑second 4K videos from text prompts with lifelike detail, potentially disrupting film and advertising. But within a week it was embroiled in copyright and ethical controversies. The Motion Picture Association (MPA) accused Sora 2 of allowing users by default to generate videos containing copyrighted characters from Disney, Netflix and others, leading to a proliferation of infringing content on social platforms. Hollywood studios and CAA agents protested collectively, calling it a theft of creative labor and demanding OpenAI strengthen filtering mechanisms immediately. More seriously, Sora 2 produced frequent outputs depicting violence and racist imagery: reports circulated that some outputs simulated realistic bullying scenarios, potentially aiding fraud and intimidation. NPR analyzed that such "deepfake" videos could become a new frontier for social media manipulation, and OpenAI was criticized for rushing the product to market.
The controversy stems from the closed ecosystem's "black box" nature: training data is opaque and users cannot easily trace the sources of infringements. Although OpenAI has pledged to improve copyright safeguards, it faces multiple lawsuits and a reported 15% user churn. This underscores the closed model's double‑edged sword: commercial barriers protect profits but amplify ethical risks, forcing giants to balance innovation against compliance.
Open Ecosystem: Chinese LLMs Top Rankings and Reshape the Global Landscape
The open ecosystem is driven by open‑source models and community collaboration; Meta's LLaMA and Mistral's Mixtral, among others, have been downloaded over 100 million times. Its appeal lies in low barriers to entry: training a medium‑sized model can cost only one‑tenth that of proprietary versions, and Hugging Face hosted more than 100,000 models in 2024, empowering startups and academic research. In 2025, the strong rise of Chinese LLMs injected new momentum into the open ecosystem.
According to Hugging Face and Design Arena rankings, Chinese models occupy the top 15 spots in open‑source rankings: Alibaba's Qwen3 series leads in multilingual capabilities; DeepSeek‑V3 (DeepSeek) rivals GPT‑4 on high‑performance inference; Zhipu AI's GLM‑4.5 and Moonshot AI's Kimi K2 follow closely. Tencent's HunyuanWorld and Baidu's Ernie also made significant contributions. A September Intuition Labs report showed Chinese open‑source LLMs accounted for 40% of global coverage. Vellum AI rankings confirmed Qwen3 placed in the top three on SOTA benchmarks, surpassing Meta LLaMA. These models excel at culturally contextualized content—such as Kling AI's Chinese‑language video generation—supporting localization across Africa and Southeast Asia.
China's open‑source dominance stems partly from national strategy: the 14th Five‑Year Plan in 2025 reportedly earmarked $50 billion to promote model open‑sourcing as a way to bypass chip controls. Community contributions are the largest globally; Interconnects.ai ranks 19 Chinese labs among the top contributors. Although there have been security concerns—five cases of open‑source models being maliciously adapted into forgery tools—transparency mechanisms have improved vetting efficiency and helped rebuild public trust.
US‑China Tech Conflict: Escalating Fractures and Geopolitical Risk
The US‑China tech conflict has pushed open and closed ecosystems into confrontation. On October 10, the Trump administration announced a 100% tariff on certain exports to China and tightened AI chip controls, effectively banning high‑end GPUs like Nvidia's H100 from Chinese sales. CNN reported that the move struck at the closed ecosystem's core: OpenAI relies on Nvidia compute, and Sora 2's training was delayed by three months. China responded by tightening chip import reviews and accelerating domestic alternatives such as Huawei's Ascend series, while pushing for greater open‑source self‑sufficiency.
The conflict affects ecosystems differently: the US views Chinese open‑source efforts as potential "technology infiltration," and a CSIS report warned that DeepSeek might circumvent export controls. The Washington Post described China's "full‑stack" AI policy—subsidizing compute and talent—as enabling its open‑source lead, while US tariff‑driven frictions have slowed domestic progress. Brookings predicts AI will dominate bilateral relations over the next five years, with closed ecosystems potentially becoming a US "digital fortress" and open ecosystems helping China build a "Global South" alliance. RAND analysis suggests Chinese industrial policy could accelerate AI advances by about 20%, intensifying ecosystem divergence.
Regulation and Ethics: Shared Tests Amid Conflict
The EU AI Act took effect in August 2025, requiring high‑risk AI systems to disclose training data, and violations can be fined up to 7% of annual turnover; the closed ecosystem was an early target—OpenAI was fined €20 million for Sora 2 GDPR violations. California's SB 1047 mandates certain open disclosures, balancing US and Chinese impacts. Ethically, both Sora 2's biases and open‑source misuse demand governance as public trust declined—surveys indicate up to a 60% drop in trust. EU AI literacy and training programs have been proposed as partial remedies.
Outlook: Convergence or Further Fragmentation?
The rivalry between open and closed ecosystems will shape AI's future. Closed ecosystems suit high‑performance, commercial scenarios such as financial AI and autonomous driving, but costs and regulatory pressure may push giants to open parts of their technology; Google has pledged to open‑source parts of Gemini in 2026. The open ecosystem, meanwhile, thrives in verticals and emerging markets—many localized AI applications in Africa and Southeast Asia rely on open‑source models. However, concentrated compute and energy challenges could consolidate incumbent advantages, and closed ecosystems might dominate up to 80% of the global AI market by 2030.
Experts suggest convergence may be the solution: giants provide compute while open communities contribute innovation. For example, Meta and Hugging Face launched a joint AI accelerator in 2025 to fund 50 startups. Whatever path is chosen, the AI ecosystem must balance innovation, fairness and responsibility to achieve inclusive benefits.
(Adapted from a The Verge podcast and combined with authoritative data from the European Commission, Hugging Face and others.)