Anthropic三个月进账170亿美元，投资人为何反而恐慌
字幕摘录
| 时间 | 英文 | 中文 |
|---|---|---|
| 0:00 | Imagine making 17 billion dollars. | 想象一下赚了170亿美元 |
| 0:02 | Right. | 对 |
| 0:03 | You make it in just three months. | 你三个月后就到了 |
| 0:04 | That is a lot of money. | 这是一大笔钱。 |
| 0:06 | It is. | 没错 |
| 0:06 | But now imagine your investors are still panicking. | 现在想象一下你的投资者还在恐慌 |
| 0:10 | Like actually panicking. | 像是惊慌失措 |
| 0:12 | Exactly. | 没错 |
| 0:12 | So | 这么说 |
| 0:13 | welcome to the deep dive. | 欢迎来到深潜 |
展开字幕全文（498 条）
| 序号 | 英文 | 中文 |
|---|---|---|
| 1 | Imagine making 17 billion dollars. | 想象一下赚了170亿美元 |
| 2 | Right. | 对 |
| 3 | You make it in just three months. | 你三个月后就到了 |
| 4 | That is a lot of money. | 这是一大笔钱。 |
| 5 | It is. | 没错 |
| 6 | But now imagine your investors are still panicking. | 现在想象一下你的投资者还在恐慌 |
| 7 | Like actually panicking. | 像是惊慌失措 |
| 8 | Exactly. | 没错 |
| 9 | So | 这么说 |
| 10 | welcome to the deep dive. | 欢迎来到深潜 |
| 11 | Glad you're joining us today. | 很高兴你今天能来 |
| 12 | Today we are looking at some leaked | 今天我们正看着一些泄露出去的人 |
| 13 | financial numbers. | 财务数字。 |
| 14 | Very fresh numbers too. | 非常新鲜的数字。 |
| 15 | Right. | 对 |
| 16 | They come from a brand new article. | 他们来自一个全新的文章。 |
| 17 | It is from the | 这是从 |
| 18 | South China Morning Post. | 南华早报. |
| 19 | The date is November 4th, 2026. | 日期为2026年11月4日. |
| 20 | And we are looking really closely at a | 我们正密切注视着 |
| 21 | 叫做Anthropic的公司。 | |
| 22 | Yes, Anthropic. | 是的,安东尼 |
| 23 | You probably know them. | 你可能认识他们 |
| 24 | 他们让Claude AI很受欢迎 | |
| 25 | model. | 型号。 |
| 26 | People usually argue about which AI is the smartest. | 人们通常会争论哪个AI是最聪明的. |
| 27 | You know, they want to know which one | 你知道,他们想知道 哪个 |
| 28 | writes the best computer code. | 写最好的计算机代码。 |
| 29 | Or which one can pass a hard test. | 或何人能通过难验. |
| 30 | Right. | 对 |
| 31 | Like a bar exam for lawyers. | 就像律师的律师考试 |
| 32 | Yeah. | 对 |
| 33 | But we have a different mission today. | 但是我们今天的任务不同 |
| 34 | A completely different angle. | 一个完全不同的角度。 |
| 35 | Yeah. | 对 |
| 36 | We are looking | 我们在寻找 |
| 37 | strictly at the business of AI. | 严格来说是AI的生意 |
| 38 | We want to know how these massive systems actually make money. | 我们想知道这些庞大的系统是如何赚钱的。 |
| 39 | If they even do make money. | 如果他们真的赚钱的话 |
| 40 | Right. | 对 |
| 41 | And Anthropic is the perfect test case right now. | 而Anthropic是完美的测试案例 现在。 |
| 42 | Because of | 因为 |
| 43 | the big event. | 大型事件。 |
| 44 | Yes. | 对 |
| 45 | They are preparing for an IPO. | 他们正在筹备一个IPO。 |
| 46 | So an IPO. | 所以是一个IPO。 |
| 47 | Let's explain that really quickly. | 让我们解释一下,真的很快。 |
| 48 | Please do. | 请吧 |
| 49 | It stands for Initial Public Offering. | 它代表了初始公开报价。 |
| 50 | Right. | 对 |
| 51 | It means a company is selling shares to the | 意思是公司要卖股票 |
| 52 | public for the first time. | 首次公开. |
| 53 | So regular people can buy a piece of the company. | 这样,普通人就可以购买公司的一块. |
| 54 | Exactly. | 没错 |
| 55 | But going | 却去 |
| 56 | public changes everything. | 公众会改变一切 |
| 57 | You cannot hide your costs anymore. | 你不能再隐瞒自己的代价了 |
| 58 | You have to open the books. | 你必须打开书。 |
| 59 | Yeah. | 对 |
| 60 | You have to open your secret financial books to the whole world. | 你必须向全世界开放你的秘密财务书籍. |
| 61 | Investors demand to see the actual | 投资者要求查看实际 |
| 62 | math. | 数学 |
| 63 | And the math in this article is wild. | 这篇文章中的数学是疯狂的。 |
| 64 | It reveals a massive gap in the global market. | 它揭示了全球市场的巨大差距。 |
| 65 | A really brutal divide. | 一个非常残酷的鸿沟。 |
| 66 | 我们一边是Anthropic。 | |
| 67 | And their top Chinese rivals are | 他们的中国顶尖对手是 |
| 68 | on the other side. | 在另一边。 |
| 69 | The main competitors mentioned are Z.AI and Minimax. | 所提到的主要竞争对手是Z.AI和Minimax. |
| 70 | Right. | 对 |
| 71 | And Z.AI is also known | Z. AI也已经知道 |
| 72 | as Xipu AI. | 作为习近平AI. |
| 73 | Just to be clear on the names. | 只是要说清楚名字 |
| 74 | Yeah. | 对 |
| 75 | Now the technology between these companies is | 现在这些公司之间的技术是 |
| 76 | relatively close. | 相对近. |
| 77 | Right. | 对 |
| 78 | It is. | 没错 |
| 79 | The AI models can do similar tasks. | AI模型可以执行类似的任务. |
| 80 | They are very smart. | 他们非常聪明。 |
| 81 | But the | 不过 |
| 82 | business side tells a completely different story. | 商业方面讲述了完全不同的故事。 |
| 83 | The financial gap is staggering. | 财政缺口惊人。 |
| 84 | We need to look | 我们需要看看 |
| 85 | at the scoreboard first. | 在记分牌第一。 |
| 86 | The basic numbers. | 基本数字。 |
| 87 | Yeah. | 对 |
| 88 | The basic score. | 基本分数. |
| 89 | The revenue numbers for the first | 第一部分的收入数字 |
| 90 | half of 2026 are completely wild. | 2026年的一半是完全野性的。 |
| 91 | Let's hear them. | 让我们听听他们。 |
| 92 | Anthropic brought in 16.3 billion dollars. | 安东尼带来了163亿美元。 |
| 93 | Wow. | 珃 |
| 94 | It is. | 没错 |
| 95 | It is just for six months of operation. | 它只运作了六个月。 |
| 96 | Just six months. | 就六个月 |
| 97 | Massive. | 大规模。 |
| 98 | It is huge. | 这是巨大的。 |
| 99 | Massive. | 大规模。 |
| 100 | Now look at the Chinese rivals. | 现在看看中国的对手 |
| 101 | During that exact same time period. | 在同一时间段 |
| 102 | OK. Z.AI made just 141 million | 摆 Z.AI只赚了1.41亿 |
| 103 | dollars. | 钱 |
| 104 | Millions not billions. | 百万亿不亿. |
| 105 | Right. | 对 |
| 106 | 而Minimax赚了1.166亿美元. | |
| 107 | So the difference in | 所以区别在于 |
| 108 | raw sales is hard to even process. | 原始销售甚至难以处理。 |
| 109 | It really is. | 没错 |
| 110 | Anthropic的操作方式完全不同 | |
| 111 | planet. | 行星 地球 行星 |
| 112 | Yeah. | 对 |
| 113 | I mean I like to think of this like a giant global supermarket. | 我的意思是我喜欢把这当成一个巨大的全球超市。 |
| 114 | OK. Yeah. | 摆 对 |
| 115 | Anthropic is | 安东尼 |
| 116 | the giant supermarket. | 巨大的超市。 |
| 117 | And the rivals are like a small local shop on the corner. | 而竞争对手就像拐角处的小地方商店. |
| 118 | That makes sense. | 这有道理。 |
| 119 | But if their AI technology is so similar why is the money so different? | 但是,如果他们的AI技术如此相似,为什么资金如此不同? |
| 120 | Well raw sales do not tell | 好吧,原销售不告诉 |
| 121 | the whole story. | 整个故事。 |
| 122 | We have to look at the gross profit margin. | 我们必须看毛利润率。 |
| 123 | Let's break that term down for | 我们把这个词拆掉吧 |
| 124 | everyone. | 每个人。 |
| 125 | Sure. | 当然 |
| 126 | Gross profit margin is basically a percentage. | 总利润率基本上是一个百分比。 |
| 127 | OK. | 摆 |
| 128 | It is the percentage of money you | 这是你的钱的百分比 |
| 129 | keep after paying the direct cost of making your product. | 在支付你的产品直接成本后继续保留。 |
| 130 | So money in minus direct costs. | 钱减去直接成本 |
| 131 | Exactly. | 没错 |
| 132 | And for an AI company the main direct cost is servers. | 而对于一个AI公司来说,主要的直接成本是服务器. |
| 133 | The computers running the whole | 电脑运行整个 |
| 134 | thing. | 东西。 |
| 135 | Right. | 对 |
| 136 | It is the electricity and the raw processing power. | 是电力和原始加工电源 |
| 137 | 那么Anthropic如何处理这个 | |
| 138 | metric? | 标尺? |
| 139 | In the tech hardware space Anthropic's numbers are unheard of. | 在技术硬件空间 Anthropic的数字是无人听说过的. |
| 140 | They keep 85.4 percent | 他们保留了85.4% |
| 141 | as gross profit. | 作为毛利润。 |
| 142 | 85 percent? | 85%吗? |
| 143 | Yeah. | 对 |
| 144 | So out of every dollar they make they keep about 85 cents. | 所以他们每赚一美元,就留了85美分 |
| 145 | Just 15 cents goes to paying for their servers. | 只要付15美分的服务器费 |
| 146 | That level of efficiency is incredible. | 这种效率水平令人难以置信。 |
| 147 | OK. Now | 摆 现在 |
| 148 | compare that to the rivals. | 和对手比较一下 |
| 149 | Right. | 对 |
| 150 | So Z.AI has a gross margin of 26.4 percent. | Z.AI的总比值为26.4%。 |
| 151 | That is a huge drop. | 这是一个巨大的下降。 |
| 152 | 而Minimax则坐着 17.9%的暴徒。 | |
| 153 | Ouch. | 哎哟 |
| 154 | 17.9 percent. | 占17.9%,占比为17.9%. |
| 155 | Yes. | 对 |
| 156 | So for Minimax over 80 cents of every dollar is just vanishing. | 所以,对于Minimax 超过80美分每美元只是消失了。 |
| 157 | It goes right into server costs. | 它直接进入服务器成本。 |
| 158 | It vanishes. | 它会消失的 |
| 159 | It is a very bad situation for them. | 这对他们来说是一个非常糟糕的情况。 |
| 160 | That does not sound sustainable. | 这听起来不可持续。 |
| 161 | It is not. | 事实并非如此。 |
| 162 | And | 还有 |
| 163 | when Anthropic released these amazing numbers the market reacted immediately. | 当Anthropic发布这些惊人的数字时,市场立即作出反应。 |
| 164 | People noticed. | 人们注意到了。 |
| 165 | Oh | 瓳 |
| 166 | yeah. | 对 |
| 167 | Investors looked at that massive efficiency gap and they got worried about the Chinese | 投资者看 巨大的效率差距 他们担心中国人 |
| 168 | companies. | 企业。 |
| 169 | The article says shares for Z.AI and Minimax dropped in Hong Kong. | 文章称Z.AI和Minimax的股份在香港下跌. |
| 170 | They took a hit. | 他们打了一顿 |
| 171 | Yeah. | 对 |
| 172 | Z.AI下降了约13%,Minimax下降了2%。 | |
| 173 | Investors clearly see the | 投资者清楚地看到 |
| 174 | danger there. | 危险 |
| 175 | High revenue does not matter if your costs eat everything. | 高收入并不重要,如果你的成本 吃的一切。 |
| 176 | This brings up a critical | 这引起了一个关键 |
| 177 | question though. | 问吧 |
| 178 | What is that? | 这是什么? |
| 179 | 我们知道Anthropic正在获胜。 | |
| 180 | But why? | 为什么? |
| 181 | Right. | 对 |
| 182 | If their AI models | 如果他们的人工智能模型 |
| 183 | are doing similar jobs why is Anthropic so much more efficient? | 为什么Anthropic的效率要高得多? |
| 184 | The answer comes down to a very | 答案是说 |
| 185 | specific word. | 具体字. |
| 186 | We have to look at how these companies manage compute. | 我们必须看看这些公司如何计算。 |
| 187 | Let's define compute | 我们来计算一下 |
| 188 | as a noun here. | 在这里作为一个名词。 |
| 189 | Yes. | 对 |
| 190 | Compute is a thing. | 计算是一回事。 |
| 191 | It is just the raw processing power required to run these | 只是运行这些设备所需的原始加工动力 |
| 192 | models right? | 模特儿 对吧? |
| 193 | Exactly. | 没错 |
| 194 | It is the raw computer power. | 这是原始的计算机动力。 |
| 195 | You need massive amounts of compute to | 你需要大量的计算 |
| 196 | train an AI. | 训练人工智能。 |
| 197 | You also need it every time a user asks a question. | 用户每次提问时也需要它. |
| 198 | Yeah. | 对 |
| 199 | Every single chat prompt | 每个聊天提示 |
| 200 | burns compute. | 燃烧计算。 |
| 201 | It's like the electricity of the AI world. | 犹如AI世界之电. |
| 202 | That is a great way to say it. | 说得真好 |
| 203 | And | 还有 |
| 204 | compute is incredibly expensive. | 计算太贵了 |
| 205 | Some analysts dug into this right? | 一些分析家发现了这个权利? |
| 206 | Yes. | 对 |
| 207 | Analysts from HSBC Bank. | 汇丰银行分析员. |
| 208 | Ritchie Sun, Charlene Lu and Wang Peishan. | 李奇孙,吕夏琳和王培山. |
| 209 | What did they find? | 他们发现了什么? |
| 210 | They looked at how much money | 他们看了多少钱 |
| 211 | companies make for every single dollar they spend on compute. | 公司每花一美元计算 |
| 212 | The return on investment. | 投资收益. |
| 213 | Exactly. | 没错 |
| 214 | In the second quarter, Anthropic made $2.08 for every $1 spent on compute. | 第二季,安特罗皮克每计算花费1美元就赚了2.08美元. |
| 215 | So they double their | 所以他们加倍 |
| 216 | money on the computer power. | 钱的电脑电源。 |
| 217 | Yes. | 对 |
| 218 | But Z.AI made only $0.46. | 但是Z.AI只赚了0.46美元. |
| 219 | Wait. | 单单 |
| 220 | $0.46 for every dollar spent. | 每花费一美元,0.46美元. |
| 221 | Yeah. | 对 |
| 222 | So Z.AI is actively losing money on their compute costs. | 因此Z.AI在计算成本上 正在积极输钱。 |
| 223 | They spend a dollar to make less | 他们花一块钱赚得少 |
| 224 | than half a dollar back. | 超过半美元回来。 |
| 225 | That is crazy. | 太疯狂了 |
| 226 | It is a big problem. | 这是一个大问题。 |
| 227 | 那么Anthropic是如何做到的? | |
| 228 | Do they | 难道他们 |
| 229 | just buy cheaper computers? | 买更便宜的电脑? |
| 230 | No. | 没有 |
| 231 | They buy the same hardware. | 他们购买同样的硬件。 |
| 232 | Everyone uses the same basic chips. | 每个人都使用同样的基本筹码. |
| 233 | So what is their secret? | 他们的秘密是什么? |
| 234 | The HSBC analysts point to a big concept. | 汇丰银行分析师指出一个大概念. |
| 235 | It is called recursive self | 这叫做自旋 |
| 236 | improvement. | 改进。 |
| 237 | Recursive self improvement. | 不断自我改进. |
| 238 | That sounds complicated. | 这听起来很复杂。 |
| 239 | Let's explain it simply. | 让我们简单解释一下 |
| 240 | Sure. | 当然 |
| 241 | It just means using the AI to write better code. | 意思是用人工智能来写更好的代码 |
| 242 | Okay. | 摆 |
| 243 | The AI finds ways to make the next | AI 找到方法 使下一个 |
| 244 | version of the AI even faster and cheaper to run. | 更快速更便宜地运行。 |
| 245 | So the tool acts as its own mechanic. | 所以这个工具可以充当自己的机械师. |
| 246 | Exactly. | 没错 |
| 247 | It rewrites parts of its own code to move data faster. | 它重写自己代码的部分内容,以更快地移动数据. |
| 248 | And faster data movement means less | 更快的数据移动意味着更少 |
| 249 | server time. | 服务器时间 。 |
| 250 | Right. | 对 |
| 251 | Less server time means lower electricity bills. | 减少服务器时间意味着电费减少. |
| 252 | Anthropic非常擅长这个. | |
| 253 | But wait, I have to push back here. | 等等,我必须推回这里 |
| 254 | Okay. | 摆 |
| 255 | Go ahead. | 来吧。 |
| 256 | The Chinese companies are not just standing | 中国公司不只是站着 |
| 257 | still, right? | 仍然,对不对? |
| 258 | Nope. | 没有 |
| 259 | You are absolutely right. | 你说得对 |
| 260 | They have smart engineers too. | 他们也有聪明的工程师。 |
| 261 | Yes. | 对 |
| 262 | And there is a | 有一个 |
| 263 | surprising fact from the source. | 出乎人们的意料 |
| 264 | Z.AI is actually catching up very fast. | Z.AI实际上赶得很快。 |
| 265 | Really? | 真的吗? |
| 266 | Yeah. | 对 |
| 267 | Because their | 因为他们 |
| 268 | total numbers looked terrible just a minute ago. | 一分钟前的总数看起来很糟糕 |
| 269 | The totals are low, but their speed of improvement | 总数很低,但是改进的速度 |
| 270 | is shocking. | 真让人吃惊 |
| 271 | How fast are we talking? | 我们说话有多快? |
| 272 | Z.AI's revenue per dollar of compute grew 23 times bigger. | Z.AI每美元计算收入增长23倍. |
| 273 | Wow. | 珃 |
| 274 | 23 times. | 诵经二十三遍. |
| 275 | Yeah. | 对 |
| 276 | Since early 2025. | 2025年初开始. |
| 277 | That is massive growth. | 这是巨大的增长。 |
| 278 | It is. | 没错 |
| 279 | For comparison, | 作为比较 |
| 280 | Anthropic's efficiency only grew 4.2 times in that same period. | 同期Anthropic的效率只增长了4.2倍. |
| 281 | So Z.AI is sprinting much faster | Z. AI的冲刺速度快得多 |
| 282 | right now. | 现在 |
| 283 | They are, but they started the race miles behind Anthropic. | 是的,但他们开始比赛 距离安特罗皮克数英里。 |
| 284 | Right. | 对 |
| 285 | So being efficient | 这么有效率 |
| 286 | today is great, but creating the next generation of AI is incredibly expensive. | 今天是伟大的,但创造下一代AI是令人难以置信的昂贵。 |
| 287 | Very expensive. | 非常昂贵。 |
| 288 | This leads to the biggest problem. | 这导致了最大的问题。 |
| 289 | Paying for new research. | 付新研究费. |
| 290 | Yes. | 对 |
| 291 | The research and development | 研究和开发 |
| 292 | trap. | 夹子。 |
| 293 | We have a quote from a tech analyst about this, Po Zhao from Hello China Tech. | 我们从一位技术分析师那里得到一则关于这个的引文,"你好中国技术公司"的宝昭. |
| 294 | Zhao suggests a very simple test for any AI company. | 赵 suggests建议对任何AI公司进行非常简单的测试. |
| 295 | Right. | 对 |
| 296 | What is the test? | 测试是什么? |
| 297 | He asks, | 他问, |
| 298 | can your gross profit pay for your research? | 你的毛利润能支付你的研究费用吗? |
| 299 | That makes sense. | 这有道理。 |
| 300 | If your current product cannot | 如果当前产品不能 |
| 301 | pay to invent your next product, you are in trouble. | 给钱发明你的下一个产品, 你有麻烦了。 |
| 302 | So how does Anthropic do on this test? | 那么,安特罗皮克如何做在这个测试? |
| 303 | Anthropic passes easily. | 安东尼很容易通过。 |
| 304 | Their gross profit is so big it covers all of their research spending. | 他们的毛利润如此之大 涵盖了他们所有的科研开支 |
| 305 | By how much? | 多少钱? |
| 306 | It covers it 1.4 times over. | 它覆盖它1.4倍超过。 |
| 307 | So they fully pay for their future research and still | 因此,他们全额支付 他们未来的研究,仍然 |
| 308 | have profit left over. | 剩下的利润。 |
| 309 | Exactly. | 没错 |
| 310 | Now look at Z.AI. | 现在看看Z.AI。 |
| 311 | Right. | 对 |
| 312 | Their gross profit covers only | 他们的毛利润只包括 |
| 313 | one eighth of their research costs. | 他们的研究成本的八分之一。 |
| 314 | One eighth. | 八分之一门. |
| 315 | That is very low. | 这是非常低的。 |
| 316 | And Minimax | 还有迷你最大 |
| 317 | is worse. | 情况更糟 |
| 318 | They cover only one 14th of their research budget. | 它们只支付研究预算的14分之一. |
| 319 | That is a scary position for | 这是一个可怕的姿势 |
| 320 | that company. | 那个公司 |
| 321 | It is like selling concert tickets for one dollar. | 这就像卖音乐会的门票 一美元。 |
| 322 | Okay. | 摆 |
| 323 | A concert? | 音乐会? |
| 324 | Yeah. | 对 |
| 325 | You sell | 你卖 |
| 326 | the tickets for a dollar. | 一块钱的票 |
| 327 | You get a huge crowd to come. | 你会有很多人来 |
| 328 | The stadium is packed. | 体育场已经打包好了 |
| 329 | Right. | 对 |
| 330 | But at the | 不过,在 |
| 331 | end of the night, you do not have enough money to pay the band. | 晚上,你没有足够的钱支付乐队。 |
| 332 | That is a perfect analogy. | 这是一个完美的类比。 |
| 333 | And that | 还有 |
| 334 | is exactly what is happening in the Chinese market. | 这正是中国市场正在发生的情况。 |
| 335 | Why are they doing that? | 他们为什么要这么做? |
| 336 | The Chinese labs are | 中国的实验室 |
| 337 | fighting a price war. | 参加价格战争 |
| 338 | They use low prices to get more users. | 他们使用低价来获得更多的用户. |
| 339 | So they are practically giving access | 因此,他们实际上提供进入 |
| 340 | away for free. | 免费离开。 |
| 341 | Yes. | 对 |
| 342 | They are desperate to build a massive user base. | 他们急于建立庞大的用户基础. |
| 343 | But Zhao warns that low prices | 但是赵紫阳警告说价格低 |
| 344 | cannot fund future research. | 无法资助未来的研究。 |
| 345 | Right. | 对 |
| 346 | The next generation of A.I. | 后世阿难. |
| 347 | will cost billions to train. | 将花费数十亿的训练。 |
| 348 | You cannot fund that with tiny prices. | 你不能用微薄的价格来资助它。 |
| 349 | So who is paying the bill right now? | 那现在是谁付帐? |
| 350 | Right now, capital | 现在,首都 |
| 351 | markets are paying. | 市场正在支付。 |
| 352 | Investors are covering the gap. | 投资者正在弥补差距。 |
| 353 | But that might not last forever. | 但这可能不会永远持续下去。 |
| 354 | Investor | 投资者 |
| 355 | patience is not infinite. | 忍耐不是无限的。 |
| 356 | No. | 没有 |
| 357 | There is another detail from Zhao. | 赵有别详. |
| 358 | It is about how they actually | 是关于他们实际上是如何 |
| 359 | purchase their computer power. | 购买他们的电脑电源。 |
| 360 | Oh, this strategy difference is huge. | 战略差异很大 |
| 361 | Yeah. | 对 |
| 362 | Anthropic commits to | 人类承诺 |
| 363 | buying computer power 12 to 18 months in advance. | 提前12至18个月购买计算机电源. |
| 364 | They sign long term contracts. | 他们签署了长期合同。 |
| 365 | Like paying for | 就像付钱一样 |
| 366 | something way ahead of time. | 远超时间 |
| 367 | Exactly. | 没错 |
| 368 | This keeps their costs safe and predictable. | 这使他们的费用安全、可预测。 |
| 369 | They are protected | 他们受到保护 |
| 370 | from sudden price jumps. | 从突然的价格跳跃。 |
| 371 | But the Chinese developers use a different strategy. | 但中国开发商使用不同的策略. |
| 372 | Very different. | 非常不一样。 |
| 373 | They buy | 他们买 |
| 374 | compute as a service. | 算作服务. |
| 375 | What does that mean? | 这是什么意思? |
| 376 | They buy it from cloud providers day by day. | 他们日日从云供养者处购买. |
| 377 | Just buying | 只是买 |
| 378 | what they need for that one day. | 有一天他们需要的 |
| 379 | Right. | 对 |
| 380 | On demand. | 随请而至. |
| 381 | But buying day by day means paying the highest | 但买日买日买日买日买日 |
| 382 | possible prices. | 可能的价格。 |
| 383 | It is like renting a car by the hour instead of buying one. | 这就像一个小时租车而不是买车。 |
| 384 | Exactly. | 没错 |
| 385 | And that high | 这么高 |
| 386 | cost just eats their profits away. | 成本只是吃他们的利润。 |
| 387 | So Anthropic's strategy seems flawless. | 所以Anthropic的策略似乎完美无缺. |
| 388 | But wait, let's stop the | 但等等,让我们停止 |
| 389 | conversation right here. | 谈话在这里。 |
| 390 | OK. | 摆 |
| 391 | I want to point out a major twist. | 我想指出一个重大的曲折。 |
| 392 | There is a twist? | 有转弯吗? |
| 393 | Yeah. | 对 |
| 394 | It sounds | 听起来 |
| 395 | like Anthropic is having a perfect year. | 就像Anthropic有一个完美的一年。 |
| 396 | They look unstoppable on paper. | 纸上看起来是无法阻挡的 |
| 397 | But they are not. | 但他们不是。 |
| 398 | The reality | 现实 |
| 399 | of this industry is hitting even the giants. | 这个行业甚至连巨人都打中了 |
| 400 | What do the new numbers show? | 新数字显示什么? |
| 401 | 我们有Anthropic的 | |
| 402 | newest numbers from an amended IPO filing. | 经修订的IPO档案中的最新数字。 |
| 403 | This is for the third quarter of 2026. | 这是2026年第三季度的 |
| 404 | OK. | 摆 |
| 405 | The very | 就是那个 |
| 406 | latest data. | 最新数据。 |
| 407 | Revenue was $17 billion. | 收入170亿美元。 |
| 408 | That was up 47%. | 涨了47% |
| 409 | So revenue is still climbing really fast. | 因此,收入仍在快速攀升。 |
| 410 | But here is the shock. | 但这里是震撼。 |
| 411 | They had an operating loss of $567 million. | 他们的运营损失为5.67亿美元. |
| 412 | They lost over half a billion | 他们损失了5亿多 |
| 413 | in just three months. | 三个月后 |
| 414 | Yes. | 对 |
| 415 | Despite $17 billion in revenue. | 虽然收入170亿美元. |
| 416 | That is crazy. | 太疯狂了 |
| 417 | Because just one quarter | 因为只有四分之一 |
| 418 | before that. | 在那个之前 |
| 419 | Yeah. | 对 |
| 420 | Anthropic had a profit of $974 million. | 安东尼有9.74亿美元的利润. |
| 421 | Going from a huge profit to a | 从巨额利润走向 |
| 422 | massive loss. | 巨大的损失。 |
| 423 | That is a violent swing. | 这是一个暴力的秋千。 |
| 424 | It really is. | 没错 |
| 425 | Why did this happen? | 为什么会这样? |
| 426 | How do you lose money | 你怎么输钱 |
| 427 | like that? | 像这样? |
| 428 | The source points to rising compute expenses. | 来文方指出计算费用不断上升。 |
| 429 | And also slowing revenue growth. | 同时也减缓了收入增长. |
| 430 | So the cost to stay at the top is hurting everyone. | 所以,留在顶端的代价 是伤害每个人。 |
| 431 | Even the biggest players. | 即使是最大的球员。 |
| 432 | You have to | 你必须 |
| 433 | constantly build new massive models to win. | 不断建立新的大规模模式以赢得胜利。 |
| 434 | The costs never stop growing. | 代价从未停止增长。 |
| 435 | This heavy spending is | 这种沉重的开支是 |
| 436 | causing major trouble for their upcoming IPO, right? | 给他们即将到来的IPO带来重大麻烦,对吧? |
| 437 | Oh, absolutely. | 当然 |
| 438 | Timing an IPO is delicate. | IPO的时机是微妙的。 |
| 439 | Investors want a safe bet. | 投资者想要一个安全的赌注。 |
| 440 | Yeah. | 对 |
| 441 | According to the Wall Street Journal, investors are pushing | 据华尔街日报报道,投资者正在推动 |
| 442 | Anthropic to lower its valuation. | 为了降低它的估价, |
| 443 | Valuation. | 估价 |
| 444 | Yeah. | 对 |
| 445 | That is what the total company is worth, right? | 这就是整个公司的价值,对不对? |
| 446 | Yes. | 对 |
| 447 | It sets the starting price for the shares. | 确定股票的起价 |
| 448 | What was the original number? | 原号码是多少? |
| 449 | Anthropic wanted to | 安东尼想 |
| 450 | be valued at $2 trillion. | 价值2万亿美元。 |
| 451 | $2 trillion. | 2万亿元. |
| 452 | That is historic. | 这是历史性的。 |
| 453 | The investors see | 投资者看到 |
| 454 | this new massive loss. | 这个新的巨大损失。 |
| 455 | They want them to drop it. | 他们想让他们放弃 |
| 456 | Drop it to what? | 丢到什么? |
| 457 | They wanted to drop it to | 他们想放弃它 |
| 458 | $1.5 or $1.6 trillion. | 1.5美元或1.6万亿美元。 |
| 459 | So erasing hundreds of billions of dollars in value, just like that. | 因此,去除价值上千亿美元,就这样。 |
| 460 | It shows a deep fear from Wall Street. | 表明华尔街的恐惧 |
| 461 | This brings everything right back to you, the listener. | 这把一切都带回你,听众。 |
| 462 | Yeah, | 对 |
| 463 | let's connect this to daily life. | 让我们把这个和日常生活联系起来 |
| 464 | When you use a cool new AI tool on your phone, you know, | 当你用一个很酷的新AI工具 在你的手机上,你知道, |
| 465 | you ask it a question. | 你问它一个问题。 |
| 466 | The answer pops up in three seconds. | 答案在三秒钟内就出来了 |
| 467 | Right. | 对 |
| 468 | It feels like magic. | 感觉就像魔法 |
| 469 | But remember, | 但是记住, |
| 470 | there is a massive, expensive money machine running behind it. | 后面有一台巨大的昂贵的货币机器 |
| 471 | Burning electricity and cash. | 烧电和现金。 |
| 472 | The ultimate test is coming. | 终极考验即将到来. |
| 473 | Will customers actually pay higher prices for the next big model? | 顾客真的会为下一个大模型支付更高的价格吗? |
| 474 | Will you pay $50 a month instead of $20? | 你会每月付50美元而不是20美元吗? |
| 475 | Exactly. | 没错 |
| 476 | If the answer is no, this whole system might break. | 如果答案是否定的,整个系统可能会崩溃. |
| 477 | It is a risky business model. | 这是一个危险的商业模式。 |
| 478 | I want to leave you with a final thought today, | 今天我想给你一个最后的想法 |
| 479 | something to explore on your own. | 独家探索的东西 |
| 480 | A big question for the future. | 未来大问. |
| 481 | Yes. | 对 |
| 482 | We just saw that a giant | 我们刚刚看到一个巨人 |
| 483 | company making $17 billion a quarter can suddenly lose $567 million. | 公司每季度赚170亿美元 可能会突然损失5.67亿美元 |
| 484 | Just trying to stay competitive. | 只是想保持竞争力 |
| 485 | Right. | 对 |
| 486 | So what happens when investor money finally runs out? | 那么当投资者的钱终于用光了会怎样呢? |
| 487 | That is the scary part. | 这是可怕的部分。 |
| 488 | Will only one or two massive AI companies be left standing in the whole world? | 全世界只有一两家大型AI公司能站在这里吗? |
| 489 | Because | 因为 |
| 490 | no one else can afford to play the game. | 没有人可以负担游戏。 |
| 491 | Exactly. | 没错 |
| 492 | Well, we just have a tiny monopoly | 嗯,我们只是有一个小垄断 |
| 493 | running all the world's AI. | 运行所有世界的AI。 |
| 494 | It is definitely something to watch as this IPO happens. | 随着IPO的出现,这绝对是值得注意的。 |
| 495 | Thank you so much for joining us on this deep dive. | 非常感谢你和我们一起下潜 |
| 496 | Keep questioning the business | 继续质疑生意 |
| 497 | behind the tools. | 工具的后面,我们去 |
| 498 | We will catch you on the next one. | 下一次我们会抓住你的 |
该视频共有字幕 501 条。解锁更多字幕为会员功能，请移动到 价格