6 月 24 日，一家叫 MirendilAI 的公司完成了 2 亿美元融资。
领投方是 a16z 和 Kleiner Perkins——硅谷最顶级的两家风投，NVIDIA 参投。目标是打造"可构建系统的系统"——能自主加速 AI 研发的自迭代 AI 系统。最夸张的是团队：20 个人，全部来自 Anthropic、xAI、DeepMind 和 OpenAI。被业内称为"AI 复仇者联盟"。
一天后，国内传来更夸张的消息：阿里前高管创立的昆仑行机器人，成立不到 90 天，连续完成三轮融资，累计达数十亿元，刷新了国内具身智能初创的融资速度纪录。
三浪叠加的中国 AI
中国 AI 行业正在出现一个罕见的"三浪叠加"现象。
第一浪是资本涌入：6 月前两周，中国 AI 领域公开融资超 45 亿元，头部企业估值达到 590 亿元。具身智能赛道尤其疯狂，几乎每周都有新的融资消息。第二浪是模型升级：国产大模型在多个维度已经逼近甚至超越海外模型，不再是"跟随者"。第三浪是数据基建：11.6 万个数据集正在构建 AI 的"石油储备"，高质量中文数据的价值正在被重新认识。
三条线索同频共振，形成了 2026 年夏天中国 AI 独特的热度。
"一级市场的 AI 项目，估值是按天涨的。晚一周签 TS，价格可能就涨了 20%。"
—— 某头部 VC 投资人
一级市场狂热，二级市场冷静
但和一级市场的狂热形成鲜明对比的，是二级市场的理性。
一个值得警惕的信号：博通发布 Jalapeño 芯片后股价仅涨 1.6%，而 AI 芯片销售预期 160 亿美元低于市场预期的 172 亿后，盘后直接跌超 10%。这说明什么？说明资本市场对 AI 概念股已经从"听故事"阶段进入"看数字"阶段了。
2023 年到 2024 年，只要公司名字里带"AI"，股价就能涨。2025 年开始要看收入。2026 年的今天，要看利润、要看增速、要看真实的市场份额。溜溜梅改名 LLM 暴涨 193% 是散户的闹剧，但机构投资者已经在用真金白银投票：没有真实业绩支撑的 AI 故事，讲不下去了。
这是一个健康的信号。泡沫不是坏事，但只有当泡沫退去，才知道谁在裸泳。一级市场的热钱会催生出伟大的公司，也会催生无数炮灰。历史告诉我们：每一次技术革命，90% 的公司会死掉，但剩下的 10% 会改变世界。
钱在往哪里流？
拆解这波融资热潮，资金的流向很有意思：
最火的赛道是具身智能——机器人。京东 618 人形机器人销量同比增长超 10 倍，华为发布 CloudRobo 具身智能平台，英伟达 Jetson AGX Thor 芯片本地算力提升 7.5 倍，特斯拉 Optimus V3 量产筹备完成。虚拟世界的 AI 打得差不多了，资本开始下注物理世界的 AI。
第二是 Agent 基础设施——MirendilAI 这类做"AI 造 AI"的公司，Cursor Origin 这类 Agent 原生工具，MCP/A2A 生态相关的项目。如果说大模型是"电力"，Agent 基础设施就是"电网"——电网的价值可能比发电厂还大。
第三是垂直应用——医疗、法律、教育、金融等行业的 AI 解决方案。当通用模型能力趋同，行业 know-how 和数据壁垒成为核心竞争力。这也是为什么很多传统行业的 AI 转型团队反而能拿到钱——他们懂行业，这是纯技术团队短期内补不上的课。
狂欢中需要保持清醒
对于创业者和从业者来说，这是最好的时代，也是最危险的时代。
钱好拿，意味着竞争也更激烈。90% 的 AI 创业公司会在未来两年死掉——不是因为技术不好，而是因为找不到真实的付费场景。To C 的产品在和免费的豆包、ChatGPT 竞争，To B 的产品在和客户自研的 Agent 竞争。找到一个"客户愿意掏钱、你比大厂做得好"的细分场景，比融多少钱都重要。
对于投资者来说，这是考验眼光的时候。热钱涌动的时候，谁都能投中几个明星项目，但只有在退潮时还能站着的基金，才是真正的赢家。
对于普通人来说，这是最好的入场时机。工具越来越便宜，资源越来越多，门槛越来越低。2010 年错过了移动互联网，2026 年别再错过 AI。不需要融资，不需要辞职，用 AI 重构你的工作流，就是对这场浪潮最好的参与。
浪潮汹涌，有人看到泡沫，有人看到冲浪的机会。你是哪一种？
明天见。
On June 24, a company called MirendilAI closed $200 million in funding.
The round was led by a16z and Kleiner Perkins -- Silicon Valley's two most prestigious VCs -- with NVIDIA participating. The goal is building "systems that build systems" -- self-iterating AI systems that can autonomously accelerate AI R&D. The most extraordinary detail: 20 people total, all from Anthropic, xAI, DeepMind, and OpenAI. The industry calls them the "AI Avengers."
A day later, even more stunning news came from China: Kunlunxing Robotics, founded by a former Alibaba executive, completed three consecutive funding rounds totaling billions of RMB less than 90 days after founding, setting a funding speed record for Chinese embodied AI startups.
Three Waves Converge in Chinese AI
A rare "three-wave convergence" phenomenon is appearing in Chinese AI.
The first wave is capital flooding in: in the first two weeks of June, public funding in Chinese AI exceeded 4.5 billion RMB; top company valuations reached 59 billion RMB. The embodied AI track is especially wild, with new funding news almost every week. The second wave is model upgrades: domestic LLMs have approached or even surpassed overseas models across multiple dimensions; they're no longer "followers." The third wave is data infrastructure: 116,000 datasets are building AI's "oil reserves," and the value of high-quality Chinese data is being re-recognized.
These three threads resonate at the same frequency, forming the unique heat of Chinese AI in summer 2026.
"In the primary market, AI project valuations rise by the day. Sign the TS a week late and the price might be up 20%."
-- A top VC investor
Primary Market Frenzy, Secondary Market Sobriety
But in sharp contrast to primary market frenzy is secondary market rationality.
A signal worth heeding: after Broadcom released the Jalapeño chip, its stock rose only 1.6%; after AI chip sales guidance of $16 billion came in below the $17.2 billion market expectation, it dropped over 10% after hours. What does this mean? Capital markets have moved from "listening to stories" to "looking at numbers" for AI concept stocks.
From 2023 to 2024, if a company had "AI" in its name the stock would rise. Starting in 2025, they want revenue. Today in 2026, they want profits, growth, real market share. 溜溜梅 renaming to LLM and surging 193% is retail investor farce, but institutional investors are voting with real money: AI stories without real performance backing can't keep selling.
This is a healthy signal. Bubbles aren't bad, but only when the tide recedes do you know who's been swimming naked. Primary market hot money will spawn great companies and countless cannon fodder. History tells us: in every technology revolution, 90% of companies die, but the remaining 10% change the world.
Where Is the Money Flowing?
Breaking down this funding boom, capital flows are interesting:
The hottest track is embodied AI -- robotics. JD 618 humanoid robot sales grew over 10x YoY; Huawei released the CloudRobo embodied intelligence platform; Nvidia Jetson AGX Thor chip boosted local compute 7.5x; Tesla Optimus V3 mass production prep is complete. AI in the virtual world has been fought over pretty thoroughly; capital is starting to bet on AI in the physical world.
Second is Agent infrastructure -- companies like MirendilAI doing "AI building AI," Agent-native tools like Cursor Origin, MCP/A2A ecosystem-related projects. If LLMs are "electricity," Agent infrastructure is the "power grid" -- the grid might be more valuable than the power plants.
Third is vertical applications -- AI solutions in healthcare, law, education, finance, and other industries. When general model capabilities converge, industry know-how and data barriers become core competitiveness. That's also why many traditional industry AI transformation teams can actually raise money -- they understand the industry, a lesson pure tech teams can't make up short-term.
Stay Awake Amid the Frenzy
For founders and practitioners, this is the best of times and the most dangerous of times.
Easy money means fiercer competition too. 90% of AI startups will die in the next two years -- not because their technology is bad, but because they can't find real paying scenarios. To-C products compete with free Doubao and ChatGPT; To-B products compete with customers' self-built Agents. Finding a niche where "customers are willing to pay and you do it better than big tech" is more important than any amount of funding.
For investors, this is when judgment is tested. When hot money surges, anyone can invest in a few star projects, but only funds still standing when the tide recedes are true winners.
For ordinary people, this is the best entry moment. Tools are getting cheaper, resources more abundant, barriers lower. If you missed mobile internet in 2010, don't miss AI in 2026. No funding needed, no quitting required; using AI to restructure your workflow is the best participation in this wave.
The tide surges; some see bubbles, some see surfing opportunities. Which are you?
See you tomorrow.
Sources · 信源 Sources
本文基于 Dawn Vision 认知引擎处理的 9 个源信号自动生成，经编辑部人工审核。素材来源包括：MirendilAI 融资公告、昆仑行融资报道、中国 AI 融资数据、博通股价表现、具身智能行业动态。
Auto-generated by Dawn Vision's cognitive engine from 9 source signals, editorially reviewed. Sources include: MirendilAI funding announcement, Kunlunxing funding coverage, Chinese AI funding data, Broadcom stock performance, embodied AI industry dynamics.
