{"slug": "goldman-sachs-just-picked-its-chinese-ai-winners-and-the-open-source-bet-is", "title": "Goldman Sachs Just Picked Its Chinese AI Winners — And the Open-Source Bet Is Massive", "summary": "Goldman Sachs initiated coverage on Chinese AI companies, naming Zhipu and DeepSeek among top picks, and projecting Chinese AI model revenue will grow 25x to $121 billion by 2030. Zhipu's founder defended open-source AI in an internal memo, positioning the company as a champion of open access amid potential government restrictions. The report highlights a distribution race between Chinese open-source models and U.S. proprietary models, with implications for global AI competition.", "body_md": "# Goldman Sachs Just Picked Its Chinese AI Winners — And the Open-Source Bet Is Massive\n\nGoldman Sachs initiated coverage on Chinese AI companies, naming Zhipu and DeepSeek among top picks. The bank projects Chinese AI model revenue growing 25x to $121B by 2030. Zhipu's founder simultaneously defended open-source AI in an internal memo.\n\n## The Three Models Goldman Is Betting On\n\nGoldman Sachs initiated coverage on Chinese AI companies this week, and the report tells you everything about where institutional money thinks AI is heading.\n\nThe investment bank named three preferred Chinese AI models. Only one is publicly traded: Zhipu, listed in Hong Kong, with a price target of HK$1,880 (about $240). Goldman rates Zhipu's open-sourced GLM-5.2 model as competitive with [Anthropic](/glossary/anthropic)'s Fable 5 on several key metrics.\n\nThe other two picks — DeepSeek and another unnamed private company — reflect Goldman's thesis that China's AI ecosystem is bifurcating: closed state-of-the-art models for domestic use, and efficient open-source variants flooding the global market to capture share.\n\n## The Numbers Behind the Bet\n\nGoldman's projection for Chinese AI model revenue is aggressive: from an estimated 35 billion yuan ($4.8 billion) in 2026 to 879 billion yuan ($121 billion) by 2030. That's a 25x increase in four years.\n\nIf you think that sounds optimistic, consider this: Goldman initiated coverage. They don't initiate coverage on companies they think will be irrelevant in four years. The report positions Chinese AI as the most significant emerging competitor to U.S. frontier models — not in five years, but right now.\n\n## Zhipu's Open-Source Gambit\n\nOn the same day Goldman dropped its report, Zhipu founder Tang Jie published an internal memo arguing that frontier AI should remain openly accessible rather than controlled by a small group of developers.\n\nThe timing is calculated. Both Washington and Beijing are considering restrictions on open-weight model releases, citing biosecurity and cybersecurity risks. Tang's memo positions Zhipu as the champion of open access — and Goldman's coverage gives that position financial backing.\n\nZhipu's strategy mirrors what we've seen from Meta with [Llama](/compare/llama-4-vs-deepseek-r1): release competitive open-source models, build an ecosystem around them, and turn distribution into a moat. When developers build on your models, switching costs go up. When your models run everywhere, everyone becomes your distribution channel.\n\n## The Global Open vs. Closed Divide\n\nThe China-U.S. AI competition is often framed as a technology race. Goldman's report frames it differently: as a distribution race.\n\nChinese labs like Zhipu and DeepSeek can't compete with OpenAI or Anthropic on the enterprise SaaS model — GTM in Western markets is too hard, trust barriers are too high. But they can compete on open-source distribution. Release a model that runs on consumer hardware, let developers build with it, and watch adoption compound.\n\nColibrì — the [inference](/glossary/inference) engine that runs GLM-5.2 on 25GB of RAM — is essentially a marketing asset for this strategy. It proves that Chinese frontier models can run anywhere, no API key required.\n\n## What This Means for U.S. AI Companies\n\nGoldman's report should wake up anyone who thinks the U.S. has a permanent lead in AI. The Chinese open-source strategy is smart:\n\n- Keep your best models closed for domestic strategic advantage\n- Release highly competitive open models to the global market\n- Build an international developer ecosystem that doesn't depend on U.S. infrastructure\n- Let the open-source community do your global distribution for free\n\nOpenAI and Anthropic are racing toward IPOs built on proprietary API businesses. If Chinese open-source models reach parity while staying free, the pricing power of proprietary models erodes. Not today, not this year — but Goldman's 25x revenue projection says the market believes it's happening.\n\n#### Q: Can I invest in Zhipu?\n\n#### A: Yes, Zhipu is listed on the Hong Kong Stock Exchange. Goldman initiated coverage with a buy rating and a price target of HK$1,880. As with any single-stock investment in an emerging technology sector, volatility should be expected.\n\n#### Q: How do GLM-5.2 and Anthropic's Fable 5 compare?\n\n#### A: Goldman rates them as competitive on several key benchmarks, though direct comparisons are complicated by different [training](/glossary/training) data and [evaluation](/glossary/evaluation) methodologies. GLM-5.2's strongest advantage is its open-weight release, which enables community [fine-tuning](/glossary/fine-tuning) and local deployment that Fable 5 doesn't support.\n\n#### Q: Does China's AI push threaten U.S. leadership?\n\n#### A: Not in the near term, but the strategic picture is shifting. The U.S. leads on frontier model capability and enterprise adoption. China's advantage is in cost efficiency and open-source distribution. If cost becomes the dominant factor in AI adoption — which history suggests it will — the advantage shifts toward the lower-cost producer.\n\nGet AI news in your inbox\n\nDaily digest of what matters in AI.\n\n## Key Terms Explained\n\n[Anthropic](/glossary/anthropic)\n\nAn AI safety company founded in 2021 by former OpenAI researchers, including Dario and Daniela Amodei.\n\n[Evaluation](/glossary/evaluation)\n\nThe process of measuring how well an AI model performs on its intended task.\n\n[Fine-Tuning](/glossary/fine-tuning)\n\nThe process of taking a pre-trained model and continuing to train it on a smaller, specific dataset to adapt it for a particular task or domain.\n\n[Inference](/glossary/inference)\n\nRunning a trained model to make predictions on new data.", "url": "https://wpnews.pro/news/goldman-sachs-just-picked-its-chinese-ai-winners-and-the-open-source-bet-is", "canonical_source": "https://www.machinebrief.com/news/goldman-sachs-chinese-ai-winners-zhipu-deepseek-open-source-2026", "published_at": "2026-07-12 13:06:59+00:00", "updated_at": "2026-07-12 13:19:45.789320+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-policy", "ai-startups", "ai-products"], "entities": ["Goldman Sachs", "Zhipu", "DeepSeek", "Tang Jie", "Anthropic", "Meta", "OpenAI", "GLM-5.2"], "alternates": {"html": "https://wpnews.pro/news/goldman-sachs-just-picked-its-chinese-ai-winners-and-the-open-source-bet-is", "markdown": "https://wpnews.pro/news/goldman-sachs-just-picked-its-chinese-ai-winners-and-the-open-source-bet-is.md", "text": "https://wpnews.pro/news/goldman-sachs-just-picked-its-chinese-ai-winners-and-the-open-source-bet-is.txt", "jsonld": "https://wpnews.pro/news/goldman-sachs-just-picked-its-chinese-ai-winners-and-the-open-source-bet-is.jsonld"}}