BV1X97Z6AEWo · 请不要叫我梁同学1 00:00:00,140 --> 00:00:02,560 美股AI量化分析工具 2 00:00:02,560 --> 00:00:03,279 大家好 3 00:00:03,279 --> 00:00:04,780 我们是美股投资网 4 00:00:04,780 --> 00:00:07,700 用大数据驱动你的交易决策 5 00:00:07,700 --> 00:00:11,080 当别人还在盯着英伟达镁光内存的时候 6 00:00:11,080 --> 00:00:14,589 我们已经画了一张AI上下游产业链地图 7 00:00:14,589 --> 00:00:16,269 阅读了海量低手 8 00:00:16,269 --> 00:00:19,309 华尔街顶级投行一手研究报告 9 00:00:19,309 --> 00:00:23,618 我们分析出芯片内部互联和长距离通信 10 00:00:23,618 --> 00:00:27,618 marvel呢将从800G到1.6TDSP 11 00:00:27,618 --> 00:00:28,778 市场份额增长 12 00:00:28,778 --> 00:00:32,429 及超大规模NPO需求提升中受益 13 00:00:32,429 --> 00:00:36,689 marvel最容易被市场低估的资产之一是其光模块 14 00:00:36,689 --> 00:00:38,320 DSP芯片 15 00:00:38,320 --> 00:00:41,219 光纤传输信号会失真衰减 16 00:00:41,219 --> 00:00:42,159 产生噪音 17 00:00:42,159 --> 00:00:45,590 DSP就像一个实时翻译官和纠错系统 18 00:00:45,590 --> 00:00:47,109 如果没有DSP 19 00:00:47,109 --> 00:00:51,509 800G和1.6T光模块几乎无法稳定工作 20 00:00:51,509 --> 00:00:54,149 mar某在2021年收购了INFI 21 00:00:54,149 --> 00:00:55,820 花了约100亿美元 22 00:00:55,820 --> 00:00:57,780 很多人当时觉得买贵了 23 00:00:57,780 --> 00:00:58,780 现在回头看 24 00:00:58,780 --> 00:01:02,450 那笔收购几乎是AI时代最成功的并购之一 25 00:01:02,450 --> 00:01:07,030 因为INFI原本就是全球光通信DSP龙头 26 00:01:07,030 --> 00:01:09,030 在1.6T光模块里 27 00:01:09,030 --> 00:01:12,049 单科DSP的价值100到200美元 28 00:01:12,049 --> 00:01:13,879 绝对是价值之王 29 00:01:13,879 --> 00:01:16,409 marvel是DSP的绝对龙头 30 00:01:16,409 --> 00:01:18,269 反而博通是老二 31 00:01:18,269 --> 00:01:19,370 基于这个结论 32 00:01:19,370 --> 00:01:23,230 我们在2025年11月和2026年1月份 33 00:01:23,230 --> 00:01:24,409 分别喊单 34 00:01:24,409 --> 00:01:29,049 以80亿美元让VIP会员重仓30%marvel 35 00:01:29,049 --> 00:01:31,290 而英伟达在3月底才宣布 36 00:01:31,290 --> 00:01:33,280 向marvel注资20亿美元 37 00:01:33,280 --> 00:01:34,140 6月1日 38 00:01:34,140 --> 00:01:35,299 华人军表态 39 00:01:35,299 --> 00:01:38,489 MAO有望成为下一家万亿美元公司 40 00:01:38,489 --> 00:01:42,150 意味着距离现在股价有五倍的上涨空间 41 00:01:42,150 --> 00:01:45,510 我们足足比黄仁勋早了四个月入股marvel 42 00:01:45,510 --> 00:01:47,620 可见我们的眼光独到 43 00:01:47,620 --> 00:01:48,599 更重要的是 44 00:01:48,599 --> 00:01:49,900 我们言出必行 45 00:01:49,900 --> 00:01:54,299 用自己的实盘账号重仓30%买入marvel 46 00:01:54,299 --> 00:01:57,379 这是我们仓位另一支核心重仓股 47 00:01:57,379 --> 00:02:00,180 是2026年必买十只股的首选 48 00:02:00,180 --> 00:02:02,280 mu和其他潜力股 49 00:02:02,280 --> 00:02:06,260 过去一年我们整个组合回报率高达300% 50 00:02:06,260 --> 00:02:09,129 过去三个月回报率127% 51 00:02:09,129 --> 00:02:11,389 从资金曲线图可以看到 52 00:02:11,389 --> 00:02:13,590 我们的收益增长相对平滑 53 00:02:13,590 --> 00:02:15,639 几乎没有出现过大回撤 54 00:02:15,639 --> 00:02:19,639 这份成绩不仅远超大盘同期约10%的涨幅 55 00:02:19,639 --> 00:02:22,979 更体现了我们在追求超额收益的同时 56 00:02:22,979 --> 00:02:24,939 始终严格控制风险 57 00:02:24,939 --> 00:02:26,430 注重资金安全 58 00:02:26,430 --> 00:02:28,330 我们把选股做到了极致 59 00:02:28,330 --> 00:02:30,810 我们严格筛选公司质量 60 00:02:30,810 --> 00:02:33,030 尽量避开基本面薄弱 61 00:02:33,030 --> 00:02:34,550 股价波动剧烈 62 00:02:34,550 --> 00:02:37,769 容易受到市场影响的小盘投机股 63 00:02:37,769 --> 00:02:40,709 正因为这种长期稳定的投资表现 64 00:02:40,709 --> 00:02:44,509 越来越多投资者选择加入美股投资网VIP会员 65 00:02:44,509 --> 00:02:45,449 截至目前 66 00:02:45,449 --> 00:02:47,629 会员人数已突破2万人 67 00:02:47,629 --> 00:02:48,969 持续创下新高 68 00:02:48,969 --> 00:02:53,360 成为全球华人美股投资研究领域龙头公司 69 00:02:53,360 --> 00:02:54,879 6月刚开始 70 00:02:54,879 --> 00:02:57,360 美股的投资主题已经很清楚 71 00:02:57,360 --> 00:03:00,139 但个股的机会仍需要大家具备专业知识 72 00:03:00,139 --> 00:03:01,409 才能更好的理解 73 00:03:01,409 --> 00:03:04,490 所以这期视频我尽量帮大家讲解清楚 74 00:03:04,490 --> 00:03:07,099 AI到底从哪里开始落地 75 00:03:07,099 --> 00:03:08,780 有哪些公司会受益 76 00:03:08,780 --> 00:03:10,379 我会给出明确代号 77 00:03:10,379 --> 00:03:11,860 记得点赞收藏 78 00:03:11,860 --> 00:03:12,819 好好了 79 00:03:12,819 --> 00:03:13,699 废话不多说 80 00:03:13,699 --> 00:03:16,709 现在开始第一条线就是AIPC 81 00:03:16,709 --> 00:03:18,429 但大家一定要搞清楚 82 00:03:18,429 --> 00:03:21,169 老黄这次推的R7x Mark 83 00:03:21,169 --> 00:03:25,079 和过去几年市场炒的AIPC根本不是一回事 84 00:03:25,079 --> 00:03:28,179 以前大家说的AIPC其实还是普通电脑 85 00:03:28,179 --> 00:03:30,378 只是在CPU旁边多了一个 86 00:03:30,378 --> 00:03:33,629 专门跑AI小任务的小芯片 87 00:03:33,629 --> 00:03:35,889 他能做的就是一些清亮功能 88 00:03:35,889 --> 00:03:37,229 比如说视频美颜 89 00:03:37,229 --> 00:03:38,069 背景虚化 90 00:03:38,069 --> 00:03:39,430 语音转文字 91 00:03:39,430 --> 00:03:42,819 简单来说就是在电脑里加了几个AI小插件 92 00:03:42,819 --> 00:03:45,719 但电脑本身还是那台普通电脑 93 00:03:45,719 --> 00:03:49,780 并没有真正变成能跑大型AI模型的工作站 94 00:03:49,780 --> 00:03:52,099 但RT x spark不一样 95 00:03:52,099 --> 00:03:54,620 它不是给电脑多加几个AI功能 96 00:03:54,620 --> 00:03:59,039 而是把PC变成一台真正的本地AI工作站 97 00:03:59,039 --> 00:04:01,479 128GB统一内存 98 00:04:01,479 --> 00:04:03,280 1pet flop AI算力 99 00:04:03,280 --> 00:04:05,379 再加上BLACKWELLGPU 100 00:04:05,379 --> 00:04:07,939 a arm CPU和库的生态 101 00:04:07,939 --> 00:04:09,139 目的很明确 102 00:04:09,139 --> 00:04:13,579 让个人电脑可以在本地运行AI模型和AI智能体 103 00:04:13,579 --> 00:04:16,449 这就是这次叙事最大的变化 104 00:04:16,449 --> 00:04:19,529 过去两年AI的主战场几乎都在云端 105 00:04:19,529 --> 00:04:20,589 你输入问题 106 00:04:20,589 --> 00:04:22,928 数据传到服务器模型 107 00:04:22,928 --> 00:04:25,129 算完再把结果返回给你 108 00:04:25,129 --> 00:04:26,809 你的电脑只是入口 109 00:04:26,809 --> 00:04:28,788 真正的大脑在数据中心 110 00:04:28,788 --> 00:04:31,809 但如果P7本身有足够强的算力 111 00:04:31,809 --> 00:04:33,000 逻辑就变了 112 00:04:33,000 --> 00:04:35,240 它可以直接读取你的文件 113 00:04:35,240 --> 00:04:36,519 理解你的工作流 114 00:04:36,519 --> 00:04:37,899 调用本地应用 115 00:04:37,899 --> 00:04:39,920 直接跨软件执行任务 116 00:04:39,920 --> 00:04:42,759 同时减少敏感数据上传云端 117 00:04:42,779 --> 00:04:45,100 这才是AIPC真正的价值 118 00:04:46,459 --> 00:04:48,639 第一个受益者当然是英伟达 119 00:04:48,639 --> 00:04:51,879 因为英伟达要做的不只是多卖一块显卡 120 00:04:51,879 --> 00:04:55,379 而是把自己在数据中心已经验证过的AI平台 121 00:04:55,379 --> 00:04:58,779 搬到个人电脑和企业终端过去 122 00:04:58,779 --> 00:05:01,139 英伟达控制的是云端AI算力 123 00:05:01,139 --> 00:05:03,009 如果R7x bug跑通 124 00:05:03,009 --> 00:05:05,949 他未来还想控制本地AI算力入口 125 00:05:05,949 --> 00:05:08,639 这就是英伟达最核心的逻辑 126 00:05:08,639 --> 00:05:10,399 第二个受益者是高通 127 00:05:10,399 --> 00:05:12,480 很多人呢会忽略高通 128 00:05:12,480 --> 00:05:15,168 但其实在英伟达真正下场之前 129 00:05:15,168 --> 00:05:17,269 windows on arm这条线 130 00:05:17,269 --> 00:05:19,519 高通是最早推动的玩家 131 00:05:19,519 --> 00:05:22,860 高通的snap dragon x电脑芯片系列 132 00:05:22,860 --> 00:05:26,480 已经先把arm架构带进windows pc 133 00:05:26,480 --> 00:05:28,360 也推动了第一波AIPC 134 00:05:28,360 --> 00:05:29,939 它的优势是什么呢 135 00:05:29,939 --> 00:05:33,079 低功耗长续航轻薄本体验好 136 00:05:33,079 --> 00:05:35,339 适合普通消费者和办公人群 137 00:05:35,339 --> 00:05:37,899 所以高通当然受益AIPC 138 00:05:37,899 --> 00:05:39,339 但现在问题来了 139 00:05:39,339 --> 00:05:40,839 英伟达下场以后 140 00:05:40,839 --> 00:05:42,918 高僧的故事就没那么独家了 141 00:05:42,918 --> 00:05:46,319 因为高通主要讲的是让windows pc更省电 142 00:05:46,319 --> 00:05:47,158 更轻薄 143 00:05:47,158 --> 00:05:48,778 更适合端侧AI 144 00:05:48,778 --> 00:05:49,978 但英伟达讲的是 145 00:05:49,978 --> 00:05:51,218 我不只是CPU 146 00:05:51,218 --> 00:05:53,858 我还有black well GPU库的生态 147 00:05:53,858 --> 00:05:56,240 本地大模型和AI智能体 148 00:05:56,240 --> 00:05:58,160 所以高通不是不受益 149 00:05:58,160 --> 00:06:01,019 而是他更偏向普通AIPC普及 150 00:06:01,019 --> 00:06:05,180 英伟达更偏向高端AI工作站和企业生产力 151 00:06:05,180 --> 00:06:07,970 第三个受益者是ARM 152 00:06:07,970 --> 00:06:11,110 这次高通和英伟达都选择arm架构 153 00:06:11,110 --> 00:06:12,389 说明一件事 154 00:06:12,389 --> 00:06:17,009 windows pc长期被X86主导的格局开始松动了 155 00:06:17,009 --> 00:06:18,069 过去几十年 156 00:06:18,069 --> 00:06:20,990 PC市场基本上是英特尔和AMD的天下 157 00:06:20,990 --> 00:06:23,839 但苹果M芯片系列已经证明 158 00:06:23,839 --> 00:06:26,300 arm不是只能做手机芯片 159 00:06:26,300 --> 00:06:28,040 也可以做高性能电脑 160 00:06:28,040 --> 00:06:31,089 现在高通先推windows on arm 161 00:06:31,089 --> 00:06:35,069 英伟达先把arm CPU带进RTX spark 162 00:06:35,069 --> 00:06:37,529 这对arm来说意义很大 163 00:06:37,529 --> 00:06:40,029 因为他的故事不再只是手机授权 164 00:06:40,029 --> 00:06:42,959 而是有机会进入更高价值的PC 165 00:06:42,959 --> 00:06:45,300 AI终端和企业设备市场 166 00:06:45,300 --> 00:06:48,220 如果windows AI p c真正跑起来 167 00:06:48,220 --> 00:06:50,879 a arm就会躺着收授权费 168 00:06:50,879 --> 00:06:52,180 在5月20日 169 00:06:52,180 --> 00:06:53,319 我们的美股大数据 170 00:06:53,319 --> 00:06:58,170 stock way AI模型就选出a arm作为AI首选股 171 00:06:58,170 --> 00:07:00,829 当时242美元提醒买入 172 00:07:00,829 --> 00:07:03,629 截至本周一获利75% 173 00:07:03,629 --> 00:07:06,100 第四个受益者是微软 174 00:07:06,100 --> 00:07:09,000 微软这次不是简单提供一个windows系统 175 00:07:09,000 --> 00:07:11,720 他真正要做的是让AI智能体能 176 00:07:11,720 --> 00:07:13,240 安全的在电脑里工作 177 00:07:13,240 --> 00:07:15,879 因为企业最担心的不是AI不够聪明 178 00:07:15,879 --> 00:07:18,199 而是AI会不会乱读文件 179 00:07:18,199 --> 00:07:19,019 乱动系统 180 00:07:19,019 --> 00:07:20,240 乱调用数据 181 00:07:20,240 --> 00:07:22,019 AI如果要进入你的电脑 182 00:07:22,019 --> 00:07:24,730 调用你的文件邮件表格 183 00:07:24,730 --> 00:07:27,689 浏览器和企业系统就必须有权限管理 184 00:07:27,689 --> 00:07:29,490 安全隔离和数据保护 185 00:07:29,490 --> 00:07:31,170 这就是微软的价值 186 00:07:31,170 --> 00:07:33,790 未来智能体如果变成数字员工 187 00:07:33,790 --> 00:07:37,850 那windows就是管理这些数字员工的操作系统 188 00:07:37,850 --> 00:07:39,449 谁控制操作系统 189 00:07:39,449 --> 00:07:41,689 谁就控制AI智能体的入口 190 00:07:41,689 --> 00:07:42,810 谁控制入口 191 00:07:42,810 --> 00:07:44,839 谁就有长期收税权 192 00:07:44,839 --> 00:07:48,220 第五个呢是电脑公司和企业硬件公司 193 00:07:48,220 --> 00:07:52,250 但这里不能简单说AIPC利好所有电脑公司 194 00:07:52,250 --> 00:07:54,410 真正要看的是谁只卖电脑 195 00:07:54,410 --> 00:07:55,910 谁还能卖服务器 196 00:07:55,910 --> 00:07:58,750 存储网络和整套企业AI方案 197 00:07:58,750 --> 00:08:01,170 我认为最值得放前面的是戴尔 198 00:08:01,170 --> 00:08:02,889 因为戴尔不只是卖电脑 199 00:08:02,889 --> 00:08:04,819 他和英伟达绑定的更深 200 00:08:04,819 --> 00:08:07,240 戴尔现在推的不是简单AIPC 201 00:08:07,240 --> 00:08:10,220 而是戴尔AI factory with英伟达 202 00:08:10,220 --> 00:08:11,509 什么意思 203 00:08:11,509 --> 00:08:13,290 企业如果想部署AI 204 00:08:13,290 --> 00:08:16,269 不一定自己从零开始买GPU配服务器 205 00:08:16,269 --> 00:08:17,839 搭网络做存储 206 00:08:17,839 --> 00:08:21,300 戴尔可以把英伟达的GPU服务器存储 207 00:08:21,300 --> 00:08:21,819 网络 208 00:08:21,819 --> 00:08:26,158 工作站和服务打包成一整套企业AI工厂方案 209 00:08:26,158 --> 00:08:28,759 所以戴尔的优势是前后通吃 210 00:08:28,759 --> 00:08:32,850 前端有商用PC和工作站能吃AIPC换机 211 00:08:32,850 --> 00:08:35,870 后端有AI服务器存储和企业服务 212 00:08:35,870 --> 00:08:38,038 能吃AI工厂建设 213 00:08:38,038 --> 00:08:41,788 这就是为什么戴尔比普通PC厂商更值得重视 214 00:08:41,788 --> 00:08:44,288 他不是只吃一台电脑的钱 215 00:08:44,288 --> 00:08:47,149 而是吃企业AI落地的一整套钱 216 00:08:47,149 --> 00:08:48,669 第二个是惠普 217 00:08:48,669 --> 00:08:50,509 惠普的逻辑更纯粹 218 00:08:50,509 --> 00:08:52,230 就是AIP交换机 219 00:08:52,230 --> 00:08:54,700 这里有一个非常关键的数据 220 00:08:54,700 --> 00:08:55,740 惠普第二季度 221 00:08:55,740 --> 00:08:59,460 AIPC已经占到PC总出货量的44% 222 00:08:59,460 --> 00:09:02,179 上一季度才刚刚超过35% 223 00:09:02,179 --> 00:09:04,600 这说明AIPC不是PPT概念 224 00:09:04,600 --> 00:09:07,100 而是已经进入真实出货结构 225 00:09:07,100 --> 00:09:09,470 但惠普的问题也很明显 226 00:09:09,470 --> 00:09:11,490 它主要指的是终端换机 227 00:09:11,490 --> 00:09:14,549 高端商用PC升级和asp提升 228 00:09:14,549 --> 00:09:17,230 如果企业开始大规模换AIPC 229 00:09:17,230 --> 00:09:18,799 惠普会直接受益 230 00:09:18,799 --> 00:09:20,240 但它不像戴尔那样 231 00:09:20,240 --> 00:09:21,559 后面还有AI服务器 232 00:09:21,559 --> 00:09:25,220 存储网络和AI工厂这条更长的链路 233 00:09:25,220 --> 00:09:29,710 所以惠普更像是最纯粹的AIPC换机标的 234 00:09:29,710 --> 00:09:31,610 第三个呢是HPE 235 00:09:31,610 --> 00:09:33,470 HPE呢不做个人电脑 236 00:09:33,470 --> 00:09:35,830 所以他不是AIPC的直接受益者 237 00:09:35,830 --> 00:09:38,330 但企业真正用上AI以后 238 00:09:38,330 --> 00:09:40,509 不是买几台AI电脑就结束了 239 00:09:40,509 --> 00:09:42,190 每个员工都要用智能体 240 00:09:42,190 --> 00:09:43,679 每台电脑都跑AI 241 00:09:43,679 --> 00:09:47,759 后面一定需要服务器存储网络和混合云来支撑 242 00:09:47,759 --> 00:09:49,669 这就是hp的机会 243 00:09:49,669 --> 00:09:52,870 最新季度hp收入已经达到107亿美元 244 00:09:52,870 --> 00:09:54,429 同比增长40% 245 00:09:54,429 --> 00:09:57,090 AI系统订单达到18亿美元 246 00:09:57,090 --> 00:09:59,350 累计订单达到164亿美元 247 00:09:59,350 --> 00:10:03,100 手里还有59亿美元AI订单等着交付 248 00:10:03,100 --> 00:10:05,779 这说明企业是真的在买设备 249 00:10:05,779 --> 00:10:08,080 准备把AI跑进自己的系统里 250 00:10:08,080 --> 00:10:11,049 而且hp也和英伟达有合作 251 00:10:11,049 --> 00:10:13,830 推出基于BLACKWELL和robin的AI服务器 252 00:10:13,830 --> 00:10:15,620 和整套企业AI方案 253 00:10:15,620 --> 00:10:17,779 所以hp的逻辑很简单 254 00:10:17,779 --> 00:10:21,789 它不吃AIPC本身吃的是AIPC和智能体 255 00:10:21,789 --> 00:10:22,769 进入企业以后 256 00:10:22,769 --> 00:10:25,039 后端基础设施升级的前 257 00:10:25,039 --> 00:10:28,379 在美股投资网的潜力爆发股服务中 258 00:10:28,379 --> 00:10:32,080 我们在5月13日就选出了hp作为潜力股 259 00:10:32,080 --> 00:10:33,860 当时股价30亿美元 260 00:10:33,860 --> 00:10:35,860 本周二股价暴涨20% 261 00:10:35,860 --> 00:10:38,279 目前累计盈利已达到80% 262 00:10:38,279 --> 00:10:40,000 这服务非常受欢迎 263 00:10:40,000 --> 00:10:42,549 非常非常多的人订阅我们的服务 264 00:10:42,549 --> 00:10:44,850 这里我还必须要提存储 265 00:10:44,850 --> 00:10:47,690 我之前在文章里反复强调过一句话 266 00:10:47,690 --> 00:10:49,830 AI时代存储是基建 267 00:10:49,830 --> 00:10:51,690 这次的观点依然成立 268 00:10:51,690 --> 00:10:52,549 为什么 269 00:10:52,549 --> 00:10:54,809 因为AIH能体要真正好用 270 00:10:54,809 --> 00:10:56,190 不能只有算力 271 00:10:56,190 --> 00:10:58,839 还必须有足够大的内存和存储 272 00:10:58,839 --> 00:11:01,519 没有内存智能体就像一个脑子很聪明 273 00:11:01,519 --> 00:11:03,009 但记忆力很差的人 274 00:11:03,009 --> 00:11:04,649 他跑不动大模型 275 00:11:04,649 --> 00:11:06,549 吃不下长上下文 276 00:11:06,549 --> 00:11:09,740 也无法长期理解你的文件和工作流 277 00:11:09,740 --> 00:11:12,340 所以在这一轮AIPC换机潮里 278 00:11:12,340 --> 00:11:16,818 内存厂商的利润弹性可能会比整机厂更早体现 279 00:11:16,818 --> 00:11:19,739 因为整机厂还要等终端销量放量 280 00:11:19,739 --> 00:11:22,538 但上游内存和高带宽存储价格 281 00:11:22,538 --> 00:11:24,578 往往会提前反映供需紧张 282 00:11:24,578 --> 00:11:25,879 这也是为什么 283 00:11:25,879 --> 00:11:30,440 我们美股投资网从年初就一直强调并加仓mu 284 00:11:30,440 --> 00:11:32,960 同时台积电也会继续受益 285 00:11:32,960 --> 00:11:37,039 R7x Mark采用的是先进制成这种高性能芯片 286 00:11:37,039 --> 00:11:40,649 对应的单价和毛利率远高于传统成熟制程 287 00:11:40,649 --> 00:11:42,929 只要端侧AI设备开始放量 288 00:11:42,929 --> 00:11:45,909 先进制程需求就不只是来自数据中心 289 00:11:45,909 --> 00:11:50,090 也会逐步扩散到高端PC和本地AI工作站 290 00:11:50,090 --> 00:11:52,250 所以这条线讲到这里 291 00:11:52,250 --> 00:11:53,950 其实已经很明白了 292 00:11:53,950 --> 00:11:56,950 R7x Mark不只是在PC上缝缝补补 293 00:11:56,950 --> 00:11:58,690 而是从底层架构上 294 00:11:58,690 --> 00:12:01,730 把PC从一个被动使用的显示终端 295 00:12:01,730 --> 00:12:05,279 变成了一个主动执行任务的本地AI工作站 296 00:12:05,279 --> 00:12:07,340 这背后算力在转移 297 00:12:07,340 --> 00:12:08,340 生态在迁移 298 00:12:08,340 --> 00:12:10,820 整条产业链的利润分配也在重组 299 00:12:10,820 --> 00:12:13,269 而这一切才刚刚开始 300 00:12:13,269 --> 00:12:15,409 如果你觉得本视频对你有帮助 301 00:12:15,409 --> 00:12:16,090 老规矩 302 00:12:16,090 --> 00:12:17,350 先点赞再收藏 303 00:12:17,350 --> 00:12:18,570 关键时刻能帮忙 304 00:12:18,570 --> 00:12:21,490 美国热线6263783637 305 00:12:21,490 --> 00:12:22,450 讲到这里 306 00:12:22,450 --> 00:12:24,090 很多人可能会有一个误解 307 00:12:24,090 --> 00:12:26,250 既然本地AIPC变强了 308 00:12:26,250 --> 00:12:28,529 是不是云端算力就不重要了 309 00:12:28,529 --> 00:12:29,568 恰恰相反 310 00:12:29,568 --> 00:12:32,729 本地AIPC不是云端AI的终结 311 00:12:32,729 --> 00:12:35,009 而是AI需求扩散的开始 312 00:12:35,009 --> 00:12:37,889 过去AI主要集中在云端入口 313 00:12:37,889 --> 00:12:39,269 很少未来不一样 314 00:12:39,269 --> 00:12:40,889 电脑会变成入口 315 00:12:40,889 --> 00:12:42,489 企业软件会变成入口 316 00:12:42,489 --> 00:12:44,279 数据平台会变成入口 317 00:12:44,279 --> 00:12:45,299 客服系统 318 00:12:45,299 --> 00:12:46,940 ERP设计软件 319 00:12:46,940 --> 00:12:49,340 工业软件都会变成AI的入口 320 00:12:49,340 --> 00:12:52,419 而这些入口最终要跑的就是智能体 321 00:12:52,419 --> 00:12:55,159 所以第二条线就是AI智能体 322 00:12:55,159 --> 00:12:57,970 但智能体不是聊天机器人 323 00:12:57,970 --> 00:13:00,190 聊天机器人只是回答问题 324 00:13:00,190 --> 00:13:03,049 智能体是能调用工具读取数据 325 00:13:03,049 --> 00:13:04,889 执行任务的数字员工 326 00:13:04,889 --> 00:13:08,070 企业真正需要的不是一个更会聊天的AI 327 00:13:08,070 --> 00:13:10,099 而是一个能安全干活的AI 328 00:13:10,099 --> 00:13:11,619 它要能读取数据 329 00:13:11,619 --> 00:13:12,418 调软件 330 00:13:12,418 --> 00:13:13,198 跑流程 331 00:13:13,198 --> 00:13:14,859 还要知道哪些文件能看 332 00:13:14,859 --> 00:13:16,629 哪些系统不能碰 333 00:13:16,629 --> 00:13:19,809 这就是为什么英伟达要推智能体相关工具 334 00:13:19,809 --> 00:13:22,529 本质上他不是只想让AI更聪明 335 00:13:22,529 --> 00:13:24,929 而是要让智能体真正进入企业系统 336 00:13:24,929 --> 00:13:26,190 变成可以被管理 337 00:13:26,190 --> 00:13:26,889 被授权 338 00:13:26,889 --> 00:13:29,169 被监督的企业级基础设施 339 00:13:29,169 --> 00:13:30,580 重点来了 340 00:13:30,580 --> 00:13:34,299 如果智能体真的进入企业软件公司不是被消灭 341 00:13:34,299 --> 00:13:36,080 而是会被重新调用 342 00:13:36,080 --> 00:13:37,419 因为智能体要干活 343 00:13:37,419 --> 00:13:38,700 离不开三样东西 344 00:13:38,700 --> 00:13:40,649 数据权限流程 345 00:13:40,649 --> 00:13:42,070 谁掌握这些入口 346 00:13:42,070 --> 00:13:43,830 谁就是智能体时代的受益者 347 00:13:43,830 --> 00:13:47,470 所以这条线第一看企业软件和应用层入口 348 00:13:47,470 --> 00:13:49,269 NOW最直接 349 00:13:49,269 --> 00:13:50,230 客服流程 350 00:13:50,230 --> 00:13:51,110 审批流程 351 00:13:51,110 --> 00:13:54,740 员工服务都是智能体最容易先接管的场景 352 00:13:54,740 --> 00:13:59,240 NOW和英伟达合作推进企业自制智能体 353 00:13:59,240 --> 00:14:01,179 重点就是让智能体在有权限 354 00:14:01,179 --> 00:14:01,820 有规则 355 00:14:01,820 --> 00:14:03,659 有治理的环境里工作 356 00:14:03,659 --> 00:14:04,620 对于NOW 357 00:14:04,620 --> 00:14:06,360 我们早在5月14日 358 00:14:06,360 --> 00:14:08,539 我们美股投资网的AI量化模型 359 00:14:08,539 --> 00:14:11,929 就在88.3美元附近选出了NOW 360 00:14:11,929 --> 00:14:14,289 并已通知VIP社群建仓 361 00:14:14,289 --> 00:14:17,370 截止本周一最高涨到138美元 362 00:14:17,370 --> 00:14:18,830 涨幅55% 363 00:14:18,830 --> 00:14:23,429 SAAP掌握企业财务采购库存智能体 364 00:14:23,429 --> 00:14:25,429 如果要真正进入企业经营 365 00:14:25,429 --> 00:14:27,730 绕不开ERPCRM 366 00:14:27,730 --> 00:14:28,750 受益于销售 367 00:14:28,750 --> 00:14:30,639 客服营销自动化 368 00:14:30,639 --> 00:14:32,799 这里还要单独提一个DOCN 369 00:14:32,799 --> 00:14:33,980 它不是应用软件 370 00:14:33,980 --> 00:14:35,299 而是开发者云 371 00:14:35,299 --> 00:14:36,820 未来智能体爆发后 372 00:14:36,820 --> 00:14:38,600 不只是大企业会用 373 00:14:38,600 --> 00:14:42,080 很多中小企业创业团队也会做客服 374 00:14:42,080 --> 00:14:44,240 电商运营营销自动化 375 00:14:44,240 --> 00:14:45,000 数据分析 376 00:14:45,000 --> 00:14:46,450 这类垂直智能体 377 00:14:46,450 --> 00:14:49,779 这些团队不一定需要复杂的大型AI云 378 00:14:49,779 --> 00:14:52,519 更需要便宜简单容易部署的平台 379 00:14:52,519 --> 00:14:54,440 这就是DOOCN的机会 380 00:14:54,440 --> 00:14:56,059 早在今年3月21日 381 00:14:56,059 --> 00:14:57,779 我们在美股投资网的视频 382 00:14:57,779 --> 00:15:00,200 2026年AI智能体爆发年 383 00:15:00,200 --> 00:15:02,519 极度利好两家美股公司中 384 00:15:02,519 --> 00:15:05,620 就把DOOCN的投资逻辑讲得非常清楚 385 00:15:05,620 --> 00:15:08,360 当时股价呢仅82.48美元 386 00:15:08,360 --> 00:15:12,139 而本周一已经冲到175.6美元 387 00:15:12,139 --> 00:15:14,480 又是一笔翻倍的涨幅 388 00:15:14,480 --> 00:15:18,139 第二呢看数据平台SNOW的逻辑 389 00:15:18,139 --> 00:15:19,139 不是蹭AI 390 00:15:19,139 --> 00:15:21,019 而是企业AI要落地 391 00:15:21,019 --> 00:15:23,940 必须先有干净统一可治理的数据 392 00:15:23,940 --> 00:15:26,470 没有数据智能体就是空转 393 00:15:26,470 --> 00:15:30,350 SNOW最近上调全年产品收入预期 394 00:15:30,350 --> 00:15:34,309 还和AWS签了5年60亿美元合作协议 395 00:15:34,309 --> 00:15:37,330 说明企业AIWORKLOUD已经开始 396 00:15:37,330 --> 00:15:39,549 真实拉动数据云消耗 397 00:15:39,549 --> 00:15:41,009 大家应该还记得 398 00:15:41,009 --> 00:15:44,370 我们之前在视频被美股市场错杀 399 00:15:44,370 --> 00:15:47,610 SARS公司是否因AH能体迎来大反转 400 00:15:47,610 --> 00:15:50,730 就重点提醒关注SNOW 401 00:15:50,730 --> 00:15:54,099 当时股价只有139.74美元 402 00:15:54,099 --> 00:15:56,379 截至周一股价280美元 403 00:15:56,379 --> 00:15:57,359 整整翻倍 404 00:15:57,359 --> 00:16:01,600 PLTR则更适合政府大企业和高安全场景 405 00:16:01,600 --> 00:16:03,480 agent要做复杂决策 406 00:16:03,480 --> 00:16:06,279 必须接入真实业务数据和安全系统 407 00:16:06,279 --> 00:16:08,360 这正是PLTR的位置 408 00:16:08,360 --> 00:16:10,610 第三看专业工具 409 00:16:10,610 --> 00:16:14,470 CDNS和SNPS是芯片设计软件 410 00:16:14,470 --> 00:16:16,269 AI芯片迭代越快 411 00:16:16,269 --> 00:16:19,110 设计验证仿真的需求越高 412 00:16:19,110 --> 00:16:21,919 EDA工具反而会被调用的更频繁 413 00:16:21,919 --> 00:16:24,320 CRWD是安全agent 414 00:16:24,320 --> 00:16:25,259 进入企业以后 415 00:16:25,259 --> 00:16:27,059 本质上多了一批数字员工 416 00:16:27,059 --> 00:16:28,580 这些数字员工有身份 417 00:16:28,580 --> 00:16:29,320 有权限 418 00:16:29,320 --> 00:16:30,620 会调用API 419 00:16:30,620 --> 00:16:31,849 会访问数据 420 00:16:31,849 --> 00:16:33,548 企业必须监控他们 421 00:16:33,548 --> 00:16:34,229 限制他们 422 00:16:34,229 --> 00:16:35,610 保护他们 423 00:16:35,610 --> 00:16:38,850 西门子和达索系统是工业软件agent的 424 00:16:38,850 --> 00:16:40,590 金融制造业是辅助设计 425 00:16:40,590 --> 00:16:42,370 仿真排产维护 426 00:16:42,370 --> 00:16:44,940 这些都离不开专业工业软件 427 00:16:44,940 --> 00:16:46,559 但这还不是全部 428 00:16:46,559 --> 00:16:49,059 如果智能体真的大规模跑起来 429 00:16:49,059 --> 00:16:50,740 受益的不只是软件公司 430 00:16:50,740 --> 00:16:53,470 底层基础设施也会继续吃紧 431 00:16:53,470 --> 00:16:56,809 因为智能体本质上会带来更多推理需求 432 00:16:56,809 --> 00:16:59,990 每一个智能体都要调用模型读取数据 433 00:16:59,990 --> 00:17:01,710 执行任务返回结果 434 00:17:01,710 --> 00:17:03,309 企业部署的智能体越多 435 00:17:03,309 --> 00:17:04,910 背后的算力存储 436 00:17:04,910 --> 00:17:08,250 网络连接和云基础设施需求就越大 437 00:17:08,250 --> 00:17:12,029 这就是为什么marvel和NBAS也要放进这条线里看 438 00:17:12,029 --> 00:17:14,609 先看marvel mao不是软件公司 439 00:17:14,609 --> 00:17:16,618 它卡的是数据中心互联 440 00:17:16,618 --> 00:17:17,919 未来智能体越多 441 00:17:17,919 --> 00:17:19,898 数据中心内部服务器之间 442 00:17:19,898 --> 00:17:20,898 机架之间 443 00:17:20,898 --> 00:17:23,710 GPU之间的数据传输压力就越大 444 00:17:23,710 --> 00:17:27,499 AI工厂不是只有GPU网络能不能撑住 445 00:17:27,499 --> 00:17:29,219 数据能不能低延迟流动 446 00:17:29,219 --> 00:17:30,818 直接决定系统效率 447 00:17:30,818 --> 00:17:34,279 这就是黄仁勋公开力挺marvel的核心原因 448 00:17:34,279 --> 00:17:35,440 更关键的是 449 00:17:35,440 --> 00:17:36,799 mono自己预计 450 00:17:36,799 --> 00:17:37,839 定制芯片业务 451 00:17:37,839 --> 00:17:40,900 到2029财年有望超过100亿美元 452 00:17:40,900 --> 00:17:42,160 这不是一句口号 453 00:17:42,160 --> 00:17:43,920 而是真实的市场空间 454 00:17:43,920 --> 00:17:46,019 再说我们的2026年必买股 455 00:17:46,019 --> 00:17:46,880 Nbs 456 00:17:46,880 --> 00:17:49,220 NBS不是普通云服务商 457 00:17:49,220 --> 00:17:51,180 而是AI云基础设施平台 458 00:17:51,180 --> 00:17:52,759 未来企业要跑智能体 459 00:17:52,759 --> 00:17:54,440 不一定都自己买GPU 460 00:17:54,440 --> 00:17:56,880 建机房或管理运维 461 00:17:56,880 --> 00:17:59,960 很多公司要直接租用NBS的AI云算力 462 00:17:59,960 --> 00:18:02,779 而且英伟达已经和NBS战略合作 463 00:18:02,779 --> 00:18:05,529 要建设下一代超大规模AI云 464 00:18:05,529 --> 00:18:07,170 并计划到2030年 465 00:18:07,170 --> 00:18:09,509 部署超过5G瓦的算力系统 466 00:18:09,509 --> 00:18:10,309 也就是说 467 00:18:10,309 --> 00:18:12,430 NBAS不只是蹭英伟达热度 468 00:18:12,430 --> 00:18:14,490 而是真正承接企业训练 469 00:18:14,490 --> 00:18:16,480 推理和智能体部署的需求 470 00:18:16,480 --> 00:18:17,599 我们美股投资网 471 00:18:17,599 --> 00:18:20,900 今年3月底就在VIP社群提示买入NBS 472 00:18:20,900 --> 00:18:22,819 当时价格只有96美元 473 00:18:22,819 --> 00:18:24,819 到今天已经快翻三倍 474 00:18:24,819 --> 00:18:27,960 我们仍然会坚定持有好了 475 00:18:27,960 --> 00:18:28,900 由于时间关系 476 00:18:28,900 --> 00:18:30,619 今天就先给大家讲到这里 477 00:18:30,619 --> 00:18:31,980 但我可以提前说一句 478 00:18:31,980 --> 00:18:33,579 华文轩这次台北演讲里 479 00:18:33,579 --> 00:18:36,329 真正值得深挖的不止今天这两条线 480 00:18:36,329 --> 00:18:37,470 还有第三条线 481 00:18:37,470 --> 00:18:38,529 我认为更狠 482 00:18:38,529 --> 00:18:40,130 它可能不是短期炒作 483 00:18:40,130 --> 00:18:44,240 而是未来10年AI产业重新分配利润的一条主线 484 00:18:44,240 --> 00:18:46,380 如果这期视频点赞超过2000 485 00:18:46,380 --> 00:18:48,920 我会继续把第三条线单独做一期 486 00:18:48,920 --> 00:18:50,319 给大家讲清楚 487 00:18:50,319 --> 00:18:53,740 如果你不想等机会涨起来以后才看到 488 00:18:53,740 --> 00:18:56,490 欢迎加入我们美股投资网VIP社群 489 00:18:56,490 --> 00:19:01,509 像s NO w d o c n NO w marvel nbas 490 00:19:01,509 --> 00:19:02,170 这些机会 491 00:19:02,170 --> 00:19:04,150 我们不是等新闻报了才说 492 00:19:04,150 --> 00:19:06,349 而是在逻辑刚冒头的时候 493 00:19:06,349 --> 00:19:08,440 就已经开始提前推演 494 00:19:08,440 --> 00:19:09,559 美股投资网 495 00:19:09,559 --> 00:19:13,619 被粉丝们称为最具前瞻性的美股研究机构 496 00:19:13,619 --> 00:19:14,700 谢谢大家支持 497 00:19:14,700 --> 00:19:16,000 我们下期再见
评论采集失败
采集完成，该视频无弹幕