<p>在 2026 年，人工智能领域的竞争将首次不再仅仅是模型能力的较量。它将演变为深层范式的冲突：美国与中国、闭源与开源、超级资本与极致效率、人工智能帝国与人工智能工业化。如果你每天关注这个行业，你会注意到一个越来越明显的变化：人工智能世界正在分裂为两条截然不同的发展路径。一边是美国。OpenAI、Anthropic 和 xAI 正在以巨额资金、全球超级数据中心和 GPU 构建一个前所未有的人工智能基础设施帝国。另一边是中国。DeepSeek、智谱 AI、Kimi 和 MiniMax 正在以不同的方式突破：更低的成本、更高的工程效率、更强的开源生态系统和更快的本地化适应。2026 年 5 月 8 日，《信息》的一份报告将这一 “中式路线” 的核心参与者推向了前台，使世界意识到了一种根本性的变化。为什么 DeepSeek 的融资震动了全球人工智能社区？报道称，DeepSeek 正在计划其首轮外部融资，金额高达 500 亿元人民币（约合 73.5 亿美元）。更值得注意的是，其创始人梁文峰可能个人投资 200 亿元，占总融资的 40%。这个一直以来不寻常的公司，拒绝寻求融资或急于商业化，突然走上了资本投资的道路。这本身就是一个强烈的信号：第一次，中国产生了一家在基础建模上真正具有全球影响力的明星公司。但比 500 亿元的估值更重要的是 DeepSeek 独特的崛起。在过去一年中，它几乎无视了硅谷人工智能公司的所有增长逻辑。回顾 OpenAI 的路径：巨额融资→超级 GPU 集群→超级数据中心→微软支持→全球资本驱动。而 DeepSeek 的路径则是：长期无融资→极低的训练成本→工程优化优先→大规模开源→增强与国内芯片的兼容性。它并不像传统的人工智能初创公司；更像是一个 “技术效率怪兽”。根据多家媒体的报道，DeepSeek 的最新估值已达到 450 亿至 500 亿美元，其接触的资本包括国家人工智能基金、国有资产和半导体行业资本。这表明 DeepSeek 正在从一个纯粹的 “技术实验室” 转变为 “中国国家人工智能团队的核心成员”。真正值得关注的不是估值，而是 “方法的分歧”。许多人仍然将人工智能竞争理解为 “谁的模型更聪明”。但现在，一个越来越明确的事实是，全球人工智能领域出现了根本性的方式分歧。我们先来看美国的方法。其核心可以总结为：超级资本 + 人工智能帝国。美国的人工智能公司正进入一个极其夸张的发展阶段，越来越像 “数字基础设施国家”，而不是 “科技公司”。OpenAI 的估值为 8520 亿美元，最近筹集了 1220 亿美元，得到了软银、微软和 NVIDIA 的支持，正在推动其总额达 5000 亿美元的 “星际门” 超级人工智能基础设施计划。估值接近 1 万亿美元的 Anthropic 正在亚马逊和谷歌的支持下，积极扩展其 Claude Code、计算机使用和企业代理生态系统，并开始与谷歌 TPU 和 AWS 云进行整合。估值为 2000 亿美元的 xAI 正在利用 X（Twitter）的实时数据、Grok Connectors 和 Agent 生态系统发展成为一个人工智能操作系统。你会发现，美国方法的核心特征非常一致：1. 超级融资：人工智能已成为一个纯粹的资本密集型行业。2. 超级计算能力：核心竞争围绕 GPU、数据中心、能源和云基础设施展开。3. 闭源生态系统：OpenAI 和 Anthropic 越来越强调专有模型、企业壁垒和平台控制。4. 人工智能帝国：他们的目标不再是 “训练一个模型”，而是构建一个完整的人工智能基础设施。然而，在大洋彼岸，正在进行一种截然不同的转型。中国路线：人工智能工业化和终极效率。中国的人工智能公司走上了一条不同的道路。这种方法最显著的特征是 “以效率为驱动”，而不是 “以资本为驱动”。我们可以总结 “中国路线” 的核心特征如下：1. 更加重视开源：DeepSeek 的全球影响力在很大程度上源于其成功的开源生态系统。2. 更加重视成本：中国的模型公司并不是 “烧更多的 GPU”，而是 “以显著更低的成本实现近乎完美的结果”。3. 更加重视工程优化：这包括 MoE（混合专家）架构、推理优化、调度优化、模型压缩等。4. 更加重视本地适应：尤其是与华为昇腾等国内芯片的深度合作，已成为中国人工智能路线图中的关键变量。然而，需要注意的是，中国的人工智能并不是单一的。DeepSeek 和智谱 AI 实际上已经开始代表两个不同的方向。许多人容易混淆大型中国模型公司，但实际上：DeepSeek 更像是一个理想主义的混合体：它有点像 “早期 OpenAI” 与 “Linux” 和 “Android” 的结合。它代表了技术革命、开源扩散、极致成本优化和开发者生态系统。它的目标是成为全球人工智能的基础模型能力层。而智谱 AI 则更像是一个务实的平台参与者：它越来越像 “中文版的 Anthropic”。它更加重视企业代理、推理能力、国内生态系统和企业级平台。特别是其 GLM、AutoGLM 和 Agent 系统明确指向 “企业人工智能操作系统化”。这两者的共存构成了中国人工智能的全新格局。全球人工智能力量地图已经改变；下一个阶段是 “基础设施战争”。这至关重要。在过去，大多数中国互联网公司都复制了美国模式。然而，在人工智能时代，第一次出现了 “不同技术路线” 可以并行发展的局面。美国路线：超级资本、超级 GPU、闭源生态系统、人工智能帝国。中国路线：开源、极致工程优化、本地化、低成本、人工智能工业化。最终谁会胜出？现在没有人知道答案。但有一点越来越明确：下一阶段的人工智能竞争不再仅仅是模型能力的竞争，而是生态系统、能源、芯片和产业组织能力的全面竞争。DeepSeek 的估值飙升至 500 亿美元，可能正是宣告新纪元真正开始的号角。</p>

深度求索

<p>DeepSeek 最近的估值为 500 亿元人民币（约合 73.5 亿美元），标志着人工智能领域的重大转变，突显了美国和中国在这一领域的不同策略。美国公司如 OpenAI 和 Anthropic 专注于大规模融资和封闭源代码模型，而中国公司如 DeepSeek 则优先考虑效率、开源生态系统和更低的成本。DeepSeek 的创始人计划投资 200 亿元人民币，表明其从技术实验室向中国国家人工智能战略的关键参与者转变。这一演变强调了全球人工智能竞争的根本变化，突显了不同的发展路径</p>

<p>In 2026, for the first time, competition in the AI ​​field will no longer be merely a contest of model capabilities. It will evolve into a clash of deep paradigms: US vs. China, closed-source vs. open-source, super-capital vs. extreme efficiency, AI empire vs. AI industrialization. If you follow this industry daily, you'll notice an increasingly obvious change: The AI ​​world is splitting into two drastically different development paths. One is the US. OpenAI, Anthropic, and xAI are building an unprecedented AI infrastructure empire with massive funding, super data centers, and GPUs worldwide. On the other side is China. DeepSeek, Zhipu AI, Kimi, and MiniMax are breaking through in a different way: lower costs, higher engineering efficiency, a stronger open-source ecosystem, and faster localization adaptation. On May 8, 2026, a report by The Information brought the core players of this &#34;Chinese route&#34; to the forefront, making the world aware of a fundamental change. Why did DeepSeek's funding shake the global AI community? Reports indicate that DeepSeek is planning its first round of external funding, a staggering 50 billion yuan (approximately US$7.35 billion). Even more noteworthy is that its founder, Liang Wenfeng, may personally invest 20 billion yuan, accounting for 40% of the total funding. This company, which has always been unconventional, refusing to seek funding or rush into commercialization, has suddenly embarked on the path of capital investment. This in itself is a strong signal: for the first time, China has produced a truly globally influential star company in basic modeling. But more important than the 50 billion yuan valuation is DeepSeek's unique rise. Over the past year, it has almost defied all the growth logic of Silicon Valley AI companies. Looking back at OpenAI's path: massive funding → super GPU clusters → super data centers → Microsoft support → global capital drive. DeepSeek's path, on the other hand, is: no funding for a long time → extremely low training costs → engineering optimization first → large-scale open source → strengthening compatibility with domestic chips. It doesn't resemble a traditional AI startup; it's more like a &#34;technological efficiency monster.&#34; According to multiple media reports, DeepSeek's latest valuation has reached $45 billion to $50 billion, and the capital it has been in contact with includes the National AI Fund, state-owned assets, and semiconductor industry capital. This indicates that DeepSeek is transforming from a purely &#34;technology lab&#34; into a &#34;core member of China's national AI team.&#34; What truly deserves attention is not valuation, but rather the &#34;divergence of approaches.&#34; Many still understand AI competition as &#34;whose model is smarter.&#34; But now, an increasingly clear fact is that a fundamental divergence in approaches has emerged globally in AI. Let's first look at the US approach. Its core can be summarized as: super capital + AI empire. American AI companies are entering an extremely exaggerated stage of development, increasingly resembling &#34;digital infrastructure nations&#34; rather than &#34;tech companies.&#34; OpenAI, valued at $852 billion and recently raising $122 billion, backed by SoftBank, Microsoft, and NVIDIA, is pushing its &#34;Stargate&#34; super AI infrastructure plan, totaling $500 billion. Anthropic, with its valuation approaching $1 trillion, is aggressively expanding its Claude Code, Computer Use, and enterprise agent ecosystem with the support of Amazon and Google, and is beginning to integrate with Google TPU and AWS cloud. xAI, valued at $200 billion, is leveraging X (Twitter)'s real-time data, Grok Connectors, and Agent ecosystem to develop into an AI operating system. You'll find that the core characteristics of the US approach are very consistent: 1. Super Funding: AI has become a purely capital-intensive industry. 2. Supercomputing Power: Core competition revolves around GPUs, data centers, power, and cloud infrastructure. 3. Closed-source ecosystem: OpenAI and Anthropic are increasingly emphasizing proprietary models, corporate barriers, and platform control. 4. AI empire: Their goal is no longer &#34;training a model,&#34; but rather building a complete AI infrastructure. However, across the ocean, a radically different transformation is underway. The Chinese route: AI industrialization and ultimate efficiency. Chinese AI companies are taking a different path. The most significant characteristic of this approach is that it is &#34;efficiency-driven&#34; rather than &#34;capital-driven.&#34; We can summarize the core characteristics of the &#34;Chinese route&#34; as follows: 1. Greater emphasis on open source: DeepSeek's global influence largely stems from its successful open-source ecosystem. 2. Greater emphasis on cost: Chinese model companies excel not at &#34;burning more GPUs,&#34; but at &#34;achieving near-perfect results at significantly lower costs.&#34; 3. Greater emphasis on engineering optimization: This includes MoE (Hybrid Expert) architecture, inference optimization, scheduling optimization, model compression, etc. 4. Greater emphasis on domestic adaptation: Especially deep cooperation with domestic chips such as Huawei Ascend has become a key variable in China's AI roadmap. However, it should be noted that China's AI is not monolithic. DeepSeek and Zhipu AI have actually begun to represent two different directions. Many people easily confuse large Chinese model companies, but in reality: DeepSeek is more like an idealistic hybrid: It's somewhat like a combination of &#34;early OpenAI&#34; with &#34;Linux&#34; and &#34;Android.&#34; It represents technological revolution, open-source diffusion, extreme cost optimization, and a developer ecosystem. Its goal is to become the foundational model capability layer for global AI. Zhipu AI, on the other hand, is more like a pragmatic platform player: It's increasingly resembling a &#34;Chinese version of Anthropic.&#34; It places greater emphasis on enterprise agents, inference capabilities, a domestic ecosystem, and enterprise-level platforms. In particular, its GLM, AutoGLM, and Agent systems clearly point towards &#34;enterprise AI operating systemization.&#34; The coexistence of these two aspects constitutes a completely new landscape for AI in China. The global AI power map has changed; the next stage is the &#34;infrastructure war.&#34; This is crucial. In the past, most Chinese internet companies copied the American model. However, in the AI ​​era, for the first time, we have seen &#34;two completely different Chinese routes&#34;: the DeepSeek route (global developers, open source, technology diffusion, extremely low cost) and the Zhipu route (enterprise market, agent platform, domestic ecosystem, enterprise closed loop). This means that China's AI is beginning to form a true ecological stratification. Turning our attention back to the global map, the current landscape, based on valuations, is very clear: The greatest significance of this review is not that &#34;China now also has large model companies,&#34; but rather that China is beginning to produce players who can truly enter the core of the global AI arena. Many are still debating whether GPT-5, Claude 4, or DeepSeek-R2 is stronger. But what truly matters in the future may not be &#34;whose model is the strongest in a single aspect,&#34; but rather &#34;who can build a complete AI ecosystem loop.&#34; The next phase of competition will evolve into a full-blown &#34;AI infrastructure war.&#34; The true determinants will be control over the following areas: chips, cloud, data centers, agents, APIs, developer ecosystem, operating systems, enterprise workflows… This is because top AI companies are increasingly resembling &#34;digital nations.&#34; They are building computing power, constructing ecosystems, cultivating developer systems, laying payment/API networks, and weaving agent systems—essentially, they are vying for the right to define and dominate the next generation of digital infrastructure. In conclusion, China's real opportunity for AI may have just begun. For the past 20 years, the Chinese internet has been catching up with the United States. But in the AI ​​era, for the first time, a situation has emerged where &#34;different technological routes&#34; can develop in parallel. The US Route: Super capital, super GPUs, closed-source ecosystem, AI empire. The Chinese Route: Open source, extreme engineering optimization, localization, low cost, AI industrialization. Who will ultimately win? Nobody knows the answer now. But one thing is becoming increasingly clear: The next stage of AI competition is no longer just a competition of model capabilities, but a comprehensive competition of ecosystem, energy, chips, and industrial organization capabilities. DeepSeek's surge to a $50 billion valuation may well be the clarion call that truly heralds the beginning of a new era.</p>
<div class="lb-trans"><p>在 2026 年，人工智能领域的竞争将首次不再仅仅是模型能力的较量。它将演变为深层范式的冲突：美国与中国、闭源与开源、超级资本与极致效率、人工智能帝国与人工智能工业化。如果你每天关注这个行业，你会注意到一个越来越明显的变化：人工智能世界正在分裂为两条截然不同的发展路径。一边是美国。OpenAI、Anthropic 和 xAI 正在以巨额资金、全球超级数据中心和 GPU 构建一个前所未有的人工智能基础设施帝国。另一边是中国。DeepSeek、智谱 AI、Kimi 和 MiniMax 正在以不同的方式突破：更低的成本、更高的工程效率、更强的开源生态系统和更快的本地化适应。2026 年 5 月 8 日，《信息》的一份报告将这一 “中式路线” 的核心参与者推向了前台，使世界意识到了一种根本性的变化。为什么 DeepSeek 的融资震动了全球人工智能社区？报道称，DeepSeek 正在计划其首轮外部融资，金额高达 500 亿元人民币（约合 73.5 亿美元）。更值得注意的是，其创始人梁文峰可能个人投资 200 亿元，占总融资的 40%。这个一直以来不寻常的公司，拒绝寻求融资或急于商业化，突然走上了资本投资的道路。这本身就是一个强烈的信号：第一次，中国产生了一家在基础建模上真正具有全球影响力的明星公司。但比 500 亿元的估值更重要的是 DeepSeek 独特的崛起。在过去一年中，它几乎无视了硅谷人工智能公司的所有增长逻辑。回顾 OpenAI 的路径：巨额融资→超级 GPU 集群→超级数据中心→微软支持→全球资本驱动。而 DeepSeek 的路径则是：长期无融资→极低的训练成本→工程优化优先→大规模开源→增强与国内芯片的兼容性。它并不像传统的人工智能初创公司；更像是一个 “技术效率怪兽”。根据多家媒体的报道，DeepSeek 的最新估值已达到 450 亿至 500 亿美元，其接触的资本包括国家人工智能基金、国有资产和半导体行业资本。这表明 DeepSeek 正在从一个纯粹的 “技术实验室” 转变为 “中国国家人工智能团队的核心成员”。真正值得关注的不是估值，而是 “方法的分歧”。许多人仍然将人工智能竞争理解为 “谁的模型更聪明”。但现在，一个越来越明确的事实是，全球人工智能领域出现了根本性的方式分歧。我们先来看美国的方法。其核心可以总结为：超级资本 + 人工智能帝国。美国的人工智能公司正进入一个极其夸张的发展阶段，越来越像 “数字基础设施国家”，而不是 “科技公司”。OpenAI 的估值为 8520 亿美元，最近筹集了 1220 亿美元，得到了软银、微软和 NVIDIA 的支持，正在推动其总额达 5000 亿美元的 “星际门” 超级人工智能基础设施计划。估值接近 1 万亿美元的 Anthropic 正在亚马逊和谷歌的支持下，积极扩展其 Claude Code、计算机使用和企业代理生态系统，并开始与谷歌 TPU 和 AWS 云进行整合。估值为 2000 亿美元的 xAI 正在利用 X（Twitter）的实时数据、Grok Connectors 和 Agent 生态系统发展成为一个人工智能操作系统。你会发现，美国方法的核心特征非常一致：1. 超级融资：人工智能已成为一个纯粹的资本密集型行业。2. 超级计算能力：核心竞争围绕 GPU、数据中心、能源和云基础设施展开。3. 闭源生态系统：OpenAI 和 Anthropic 越来越强调专有模型、企业壁垒和平台控制。4. 人工智能帝国：他们的目标不再是 “训练一个模型”，而是构建一个完整的人工智能基础设施。然而，在大洋彼岸，正在进行一种截然不同的转型。中国路线：人工智能工业化和终极效率。中国的人工智能公司走上了一条不同的道路。这种方法最显著的特征是 “以效率为驱动”，而不是 “以资本为驱动”。我们可以总结 “中国路线” 的核心特征如下：1. 更加重视开源：DeepSeek 的全球影响力在很大程度上源于其成功的开源生态系统。2. 更加重视成本：中国的模型公司并不是 “烧更多的 GPU”，而是 “以显著更低的成本实现近乎完美的结果”。3. 更加重视工程优化：这包括 MoE（混合专家）架构、推理优化、调度优化、模型压缩等。4. 更加重视本地适应：尤其是与华为昇腾等国内芯片的深度合作，已成为中国人工智能路线图中的关键变量。然而，需要注意的是，中国的人工智能并不是单一的。DeepSeek 和智谱 AI 实际上已经开始代表两个不同的方向。许多人容易混淆大型中国模型公司，但实际上：DeepSeek 更像是一个理想主义的混合体：它有点像 “早期 OpenAI” 与 “Linux” 和 “Android” 的结合。它代表了技术革命、开源扩散、极致成本优化和开发者生态系统。它的目标是成为全球人工智能的基础模型能力层。而智谱 AI 则更像是一个务实的平台参与者：它越来越像 “中文版的 Anthropic”。它更加重视企业代理、推理能力、国内生态系统和企业级平台。特别是其 GLM、AutoGLM 和 Agent 系统明确指向 “企业人工智能操作系统化”。这两者的共存构成了中国人工智能的全新格局。全球人工智能力量地图已经改变；下一个阶段是 “基础设施战争”。这至关重要。在过去，大多数中国互联网公司都复制了美国模式。然而，在人工智能时代，第一次出现了 “不同技术路线” 可以并行发展的局面。美国路线：超级资本、超级 GPU、闭源生态系统、人工智能帝国。中国路线：开源、极致工程优化、本地化、低成本、人工智能工业化。最终谁会胜出？现在没有人知道答案。但有一点越来越明确：下一阶段的人工智能竞争不再仅仅是模型能力的竞争，而是生态系统、能源、芯片和产业组织能力的全面竞争。DeepSeek 的估值飙升至 500 亿美元，可能正是宣告新纪元真正开始的号角。</p>
</div>

深度求索的 500 亿评级：中国的人工智能正在开始发生变化