1/156 0.44-6.92 <font color=red>n</font> 现在全网最炸裂也最扎心的 a i 对比就是中美两国完全割裂的 token 消耗结构。09-07 18:22:37-09-07 18:22:42<font color=blue>Currently, the most explosive and painful AI comparison is the complete separation of token consumption structures between China and the United States.</font><br>2/156 7.76-9.32 <font color=red>n</font> 很多人都听过一种说法。09-07 18:22:37-09-07 18:22:40<font color=blue>Many people have heard of this.</font><br>3/156 9.74-13.46 <font color=red>xss</font> 美国接近一半的 token用在编程开发和企业自动化，09-07 18:22:38-09-07 18:23:33<font color=blue>Approximately half of the tokens used in the United States are for programming and business automation.</font><br>4/156 13.82-19.28 <font color=red>aaa</font> 而中国的大量算力消耗在短剧短视频和电商用于研发的比例极低。09-07 18:22:39-09-07 18:22:44<font color=blue>However, a large proportion of China's computing power is used for short videos and e-commerce research and development, while the use in other areas is very low.</font><br>5/156 19.94-20.96 <font color=red>hp</font> 首先要说明的是，09-07 18:22:39-09-07 18:22:43<font color=blue>First of all, it should be noted that...</font><br>6/156 21.26-25.76 <font color=red>de</font> 这套流传数据只是行业传播中简化取整后的通俗版本，09-07 18:22:40-09-07 18:24:03<font color=blue>This dataset is a simplified and rounded version of the data that has been circulated in industry,</font><br>7/156 26.06-27.08 <font color=red>n</font> 并不完全精准，09-07 18:22:41-09-07 18:22:42<font color=blue>Not completely accurate.</font><br>8/156 27.32-30.26 <font color=red>n</font> 但它折射出的产业真相却真实而残酷。09-07 18:22:43-09-07 18:22:47<font color=blue>But the truth it reflects is real and harsh.</font><br>9/156 30.8-33.98 <font color=red>n</font> 今天我们就用更接近官方原始调研的口径，09-07 18:22:44-09-07 18:22:47<font color=blue>Today, we will use a more accurate and original survey perspective.</font><br>10/156 34.46-37.76 <font color=red>hp</font> 一次性讲透中美 a i 真正的差距到底藏在哪里？09-07 18:22:44-09-07 18:22:48<font color=blue>What are the real differences between China and the US in AI?</font><br>11/156 38.36-38.42 <font color=red>aaa</font> 一、09-07 18:22:45-09-07 18:22:47<font color=blue>1.</font><br>12/156 38.78-40.16 <font color=red>aaa</font> 先破除全网误区，09-07 18:22:47-09-07 18:22:49<font color=blue>First, dispel the widespread misconception on the internet.</font><br>13/156 40.7-42.02 <font color=red>n</font> 还原真实算力结构。09-07 18:22:48-09-07 18:22:50<font color=blue>Restore the real computing structure.</font><br>14/156 42.68-46.28 <font color=red>n</font> 先看国内真实的公有 a p i 流量分布第一，09-07 18:22:49-09-07 18:22:53<font color=blue>First, look at the real domestic public API traffic distribution first.</font><br>15/156 46.76-50.66 <font color=red>hp</font> a i 短视频短剧和视觉内容生成约占百分之五十五。09-07 18:22:49-09-07 18:22:55<font color=blue>a i Short video and short drama, as well as visual content generation accounts for about 55%.</font><br>16/156 51.26-51.44 <font color=red>aaa</font> 第二，09-07 18:22:50-09-07 18:22:51<font color=blue>Second,</font><br>17/156 51.92-54.86 <font color=red>n</font> 电商素材广告营销内容量产约占百分之二十四。09-07 18:22:51-09-07 18:22:54<font color=blue>Approximately 24% of e-commerce content, advertising and marketing materials are produced.</font><br>18/156 55.7-55.88 <font color=red>aaa</font> 第三，09-07 18:22:51-09-07 18:22:52<font color=blue>Third,</font><br>19/156 56.42-60.57 <font color=red>n</font> 真正的软件开发代码迭代和技术研发仅占百百分之十五，09-07 18:22:55-09-07 18:23:00<font color=blue>The actual software development code iteration and technical research only account for 15% of the time.</font><br>20/156 61.08-64.32 <font color=red>n</font> 剩下约百分之六用于普通闲聊和基础文案改写。09-07 18:22:55-09-07 18:23:00<font color=blue>The remaining approximately 6% is used for general conversation and basic text rewriting.</font><br>21/156 64.98-65.4 <font color=red>hp</font> 也就是说，09-07 18:22:56-09-07 18:22:59<font color=blue>That is to say,</font><br>22/156 65.7-70.2 <font color=red>aaa</font> 国内接近百分之八十的 a i 算力都在服务内容变现和电商流量，09-07 18:22:57-09-07 18:23:00<font color=blue>Nearly 80% of AI computing power in China is used for content monetization and e-commerce traffic.</font><br>23/156 70.56-74.64 <font color=red>de</font> 真正用于底层技术迭代和硬核研发的算力只有约百分之十五。09-07 18:22:58-09-07 18:24:31<font color=blue>Approximately 15% of the computing power is actually used for underlying technology iteration and hardcore research.</font><br>24/156 75.34-78.96 <font color=red>hp</font> 网传的短剧电商百分之三十五编程百分之十，09-07 18:23:00-09-07 18:23:08<font color=blue>According to a leaked short video, the percentage of people who work in e-commerce is 35%, while only 10% are programmers.</font><br>25/156 79.51-81.43 <font color=red>n</font> 本质上是行业早期的粗略估算，09-07 18:23:01-09-07 18:23:04<font color=blue>It is essentially an early and rough estimate in the industry.</font><br>26/156 81.79-85.21 <font color=red>n</font> 经过大量自媒体传播后简化取整形成的通用版本。09-07 18:23:01-09-07 18:23:04<font color=blue>Simplified and generalized version, formed through extensive self-media dissemination.</font><br>27/156 85.93-87.19 <font color=red>aaa</font> 虽然数值不够精准，09-07 18:23:02-09-07 18:23:04<font color=blue>Although the numbers are not precise enough,</font><br>28/156 87.49-90.73 <font color=red>aaa</font> 但中美 a i 应用场景的结构性分化完全属实。09-07 18:23:04-09-07 18:23:07<font color=blue>However, the structural differentiation of application scenarios between China and the US is completely accurate.</font><br>29/156 91.45-93.67 <font color=red>n</font> 再看美国市场没有任何官方报告，09-07 18:23:05-09-07 18:23:07<font color=blue>Without further official reports from the US market,</font><br>30/156 93.79-94.33 <font color=red>n</font> 明确规定，09-07 18:23:06-09-07 18:23:07<font color=blue>Clearly defined,</font><br>31/156 94.51-96.31 <font color=red>n</font> 刚好百分之五十的算力用于编程。09-07 18:23:08-09-07 18:23:11<font color=blue>Only 50% of the computing power is used for programming.</font><br>32/156 97.0-100.72 <font color=red>hp</font> 但从海外 a p i 平台企业付费结构和开发者生态来看，09-07 18:28:10-09-07 18:28:14<font color=blue>However, from the perspective of overseas API platform companies' pricing structure and developer ecosystem,</font><br>33/156 101.13-102.69 <font color=red>bg</font> 可以得出一个相对一致的结论。09-07 18:23:09-09-07 18:23:53<font color=blue>We can arrive at a relatively consistent conclusion.</font><br>34/156 103.17-111.57 <font color=red>n</font> 美国公有 a p i 流量中代码开发程序调试自动化智能体和企业深度研发工作流合计约占百分之五十五至百分之六十，09-07 18:23:10-09-07 18:23:18<font color=blue>In the United States, APIs (Application Programming Interfaces) account for approximately 55% to 60% of code development, debugging, automation, and intelligent enterprise workflows.</font><br>35/156 111.9-114.48 <font color=red>aaa</font> 并贡献了全美百分之七十以上的付费收入。09-07 18:23:11-09-07 18:23:14<font color=blue>And contributed to over 70% of the total revenue.</font><br>36/156 115.19-117.53 <font color=red>hp</font> 网上所说的一半算力用于编程，09-07 18:23:11-09-07 18:23:16<font color=blue>According to what is said online, half of the computing power is used for programming.</font><br>37/156 117.95-121.91 <font color=red>n</font> 其实是对研发加企业智能体深度任务的通俗概括。09-07 18:23:12-09-07 18:23:16<font color=blue>This is a simple explanation of deep learning and enterprise-level intelligent tasks.</font><br>38/156 122.55-122.61 <font color=red>aaa</font> 二、09-07 18:23:15-09-07 18:23:16<font color=blue>II.</font><br>39/156 123.15-129.8 <font color=red>aaa</font> 看懂本质中美 a i 正在走两条截然相反的路 token 是 a i 时代最真实的产业，09-07 18:23:17-09-07 18:23:21<font color=blue>Understanding the Essence of AI: A is walking two completely opposite paths, and tokens are the most authentic product of the AI era.</font><br>40/156 129.98-131.63 <font color=red>n</font> 脚印比总量差距。09-07 18:23:17-09-07 18:23:21<font color=blue>The footprints are different from the total.</font><br>41/156 131.75-134.33 <font color=red>hp</font> 更可怕的是两国算力用途的错位。09-07 18:23:18-09-07 18:23:23<font color=blue>What's even more frightening is the misalignment of computing power usage between the two countries.</font><br>42/156 134.88-136.679 <font color=red>n</font> 中国的 a i 算力主要换流水，09-07 18:23:19-09-07 18:23:21<font color=blue>China's AI computing power is mainly used for data processing.</font><br>43/156 137.04-137.64 <font color=red>n</font> 优先落地，09-07 18:23:21-09-07 18:23:22<font color=blue>Prioritize implementation,</font><br>44/156 137.82-138.24 <font color=red>aaa</font> 商业变现。09-07 18:23:22-09-07 18:23:23<font color=blue>Business monetization.</font><br>45/156 138.96-142.74 <font color=red>n</font> 国内短剧 a i 视频和电商素材生成有一个非常关键，09-07 18:23:22-09-07 18:23:27<font color=blue>In the domestic short video, AI-generated videos and e-commerce materials are very important.</font><br>46/156 143.14-144.9 <font color=red>n</font> 却常被忽略的特点多，09-07 18:23:23-09-07 18:23:26<font color=blue>However, it has many overlooked characteristics.</font><br>47/156 145.02-146.88 <font color=red>aaa</font> 模态内容生成极度消耗 token,09-07 18:23:24-09-07 18:23:26<font color=blue>Modal content generation consumes a lot of tokens.</font><br>48/156 147.06-148.08 <font color=red>hp</font> 也极度吃算力。09-07 18:23:24-09-07 18:23:28<font color=blue>It also consumes a lot of computing power.</font><br>49/156 148.56-149.52 <font color=red>aaa</font> 一分钟 a i 视频，09-07 18:23:27-09-07 18:23:28<font color=blue>One minute AI video,</font><br>50/156 149.7-154.8 <font color=red>n</font> 一套动态分镜或一批广告素材消耗的算力可能是普通代码编写的几十倍。09-07 18:23:28-09-07 18:23:35<font color=blue>The computational resources required for a dynamic layout or a batch of advertising materials could be tens to hundreds of times greater than those needed for ordinary code.</font><br>51/156 155.26-157.36 <font color=red>n</font> 这也解释了国内独特的产业现状。09-07 18:23:28-09-07 18:23:32<font color=blue>This also explains the unique industrial situation in China.</font><br>52/156 157.66-159.82 <font color=red>hp</font> 我们的 a i token 总量位居全球前列，09-07 18:23:29-09-07 18:23:32<font color=blue>Our AI token ranks among the top globally.</font><br>53/156 160.28-161.66 <font color=red>aaa</font> 算力规模也非常庞大，09-07 18:23:30-09-07 18:23:32<font color=blue>The scale of computing power is also very large.</font><br>54/156 162.02-164.79 <font color=red>n</font> 但绝大部分算力换来的是短期内容流水，09-07 18:23:32-09-07 18:23:36<font color=blue>However, most of the computing power resulted in a stream of short-term content.</font><br>55/156 165.15-166.5 <font color=red>aaa</font> 而不是长期技术壁垒。09-07 18:23:33-09-07 18:23:34<font color=blue>Not long-term technical barriers.</font><br>56/156 166.98-169.2 <font color=red>hp</font> 国内 a i 的第一轮爆发极度务实，09-07 18:23:34-09-07 18:23:37<font color=blue>The first wave of AI in China was extremely pragmatic.</font><br>57/156 169.5-170.04 <font color=red>xss</font> 快速落地，09-07 18:23:34-09-07 18:23:53<font color=blue>Quickly land,</font><br>58/156 170.28-170.76 <font color=red>aaa</font> 快速变现、09-07 18:23:35-09-07 18:23:36<font color=blue>Quickly monetize,</font><br>59/156 171.06-171.66 <font color=red>n</font> 快速跑通、09-07 18:23:36-09-07 18:23:37<font color=blue>Quickly run through,</font><br>60/156 171.78-172.32 <font color=red>aaa</font> 商业闭环、09-07 18:23:36-09-07 18:23:37<font color=blue>Business Loop,</font><br>61/156 172.97-173.69 <font color=red>n</font> 短剧量产、09-07 18:23:37-09-07 18:23:39<font color=blue>Short-form video production,</font><br>62/156 173.99-174.17 <font color=red>n</font> 直播、09-07 18:23:38-09-07 18:23:39<font color=blue>Live streaming,</font><br>63/156 174.35-174.53 <font color=red>aaa</font> 脚本、09-07 18:23:38-09-07 18:23:39<font color=blue>Script,</font><br>64/156 174.95-177.74 <font color=red>hp</font> 电商迭代和广告素材更新都是短平快、09-07 18:23:39-09-07 18:23:45<font color=blue>The iterative process of e-commerce and the update of advertising materials are short, fast, and quick.</font><br>65/156 177.98-178.4 <font color=red>n</font> 低门槛，09-07 18:23:40-09-07 18:23:40<font color=blue>Low threshold,</font><br>66/156 178.7-180.32 <font color=red>n</font> 能立刻赚钱的应用层场景。09-07 18:23:40-09-07 18:23:42<font color=blue>Applications that can generate immediate income.</font><br>67/156 180.92-183.62 <font color=red>aaa</font> 这套模式让国产 a i 在最短时间内活了下来，09-07 18:23:41-09-07 18:23:44<font color=blue>This pattern allowed domestic AI to survive in the shortest time.</font><br>68/156 183.94-185.65 <font color=red>n</font> 也撑起了整条产业链的现金流，09-07 18:23:42-09-07 18:23:45<font color=blue>It also supported the cash flow of the entire industrial chain.</font><br>69/156 186.12-187.23 <font color=red>n</font> 但代价同样明显，09-07 18:23:42-09-07 18:23:45<font color=blue>But the cost is also obvious.</font><br>70/156 187.73-189.589 <font color=red>aaa</font> 海量算力堆积在低壁垒高、09-07 18:23:45-09-07 18:23:48<font color=blue>A vast amount of computing power is concentrated in low-barrier, high-volume areas.</font><br>71/156 189.65-191.39 <font color=red>n</font> 内卷可替代性强的应用层，09-07 18:23:46-09-07 18:23:47<font color=blue>Applications with high substitutability,</font><br>72/156 191.81-195.29 <font color=red>hp</font> 真正支撑下一代技术革命的底层研发算力却严重不足。09-07 18:23:46-09-07 18:23:53<font color=blue>The underlying R&D computing power that truly supports the next generation of technological revolution is severely insufficient.</font><br>73/156 195.84-199.8 <font color=red>n</font> 反观美国 a i 算力更多是在换技术堆长期壁垒，09-07 18:23:47-09-07 18:23:53<font color=blue>In contrast, the United States' AI computing power is more focused on technological breakthroughs and long-term barriers.</font><br>74/156 200.24-202.34 <font color=red>n</font> 其绝大多数 token消耗在长链条、09-07 18:23:48-09-07 18:23:51<font color=blue>The vast majority of tokens are consumed by long chains.</font><br>75/156 202.7-203.18 <font color=red>aaa</font> 深推里、09-07 18:23:48-09-07 18:23:50<font color=blue>Deeply,</font><br>76/156 203.53-205.75 <font color=red>aaa</font> 多轮循环和高频试错的硬核任务上，09-07 18:23:51-09-07 18:23:53<font color=blue>In the hardcore tasks that require multiple iterations and high-frequency trial and error,</font><br>77/156 206.27-210.83 <font color=red>n</font> 代码智能体自主编程排错和迭代企业智能体自动调研复盘，09-07 18:23:52-09-07 18:23:55<font color=blue>Code intelligent body autonomous programming debugging and iteration enterprise intelligent body automatic research review,</font><br>78/156 210.95-215.41 <font color=red>hp</font> 并跑通完整工作流研发团队依靠大模型持续进行逻辑推演模型、09-07 18:23:54-09-07 18:24:01<font color=blue>And run the complete workflow, development teams rely on large models for continuous logical reasoning model.</font><br>79/156 215.53-216.55 <font color=red>xss</font> 微调和方案试错。09-07 18:23:54-09-07 18:24:27<font color=blue>Fine-tuning and trial-and-error.</font><br>80/156 217.13-219.47 <font color=red>n</font> 这些任务短期内未必能带来直接流水，09-07 18:23:55-09-07 18:23:57<font color=blue>These tasks may not necessarily bring immediate production in the short term.</font><br>81/156 219.93-224.11 <font color=red>bg</font> 但每一次 token 消耗都在沉淀技术迭代能力逐步拉开带插。09-07 18:23:56-09-07 18:25:08<font color=blue>However, each token consumption is gradually accumulating technical iteration capabilities and widening the gap with others.</font><br>82/156 224.71-225.49 <font color=red>aaa</font> 用一句话总结，09-07 18:23:56-09-07 18:23:58<font color=blue>Summarize in one sentence.</font><br>83/156 225.96-227.94 <font color=red>n</font> 中国 a i 烧算力赚今天的钱。09-07 18:23:57-09-07 18:24:01<font color=blue>China is burning through computing power to make money today.</font><br>84/156 228.36-229.68 <font color=red>n</font> 美国 a i 烧算力铺，09-07 18:23:58-09-07 18:24:01<font color=blue>US AI computing center,</font><br>85/156 229.8-236.32 <font color=red>aaa</font> 明天的路三百分之九十的人忽略的致命盲区私有化部署带来的巨大暗差。09-07 18:23:59-09-07 18:24:02<font color=blue>The 90% of people who ignore the deadly blind spot brought by private deployment tomorrow.</font><br>86/156 236.82-237.72 <font color=red>hp</font> 这是全网解说，09-07 18:24:01-09-07 18:24:04<font color=blue>This is a comprehensive explanation,</font><br>87/156 237.96-239.52 <font color=red>n</font> 几乎都会漏掉的关键真相。09-07 18:24:03-09-07 18:24:04<font color=blue>The key truth that is almost always overlooked.</font><br>88/156 240.14-244.15 <font color=red>aaa</font> 今天所有公开的 token 统计主要包含公有 a p i 流量。09-07 18:24:03-09-07 18:24:05<font color=blue>Today, the main token statistics available are public API traffic.</font><br>89/156 244.55-254.29 <font color=red>de</font> 而美国硅谷巨头科技企业和软件公司大量高端 a i 模型研发系统及企业智能体都部署在私有云内网和独立集群中，09-07 18:24:04-09-07 18:26:06<font color=blue>However, many high-end AI models and enterprise intelligent systems developed by US tech giants and software companies are deployed in private cloud internal networks and independent clusters.</font><br>90/156 254.97-257.49 <font color=red>n</font> 高端研发代码迭代和企业自动化，09-07 18:24:05-09-07 18:24:08<font color=blue>High-end R&D code iteration and enterprise automation,</font><br>91/156 257.73-260.9 <font color=red>hp</font> 完全不走公开 a p i 也就不会被记入榜单数据，09-07 18:24:05-09-07 18:24:12<font color=blue>Since it doesn't use public APIs, it won't be included in the ranking data.</font><br>92/156 261.62-262.16 <font color=red>n</font> 这意味着，09-07 18:24:06-09-07 18:24:08<font color=blue>This means that,</font><br>93/156 262.61-265.73 <font color=red>aaa</font> 美国真实的研发型 token 总量可能被严重低估。09-07 18:24:07-09-07 18:24:09<font color=blue>The actual research and development token supply in the United States may be significantly underestimated.</font><br>94/156 266.29-268.03 <font color=red>n</font> 而国内 a i 生态恰好相反，09-07 18:24:09-09-07 18:24:12<font color=blue>However, the domestic AI ecosystem is quite different.</font><br>95/156 268.39-273.42 <font color=red>n</font> 绝大多数流量来自公有 a p i 终端用户和中小商家的私有化部署比例较低，09-07 18:24:10-09-07 18:24:16<font color=blue>The vast majority of traffic comes from public API endpoints, and the proportion of private deployments by small businesses is relatively low.</font><br>96/156 273.88-275.56 <font color=red>aaa</font> 真实数据几乎完全暴露，09-07 18:24:10-09-07 18:24:12<font color=blue>Real data is almost completely exposed.</font><br>97/156 276.32-279.71 <font color=red>hp</font> 因此真实差距可能比我们看到的数据更加残酷。09-07 18:24:13-09-07 18:24:18<font color=blue>Therefore, the real gap may be even more severe than what we see in the data.</font><br>98/156 280.31-283.01 <font color=red>aaa</font> 国内百分之十五的研发算力几乎就是全部。09-07 18:24:13-09-07 18:24:16<font color=blue>Fifteen percent of the R&D computing power in China accounts for almost all.</font><br>99/156 283.44-286.89 <font color=red>n</font> 而美国百分之五十以上的研发算力可能只是冰山一角，09-07 18:24:14-09-07 18:24:19<font color=blue>However, the 50% or more of R&D capabilities in the United States may only be a small part.</font><br>100/156 287.63-289.37 <font color=red>n</font> 还有一个容易被忽略的小干扰。09-07 18:24:16-09-07 18:24:18<font color=blue>Another easily overlooked disturbance.</font><br>101/156 289.72-293.08 <font color=red>aaa</font> 现在有不少美国企业会调用国内大模型完成基础任务。09-07 18:24:17-09-07 18:24:19<font color=blue>Currently, many American companies are using domestic large models to complete basic tasks.</font><br>102/156 293.52-294.72 <font color=red>n</font> 按照目前的统计规则，09-07 18:24:19-09-07 18:24:20<font color=blue>According to the current statistical rules,</font><br>103/156 295.08-297.51 <font color=red>hp</font> 这部分流量会被计入中国模型的 token 总量，09-07 18:24:20-09-07 18:24:24<font color=blue>This portion of traffic will be counted towards the total token count for Chinese models.</font><br>104/156 297.69-299.73 <font color=red>n</font> 在一定程度上抬高我们的统计盘子。09-07 18:24:20-09-07 18:24:23<font color=blue>To some extent, it raises our statistical plate.</font><br>105/156 300.37-300.55 <font color=red>aaa</font> 不过，09-07 18:24:21-09-07 18:24:22<font color=blue>However,</font><br>106/156 300.91-302.89 <font color=red>n</font> 这些任务大多属于基础应用，09-07 18:24:22-09-07 18:24:23<font color=blue>These tasks are mostly basic applications.</font><br>107/156 303.28-304.6 <font color=red>aaa</font> 真正的核心深度研发，09-07 18:24:22-09-07 18:24:24<font color=blue>True core in-depth research,</font><br>108/156 304.9-306.94 <font color=red>n</font> 依然主要运行在美国本土模型上，09-07 18:24:23-09-07 18:24:27<font color=blue>Still primarily running on models based in the United States.</font><br>109/156 307.56-309.24 <font color=red>n</font> 即使扣除这部分海外流量，09-07 18:24:24-09-07 18:24:27<font color=blue>Even after deducting this portion of overseas traffic,</font><br>110/156 309.54-312.54 <font color=red>aaa</font> 国内本土用于硬核研发的算力占比依旧偏低，09-07 18:24:25-09-07 18:24:27<font color=blue>The proportion of computing power used for hardcore research in domestic areas is still relatively low.</font><br>111/156 313.03-314.35 <font color=red>hp</font> 结构性差距并没有改变。09-07 18:24:25-09-07 18:24:28<font color=blue>Structural gaps have not changed.</font><br>112/156 314.84-314.9 <font color=red>n</font> 四、09-07 18:24:28-09-07 18:24:29<font color=blue>Fourth,</font><br>113/156 315.5-318.02 <font color=red>n</font> 数量狂欢 v 质量红利两套，09-07 18:24:29-09-07 18:24:30<font color=blue>Quantity and quality, two sets of advantages.</font><br>114/156 318.2-319.46 <font color=red>xss</font> 完全不同的商业账本。09-07 18:24:29-09-07 18:24:54<font color=blue>Completely different business ledger.</font><br>115/156 320.1-321.18 <font color=red>aaa</font> 从数据表面看，09-07 18:24:30-09-07 18:24:31<font color=blue>From the surface of the data,</font><br>116/156 321.54-323.76 <font color=red>hp</font> 中国的 token 调用量早已超过美国，09-07 18:24:31-09-07 18:24:35<font color=blue>China's token call volume has already exceeded that of the United States.</font><br>117/156 324.06-325.08 <font color=red>n</font> 稳居全球第一，09-07 18:24:31-09-07 18:24:33<font color=blue>Consistently ranked first globally,</font><br>118/156 325.63-328.39 <font color=red>n</font> 但繁荣背后是完全不对等的商业价值。09-07 18:24:32-09-07 18:24:34<font color=blue>However, behind the prosperity is a completely unequal commercial value.</font><br>119/156 328.93-330.67 <font color=red>hp</font> 国内算利的特征是量大、09-07 18:29:34-09-07 18:29:39<font color=blue>The characteristic of domestic arbitrage is that it involves large amounts.</font><br>120/156 331.03-331.45 <font color=red>de</font> 价格低、09-07 18:24:33-09-07 18:24:55<font color=blue>Low price,</font><br>121/156 331.75-332.11 <font color=red>aaa</font> 利润薄，09-07 18:24:34-09-07 18:24:35<font color=blue>Low profit,</font><br>122/156 332.66-335.24 <font color=red>n</font> 大量短剧视频和电商调用的单价极低。09-07 18:24:35-09-07 18:24:39<font color=blue>The price of a large number of short video and e-commerce promotional videos is very low.</font><br>123/156 335.71-336.67 <font color=red>n</font> 看似数据爆炸，09-07 18:24:35-09-07 18:24:39<font color=blue>Appears to be a data explosion,</font><br>124/156 337.0-339.88 <font color=red>aaa</font> 实际上很难沉淀高质量利润和技术迭代能力。09-07 18:24:36-09-07 18:24:39<font color=blue>In reality, it is very difficult to accumulate high-quality profits and technological innovation capabilities.</font><br>125/156 340.48-342.34 <font color=red>hp</font> 美国算力的特征则完全相反，09-07 18:24:37-09-07 18:24:40<font color=blue>The characteristic of American computing power is completely opposite.</font><br>126/156 342.84-343.68 <font color=red>n</font> 调用量相对稳定，09-07 18:24:39-09-07 18:24:40<font color=blue>Stable call volume,</font><br>127/156 344.04-344.58 <font color=red>n</font> 价格较高，09-07 18:24:40-09-07 18:24:41<font color=blue>Expensive,</font><br>128/156 344.94-345.54 <font color=red>aaa</font> 付费质量高、09-07 18:24:41-09-07 18:24:42<font color=blue>High quality, paid</font><br>129/156 345.84-346.5 <font color=red>hp</font> 技术粘性强。09-07 18:24:41-09-07 18:24:44<font color=blue>Strong technical stickiness.</font><br>130/156 347.25-349.17 <font color=red>n</font> 美国企业不只是按量购买算利，09-07 18:24:42-09-07 18:24:46<font color=blue>American businesses don't just buy based on quantity,</font><br>131/156 349.56-354.12 <font color=red>n</font> 更多采用企业订阅岗位绑定和深度嵌入研发流程的方式。09-07 18:24:43-09-07 18:24:46<font color=blue>Adopt more enterprise subscription models and deeply integrate with the development process.</font><br>132/156 354.78-357.3 <font color=red>aaa</font> 他们付费购买的不是一次对话或一次生成，09-07 18:24:43-09-07 18:24:46<font color=blue>They are not paying for a single conversation or generation.</font><br>133/156 357.6-358.92 <font color=red>hp</font> 而是整体效率升级、09-07 18:24:44-09-07 18:24:48<font color=blue>Instead, it is an overall efficiency upgrade.</font><br>134/156 359.22-361.38 <font color=red>n</font> 技术迭代能力和企业自动化壁垒。09-07 18:24:47-09-07 18:24:49<font color=blue>Technical iteration capability and enterprise automation barriers.</font><br>135/156 361.98-364.32 <font color=red>aaa</font> 这也印证了我们之前讲过的智能体困境，09-07 18:24:47-09-07 18:24:49<font color=blue>This also confirms the agent dilemma we talked about before,</font><br>136/156 364.79-367.79 <font color=red>n</font> 国内用户愿意免费使用海量工具来提升活跃度，09-07 18:24:48-09-07 18:24:54<font color=blue>Domestic users are willing to use a large number of tools for free in order to increase their activity.</font><br>137/156 368.17-371.74 <font color=red>hp</font> 却很难为长期技术价值和深度企业服务持续付费，09-07 18:24:49-09-07 18:24:55<font color=blue>However, it is difficult to ensure long-term technical value and in-depth enterprise services are paid for continuously.</font><br>138/156 372.21-375.33 <font color=red>n</font> 美国市场则已经跑通了为技术效率和壁垒。09-07 18:24:49-09-07 18:24:54<font color=blue>The US market has already run through the process of achieving technological efficiency and barriers.</font><br>139/156 375.51-382.46 <font color=red>aaa</font> 付费的成熟生态终极总结结构决定两国 a i 的未来重新客观复盘网传数据。09-07 18:24:53-09-07 18:24:57<font color=blue>A comprehensive summary of the ultimate structure of paid mature ecosystems determines the future of AI in two countries, objectively reviewing and re-analyzing online data.</font><br>140/156 382.93-384.43 <font color=red>n</font> 虽然它经过了简化取整，09-07 18:24:55-09-07 18:24:57<font color=blue>Although it has been simplified and rounded,</font><br>141/156 384.67-386.11 <font color=red>n</font> 但核心真相并没有错。09-07 18:24:56-09-07 18:24:57<font color=blue>But the core truth is not wrong.</font><br>142/156 386.47-388.93 <font color=red>xss</font> 中国 a i 优先完成了应用层的商业普及。09-07 18:24:56-09-07 18:25:40<font color=blue>China has prioritized the commercialization of applications at the application layer.</font><br>143/156 389.35-392.17 <font color=red>de</font> 美国 a i 则率先完成了底层技术的深度扎根。09-07 18:24:57-09-07 18:25:40<font color=blue>The United States has been the first to deeply integrate its underlying technology.</font><br>144/156 392.65-396.01 <font color=red>hp</font> 我们用大约八成算力换来了产业生存和短期流水，09-07 18:24:58-09-07 18:25:03<font color=blue>We used about 80% of our computing power to achieve industrial survival and short-term production.</font><br>145/156 396.37-398.89 <font color=red>n</font> 却只有很少一部分算力投入底层创新。09-07 18:24:58-09-07 18:25:02<font color=blue>However, only a small portion of the computing power is invested in fundamental innovation.</font><br>146/156 399.52-402.4 <font color=red>aaa</font> 美国则用过半算力持续深耕研发与自动化，09-07 18:24:59-09-07 18:25:03<font color=blue>The United States has also been deeply involved in research and development, as well as automation, using more than half of its computing power.</font><br>147/156 402.76-404.8 <font color=red>n</font> 默默堆叠下一代技术的代差优势。09-07 18:24:59-09-07 18:25:02<font color=blue>Accumulate silently and create a technological advantage for the next generation.</font><br>148/156 405.47-408.71 <font color=red>n</font> a i 的下半场早就不再是比拼谁的 token 烧的多，09-07 18:25:02-09-07 18:25:06<font color=blue>The second half of the game between A and I was no longer about who could burn more tokens.</font><br>149/156 409.42-410.92 <font color=red>n</font> 真正的终极胜负手，09-07 18:25:03-09-07 18:25:06<font color=blue>The ultimate winning strategy,</font><br>150/156 411.1-414.98 <font color=red>aaa</font> 只有一个你的海量算力究竟是用来换短期流量，09-07 18:25:04-09-07 18:25:08<font color=blue>Instead of focusing on long-term growth, you are only using your massive computing power to generate short-term traffic.</font><br>151/156 415.28-417.2 <font color=red>hp</font> 还是用来换长期技术壁垒。09-07 18:25:04-09-07 18:25:07<font color=blue>Used to break long-term technical barriers.</font><br>152/156 417.67-420.13 <font color=red>n</font> 中国 a i 的下一轮拐点已经非常清晰，09-07 18:25:07-09-07 18:25:11<font color=blue>The next turning point for China's AI is very clear.</font><br>153/156 420.59-425.21 <font color=red>n</font> 从算力换变现全面转向算力换迭代换底层突破。09-07 18:25:08-09-07 18:25:14<font color=blue>From a comprehensive shift from computing power to monetization, to iterative improvements driven by computing power, and finally to breakthroughs in the underlying technology.</font><br>154/156 426.05-430.43 <font color=red>hp</font> 只有当研发编程和企业深度智能体所占的算力比例持续提升。09-07 18:25:08-09-07 18:25:15<font color=blue>Only when the proportion of computing power occupied by research and development programming and enterprise-level intelligent systems continues to increase.</font><br>155/156 430.79-433.07 <font color=red>aaa</font> 我们的 a i 产业才能从应用层内卷，09-07 18:25:09-09-07 18:25:11<font color=blue>Our AI industry can only thrive from the application layer,</font><br>156/156 433.43-435.05 <font color=red>bg</font> 真正走向底层技术崛起。09-07 18:25:10-09-07 18:25:46<font color=blue>Truly embracing the rise of underlying technologies.</font><br>