498 lines
30 KiB
HTML
498 lines
30 KiB
HTML
<!DOCTYPE html>
|
||
<html lang="zh-CN" data-theme="dark">
|
||
<head>
|
||
<meta charset="UTF-8">
|
||
<meta name="viewport" content="width=device-width, initial-scale=1.0">
|
||
<title>深度研报：MOE模型 (Mixture of Experts)</title>
|
||
<script src="https://cdn.tailwindcss.com"></script>
|
||
<link href="https://cdn.jsdelivr.net/npm/daisyui@4.10.1/dist/full.min.css" rel="stylesheet" type="text/css" />
|
||
<script defer src="https://cdn.jsdelivr.net/npm/alpinejs@3.x.x/dist/cdn.min.js"></script>
|
||
<script src="https://cdn.jsdelivr.net/npm/echarts@5.5.0/dist/echarts.min.js"></script>
|
||
<style>
|
||
@import url('https://fonts.googleapis.com/css2?family=Manrope:wght@300;400;500;700&display=swap');
|
||
|
||
body {
|
||
font-family: 'Manrope', sans-serif;
|
||
background-color: #05050A;
|
||
color: #E0E0E0;
|
||
overflow-x: hidden;
|
||
position: relative;
|
||
}
|
||
|
||
.glass-card {
|
||
background: rgba(10, 10, 15, 0.5);
|
||
backdrop-filter: blur(24px);
|
||
-webkit-backdrop-filter: blur(24px);
|
||
border: 1px solid rgba(255, 255, 255, 0.1);
|
||
border-radius: 2rem; /* 极致圆角 */
|
||
transition: all 0.3s ease;
|
||
}
|
||
|
||
.glass-card:hover {
|
||
border-color: rgba(0, 255, 255, 0.3);
|
||
transform: translateY(-5px);
|
||
box-shadow: 0 10px 30px rgba(0, 255, 255, 0.05);
|
||
}
|
||
|
||
.glow-bg {
|
||
position: fixed;
|
||
top: 0;
|
||
left: 0;
|
||
width: 100vw;
|
||
height: 100vh;
|
||
z-index: -1;
|
||
background: radial-gradient(circle at 20% 20%, rgba(0, 150, 255, 0.15) 0%, transparent 30%),
|
||
radial-gradient(circle at 80% 70%, rgba(255, 0, 150, 0.15) 0%, transparent 30%);
|
||
animation: float 20s ease-in-out infinite;
|
||
}
|
||
|
||
@keyframes float {
|
||
0% { transform: translate(0px, 0px); }
|
||
50% { transform: translate(30px, -40px); }
|
||
100% { transform: translate(0px, 0px); }
|
||
}
|
||
|
||
h1, h2, h3 {
|
||
color: #FFFFFF;
|
||
font-weight: 700;
|
||
}
|
||
|
||
.bento-grid {
|
||
display: grid;
|
||
gap: 1.5rem;
|
||
grid-template-columns: repeat(12, 1fr);
|
||
grid-auto-rows: auto;
|
||
}
|
||
|
||
.bento-item-1 { grid-column: span 12; md:grid-column: span 8; grid-row: span 2; }
|
||
.bento-item-2 { grid-column: span 12; md:grid-column: span 4; grid-row: span 1; }
|
||
.bento-item-3 { grid-column: span 12; md:grid-column: span 4; grid-row: span 1; }
|
||
.bento-item-4 { grid-column: span 12; md:grid-column: span 6; grid-row: span 1; }
|
||
.bento-item-5 { grid-column: span 12; md:grid-column: span 6; grid-row: span 1; }
|
||
.bento-item-6 { grid-column: span 12; grid-row: span 1; }
|
||
.bento-item-7 { grid-column: span 12; md:grid-column: span 7; grid-row: span 2; }
|
||
.bento-item-8 { grid-column: span 12; md:grid-column: span 5; grid-row: span 2; }
|
||
|
||
@media (min-width: 768px) {
|
||
.bento-item-1 { grid-column: span 8; grid-row: span 2; }
|
||
.bento-item-2 { grid-column: span 4; grid-row: span 1; }
|
||
.bento-item-3 { grid-column: span 4; grid-row: span 1; }
|
||
.bento-item-4 { grid-column: span 6; grid-row: span 1; }
|
||
.bento-item-5 { grid-column: span 6; grid-row: span 1; }
|
||
.bento-item-6 { grid-column: span 12; grid-row: span 1; }
|
||
.bento-item-7 { grid-column: span 7; grid-row: span 2; }
|
||
.bento-item-8 { grid-column: span 5; grid-row: span 2; }
|
||
}
|
||
|
||
.text-cyan-glow {
|
||
color: #00FFFF;
|
||
text-shadow: 0 0 8px rgba(0, 255, 255, 0.7);
|
||
}
|
||
|
||
.collapse-title {
|
||
font-weight: 500;
|
||
}
|
||
.daisy-table {
|
||
width: 100%;
|
||
border-collapse: collapse;
|
||
}
|
||
.daisy-table th, .daisy-table td {
|
||
padding: 1rem 1.5rem;
|
||
text-align: left;
|
||
border-bottom: 1px solid rgba(255, 255, 255, 0.1);
|
||
}
|
||
.daisy-table th {
|
||
color: #FFFFFF;
|
||
font-weight: 500;
|
||
background-color: rgba(255, 255, 255, 0.05);
|
||
}
|
||
.daisy-table tr:last-child td {
|
||
border-bottom: none;
|
||
}
|
||
.daisy-table tr:hover {
|
||
background-color: rgba(255, 255, 255, 0.03);
|
||
}
|
||
|
||
</style>
|
||
</head>
|
||
<body class="min-h-screen">
|
||
<div class="glow-bg"></div>
|
||
|
||
<div class="container mx-auto px-4 sm:px-6 lg:px-8 py-12 md:py-20">
|
||
|
||
<header class="text-center mb-16">
|
||
<p class="text-sm font-light text-cyan-400 mb-2">北京价值前沿科技有限公司 AI投研agent：“价小前投研”</p>
|
||
<h1 class="text-4xl md:text-6xl font-bold tracking-tight bg-clip-text text-transparent bg-gradient-to-r from-white to-cyan-400">
|
||
MOE模型 (Mixture of Experts)
|
||
</h1>
|
||
<p class="mt-4 max-w-3xl mx-auto text-lg text-gray-300">
|
||
突破算力与成本瓶颈，开启万亿参数时代，重塑AI商业格局的核心技术路径。
|
||
</p>
|
||
</header>
|
||
|
||
<main class="bento-grid">
|
||
|
||
<!-- 核心观点摘要 -->
|
||
<section class="bento-item-1 p-8 glass-card flex flex-col">
|
||
<h2 class="text-2xl mb-4 text-cyan-glow">核心观点摘要</h2>
|
||
<p class="text-gray-300 leading-relaxed flex-grow">
|
||
MOE模型是当前AI大模型突破算力与成本瓶颈、实现参数规模跃迁（万亿级）的核心技术路径。其本质是通过“稀疏激活”机制，以远低于稠密模型的计算成本实现更强的模型能力。目前，该概念正从技术竞赛的“军备竞赛”阶段，快速过渡到由成本下降驱动的商业化应用爆发前期。核心驱动力在于极致的“性价比”，未来潜力在于重塑AI应用层的成本结构和商业模式，尤其是在国产算力生态中扮演着关键的战略角色。
|
||
</p>
|
||
</section>
|
||
|
||
<!-- 2024年发展时间轴 -->
|
||
<section class="bento-item-2 p-8 glass-card">
|
||
<h3 class="text-xl mb-4 font-bold">2024：爆发与竞赛之年</h3>
|
||
<div id="timeline-chart" class="w-full h-64"></div>
|
||
</section>
|
||
|
||
<!-- API成本对比 -->
|
||
<section class="bento-item-3 p-8 glass-card">
|
||
<h3 class="text-xl mb-4 font-bold">颠覆性成本优势</h3>
|
||
<div id="cost-chart" class="w-full h-64"></div>
|
||
</section>
|
||
|
||
<!-- 关键催化剂与未来路径 -->
|
||
<section class="bento-item-6 p-8 glass-card">
|
||
<h2 class="text-2xl mb-4 text-cyan-glow">关键催化剂与未来发展路径</h2>
|
||
<div class="grid md:grid-cols-2 gap-6">
|
||
<div>
|
||
<h3 class="font-semibold text-lg text-white mb-2">近期催化剂 (3-6个月)</h3>
|
||
<ul class="list-disc list-inside space-y-2 text-gray-300">
|
||
<li><strong>价格战深化与应用爆发：</strong>以DeepSeek为首的低价API引发云厂商全面跟进，催生爆款AI原生应用。</li>
|
||
<li><strong>下一代架构发布：</strong>阿里Qwen3-Next、字节UltraMem等新架构发布，树立性能新标杆。</li>
|
||
<li><strong>国产算力标杆落地：</strong>基于昇腾等国产平台的MoE模型在金融、政务等领域斩获重大商业合同。</li>
|
||
</ul>
|
||
</div>
|
||
<div>
|
||
<h3 class="font-semibold text-lg text-white mb-2">长期发展路径</h3>
|
||
<ul class="list-disc list-inside space-y-2 text-gray-300">
|
||
<li><strong>2024-2025 (架构优化期):</strong> 竞争聚焦于路由算法、训练框架及软硬件协同优化。</li>
|
||
<li><strong>2025-2027 (垂直深化期):</strong> 与多模态、具身智能深度融合，涌现大量行业垂直MoE模型。</li>
|
||
<li><strong>2027+ (架构混合期):</strong> MoE与Agent Swarm等架构混合，实现更高阶的智能涌现。</li>
|
||
</ul>
|
||
</div>
|
||
</div>
|
||
</section>
|
||
|
||
<!-- 核心逻辑与市场认知分析 -->
|
||
<section class="bento-item-7 p-8 glass-card">
|
||
<h2 class="text-2xl mb-4 text-cyan-glow">核心逻辑与预期差分析</h2>
|
||
<div class="space-y-6 text-gray-300 leading-relaxed">
|
||
<div>
|
||
<h3 class="font-semibold text-lg text-white mb-2">三大核心驱动力</h3>
|
||
<ol class="list-decimal list-inside space-y-1">
|
||
<li><strong>技术驱动的成本下降：</strong>通过“稀疏激活”解耦参数规模与计算量，用稠密模型约1/6的计算成本实现同等性能。</li>
|
||
<li><strong>商业化落地的现实需求：</strong>颠覆性降价（如DeepSeek）破除AI应用成本壁垒，推动规模化落地。</li>
|
||
<li><strong>国产算力生态的战略选择：</strong>在高端算力受限背景下，以MoE架构最大化利用现有算力，成为关键战略路径。</li>
|
||
</ol>
|
||
</div>
|
||
<div>
|
||
<h3 class="font-semibold text-lg text-white mb-2">市场热度与预期差</h3>
|
||
<p>市场对MoE降本增效前景普遍乐观，但存在三大预期差：</p>
|
||
<ul class="list-disc list-inside mt-2 space-y-2">
|
||
<li><strong class="text-cyan-400">真实竞争力 vs. 宣传叙事：</strong>Meta的Llama 405B（非MoE）性能匹敌国产MoE模型，暗示国内模型底层能力或有差距。“参数虚胖”可能掩盖真实性能，华为与阿里模型“惊人一致”的争议也暗示创新不足。</li>
|
||
<li><strong class="text-cyan-400">工程复杂度的低估：</strong>市场可能忽略训练稳定性、负载均衡、路由策略等核心工程难题，低估了高质量MoE模型的部署门槛。</li>
|
||
<li><strong class="text-cyan-400">受益环节的认知模糊：</strong>市场未能清晰区分“模型开发者”（华为）、“模型应用者”（汤姆猫）和“基础设施提供商”（瑞斯康达、恒为科技）三者逻辑的纯粹性与弹性差异。</li>
|
||
</ul>
|
||
</div>
|
||
</div>
|
||
</section>
|
||
|
||
<!-- 产业链与核心公司剖析 -->
|
||
<section class="bento-item-8 p-8 glass-card">
|
||
<h2 class="text-2xl mb-4 text-cyan-glow">产业链与核心玩家</h2>
|
||
<div class="space-y-6">
|
||
<div>
|
||
<h3 class="font-semibold text-lg text-white mb-2">产业链图谱</h3>
|
||
<ul class="space-y-2 text-gray-300">
|
||
<li><strong>上游 (AI基础设施):</strong> <span class="font-semibold text-white">算力芯片</span> (昇腾, 海光), <span class="font-semibold text-white">网络设备 (交换机/光交换)</span> (瑞斯康达, 共进股份, 恒为科技)。</li>
|
||
<li><strong>中游 (模型开发与服务):</strong> <span class="font-semibold text-white">第一梯队</span> (华为, 阿里, 腾讯, DeepSeek), <span class="font-semibold text-white">垂直领域</span> (科大讯飞, 蚂蚁集团)。</li>
|
||
<li><strong>下游 (模型应用):</strong> <span class="font-semibold text-white">边缘计算</span> (云天励飞), <span class="font-semibold text-white">垂直行业</span> (东方通, 汤姆猫)。</li>
|
||
</ul>
|
||
</div>
|
||
<div>
|
||
<h3 class="font-semibold text-lg text-white mb-2">核心玩家对比</h3>
|
||
<div class="space-y-3">
|
||
<p><strong class="text-white">华为:</strong> 逻辑最“硬”，全栈自主可控（昇腾+盘古），契合国产替代战略。</p>
|
||
<p><strong class="text-white">DeepSeek:</strong> 逻辑最“纯”，算法创新（MLA）与极致商业化效率（低价API）的典范。</p>
|
||
<p><strong class="text-white">阿里巴巴:</strong> 逻辑最“开放”，持续开源Qwen系列，构建开发者生态。</p>
|
||
<p><strong class="text-white">字节跳动:</strong> 逻辑最“务实”，聚焦训练成本与推理效率优化（COMET, UltraMem）。</p>
|
||
</div>
|
||
</div>
|
||
</div>
|
||
</section>
|
||
|
||
<!-- 深度资料库 -->
|
||
<section class="bento-item-4 p-8 glass-card" x-data="{ tab: 'news' }">
|
||
<h2 class="text-2xl mb-4 text-cyan-glow">深度资料库</h2>
|
||
<div class="tabs tabs-boxed bg-black/20 mb-4">
|
||
<a class="tab" :class="{'tab-active': tab === 'news'}" @click="tab = 'news'">新闻洞察</a>
|
||
<a class="tab" :class="{'tab-active': tab === 'roadshow'}" @click="tab = 'roadshow'">路演纪要</a>
|
||
<a class="tab" :class="{'tab-active': tab === 'report'}" @click="tab = 'report'">研报精粹</a>
|
||
</div>
|
||
<div class="max-h-96 overflow-y-auto pr-2">
|
||
<div x-show="tab === 'news'" class="space-y-3 text-gray-300">
|
||
<p><strong>技术趋势:</strong> MoE被定位为Transformer架构优化的主流路线，能以更低计算需求大幅提升模型能力，有望成为全球主流技术。</p>
|
||
<p><strong>华为 (盘古):</strong> 发布准万亿参数盘古Ultra MoE，完全基于昇腾平台训练，实现全栈自主可控。同时发布盘古大模型5.5，覆盖多种MoE架构。</p>
|
||
<p><strong>字节跳动 (豆包):</strong> 开源训练优化技术COMET，效率提升1.7倍；并提出超越MoE的UltraMem架构，推理速度提升2-6倍。</p>
|
||
<p><strong>阿里巴巴 (通义千问):</strong> 发布Qwen2.5-Max，并开源Qwen3系列，包含首个超2000亿级MoE模型，实现以10%激活参数超越前代72B模型。</p>
|
||
<p><strong>腾讯 (混元):</strong> 开源混元Large（389B参数）和Hunyuan-A13B，探索混合推理。</p>
|
||
<p><strong>DeepSeek:</strong> 作为MoE典范，V2引入MLA算法提升推理效率，并开源通信库DeepEP解决通信瓶颈。</p>
|
||
<p><strong>科大讯飞 & 蚂蚁集团:</strong> 均在国产算力上取得突破，科大讯飞与华为合作攻克推理难题，蚂蚁集团验证了国产GPU训练3000亿参数MoE的可行性。</p>
|
||
</div>
|
||
<div x-show="tab === 'roadshow'" class="space-y-3 text-gray-300">
|
||
<p><strong>核心优势:</strong> 核心是“动态路由”，能显著降低推理成本（至稠密模型1/6-1/8），提升训练效率，平衡性能与效率。</p>
|
||
<p><strong>技术挑战:</strong> 训练稳定性、显存优化（可能牺牲速度）、路由精度、负载均衡是核心难题。</p>
|
||
<p><strong>MoE vs. Dense:</strong> Llama3采用Dense架构并表现优异，显示MoE并非唯一路径。Dense在长文本处理上更优，MoE更适合通用推理，两者是技术战略选择。</p>
|
||
<p><strong>国内外玩家:</strong> 国外OpenAI(GPT-4)、Google(Gemini)是先行者；国内阶跃星辰、阿里、华为、字节等均已布局，但多数仍处跟随阶段。</p>
|
||
<p><strong>未来趋势:</strong> 升级方向为门控精度、专家规模差异化、训练稳定性。稀疏化将从层内扩展到层间（MOLD架构）和模型级（Coe方案）。</p>
|
||
</div>
|
||
<div x-show="tab === 'report'" class="space-y-3 text-gray-300">
|
||
<p><strong>技术原理:</strong> 核心是通过门控函数（路由）稀疏激活部分专家网络，在扩大总参数的同时降低单次计算量。负载均衡是路由策略关键。</p>
|
||
<p><strong>突破万亿参数的关键路径:</strong> MoE是当前大模型突破万亿参数规模，同时有效控制训练和部署成本的核心技术。</p>
|
||
<p><strong>DeepSeek-V2:</strong> 2360亿参数，采用自研DeepSeekMoE和MLA架构，中文能力超GPT-4，API定价仅为GPT-4-Turbo的近百分之一，引爆商业化。</p>
|
||
<p><strong>华为Pangu Ultra MoE:</strong> 7180亿参数，基于昇腾平台，通过DSSN架构等创新解决训练稳定性难题，标志“国产算力+国产模型”全流程自主可控。</p>
|
||
<p><strong>腾讯混元Large:</strong> 提出MoE的Scaling Law，采用混合路由、专家回收等高级策略，支持256K长文本。</p>
|
||
<p><strong>行业影响:</strong> 凭借“参数更多、能力更强、成本更低”的特点，推动下游应用商业化爆发，催生对AI算力基础设施（尤其网络设备）的更高需求。</p>
|
||
</div>
|
||
</div>
|
||
</section>
|
||
|
||
<!-- 潜在风险与挑战 -->
|
||
<section class="bento-item-5 p-8 glass-card">
|
||
<h2 class="text-2xl mb-4 text-cyan-glow">潜在风险与挑战</h2>
|
||
<div class="max-h-96 overflow-y-auto pr-2">
|
||
<ul class="list-disc list-inside space-y-3 text-gray-300">
|
||
<li><strong>技术瓶颈:</strong> 训练稳定性、推理速度与显存成本的权衡、路由精度等仍是待优化的技术难题。国内厂商可能存在“底层创新”不足的风险。</li>
|
||
<li><strong>商业化风险:</strong> 高昂的显存（VRAM）成本是商业化部署的隐性门槛。MoE并非所有场景最优解，存在市场过度泛化的乐观预期。</li>
|
||
<li><strong>竞争与路线之争:</strong> 高性能开源MoE模型（Qwen, DeepSeek）和强大的Dense模型（Llama3）将持续挤压技术实力较弱厂商的空间，主流技术路线并非一成不变。</li>
|
||
<li><strong>信息交叉验证风险:</strong> 市场宣传（“超越国际水平”）与产业实际（“Llama稠密模型可匹敌国产MoE”）存在显著差距，需警惕被过于乐观的叙事误导。</li>
|
||
</ul>
|
||
</div>
|
||
</section>
|
||
|
||
</main>
|
||
|
||
<!-- 关联个股 -->
|
||
<section class="mt-16">
|
||
<h2 class="text-3xl font-bold text-center mb-8 text-cyan-glow">产业链核心标的</h2>
|
||
<div class="glass-card overflow-x-auto">
|
||
<table class="daisy-table">
|
||
<thead>
|
||
<tr>
|
||
<th>股票名称</th>
|
||
<th>股票代码</th>
|
||
<th>核心逻辑</th>
|
||
<th>流通市值(亿元)</th>
|
||
</tr>
|
||
</thead>
|
||
<tbody>
|
||
<tr>
|
||
<td>云天励飞</td>
|
||
<td><a href="https://valuefrontier.cn/company?scode=688343" target="_blank" class="text-cyan-400 hover:underline">688343</a></td>
|
||
<td>通过能力蒸馏研发轻量化MoE大模型，支撑大模型在边缘设备的广泛落地。</td>
|
||
<td>211.5</td>
|
||
</tr>
|
||
<tr>
|
||
<td>东方通</td>
|
||
<td><a href="https://valuefrontier.cn/company?scode=300379" target="_blank" class="text-cyan-400 hover:underline">300379</a></td>
|
||
<td>发布基于MoE的深度安全大模型，在LLM+AI Agent框架下为客户快速部署AI安全能力。</td>
|
||
<td>98.4</td>
|
||
</tr>
|
||
<tr>
|
||
<td>汤姆猫</td>
|
||
<td><a href="https://valuefrontier.cn/company?scode=300459" target="_blank" class="text-cyan-400 hover:underline">300459</a></td>
|
||
<td>AI情感陪伴机器人采用MoE架构，搭载自研垂直模型并调用豆包、DeepSeek等模型能力。</td>
|
||
<td>219.8</td>
|
||
</tr>
|
||
<tr>
|
||
<td>瑞斯康达</td>
|
||
<td><a href="https://valuefrontier.cn/company?scode=603803" target="_blank" class="text-cyan-400 hover:underline">603803</a></td>
|
||
<td>MoE推动AI大模型降本，对网络通信需求剧增，交换机迎来新发展机遇。公司布局数据中心交换机等产品。</td>
|
||
<td>53.1</td>
|
||
</tr>
|
||
<tr>
|
||
<td>共进股份</td>
|
||
<td><a href="https://valuefrontier.cn/company?scode=603118" target="_blank" class="text-cyan-400 hover:underline">603118</a></td>
|
||
<td>MoE架构对网络通信提出更高要求，交换机需求受益。公司产品覆盖100G、核心交换机等。</td>
|
||
<td>101.4</td>
|
||
</tr>
|
||
</tbody>
|
||
</table>
|
||
</div>
|
||
</section>
|
||
|
||
<!-- 涨幅分析补充 -->
|
||
<section class="mt-16">
|
||
<h2 class="text-3xl font-bold text-center mb-8 text-cyan-glow">概念热点追踪</h2>
|
||
<div class="glass-card overflow-x-auto">
|
||
<table class="daisy-table">
|
||
<thead>
|
||
<tr>
|
||
<th>股票名称</th>
|
||
<th>股票代码</th>
|
||
<th>异动日期</th>
|
||
<th>涨幅</th>
|
||
<th>核心驱动逻辑</th>
|
||
</tr>
|
||
</thead>
|
||
<tbody>
|
||
<tr>
|
||
<td>恒为科技</td>
|
||
<td><a href="https://valuefrontier.cn/company?scode=603496" target="_blank" class="text-cyan-400 hover:underline">603496</a></td>
|
||
<td>2025-09-10</td>
|
||
<td class="text-green-400">+5.19%</td>
|
||
<td>核心技术突破：发布OCS光交换机方案，为MoE训练和智算超节点提供更优技术路径，精准卡位算力基建新赛道。</td>
|
||
</tr>
|
||
<tr>
|
||
<td>中际旭创</td>
|
||
<td><a href="https://valuefrontier.cn/company?scode=300308" target="_blank" class="text-cyan-400 hover:underline">300308</a></td>
|
||
<td>2025-06-12</td>
|
||
<td class="text-green-400">+7.43%</td>
|
||
<td>券商研报关注MoE架构下对光组网（800G/1.6T）需求的拉动，公司作为光通信龙头直接受益。</td>
|
||
</tr>
|
||
<tr>
|
||
<td>巨人网络</td>
|
||
<td><a href="https://valuefrontier.cn/company?scode=002558" target="_blank" class="text-cyan-400 hover:underline">002558</a></td>
|
||
<td>2025-06-12</td>
|
||
<td class="text-green-400">+9.98%</td>
|
||
<td>游戏AI大模型GiantGPT完成备案，专注游戏场景，是MoE等大模型技术在垂直应用领域的典型落地。</td>
|
||
</tr>
|
||
<tr>
|
||
<td>中科曙光</td>
|
||
<td><a href="https://valuefrontier.cn/company?scode=603019" target="_blank" class="text-cyan-400 hover:underline">603019</a></td>
|
||
<td>2025-11-06</td>
|
||
<td class="text-green-400">+7.39%</td>
|
||
<td>发布全球首个单机柜级640卡超节点，在MoE万亿参数大模型等场景性能提升30-40%，引领国产算力。</td>
|
||
</tr>
|
||
<tr>
|
||
<td>因赛集团</td>
|
||
<td><a href="https://valuefrontier.cn/company?scode=300781" target="_blank" class="text-cyan-400 hover:underline">300781</a></td>
|
||
<td>2025-07-25</td>
|
||
<td class="text-green-400">+19.99%</td>
|
||
<td>与月之暗面(Kimi)战略合作，为其提供AIGC营销内容及数据服务，深度参与大模型应用生态。</td>
|
||
</tr>
|
||
</tbody>
|
||
</table>
|
||
</div>
|
||
</section>
|
||
|
||
<footer class="text-center mt-20 text-gray-500 text-sm">
|
||
<p>北京价值前沿科技有限公司出品</p>
|
||
<p class="mt-1">本报告为AI合成数据，不构成任何投资建议，投资需谨慎。</p>
|
||
</footer>
|
||
|
||
</div>
|
||
|
||
<script>
|
||
document.addEventListener('DOMContentLoaded', function () {
|
||
// ECharts 主题
|
||
const echartsTheme = {
|
||
color: ['#00FFFF', '#FF00FF', '#00FF00', '#FFFF00', '#FF0000', '#0000FF'],
|
||
backgroundColor: 'transparent',
|
||
textStyle: {
|
||
color: '#E0E0E0'
|
||
},
|
||
title: {
|
||
textStyle: {
|
||
color: '#FFFFFF'
|
||
}
|
||
},
|
||
legend: {
|
||
textStyle: {
|
||
color: '#E0E0E0'
|
||
}
|
||
},
|
||
tooltip: {
|
||
backgroundColor: 'rgba(10, 10, 15, 0.8)',
|
||
borderColor: 'rgba(255, 255, 255, 0.1)',
|
||
textStyle: {
|
||
color: '#E0E0E0'
|
||
}
|
||
},
|
||
xAxis: {
|
||
axisLine: { lineStyle: { color: 'rgba(255, 255, 255, 0.3)' } },
|
||
axisLabel: { color: '#E0E0E0' },
|
||
splitLine: { show: false }
|
||
},
|
||
yAxis: {
|
||
axisLine: { lineStyle: { color: 'rgba(255, 255, 255, 0.3)' } },
|
||
axisLabel: { color: '#E0E0E0' },
|
||
splitLine: { lineStyle: { color: 'rgba(255, 255, 255, 0.1)' } }
|
||
}
|
||
};
|
||
echarts.registerTheme('dark_sci_fi', echartsTheme);
|
||
|
||
// Timeline Chart
|
||
var timelineChartDom = document.getElementById('timeline-chart');
|
||
if (timelineChartDom) {
|
||
var timelineChart = echarts.init(timelineChartDom, 'dark_sci_fi');
|
||
var timelineOption = {
|
||
tooltip: {
|
||
trigger: 'axis',
|
||
formatter: function (params) {
|
||
return params[0].name + '<br/>' + params[0].value;
|
||
}
|
||
},
|
||
grid: { left: '3%', right: '4%', bottom: '3%', top: '15%', containLabel: true },
|
||
xAxis: { type: 'category', data: ['Q1 技术验证', 'Q2 开源与价格战', 'Q3-Q4 国产算力突破'], axisTick: { alignWithLabel: true } },
|
||
yAxis: { type: 'value', show: false },
|
||
series: [{
|
||
name: '事件',
|
||
type: 'line',
|
||
step: 'start',
|
||
symbol: 'circle',
|
||
symbolSize: 10,
|
||
lineStyle: { color: '#00FFFF', width: 2, type: 'dashed' },
|
||
itemStyle: { color: '#00FFFF', borderColor: '#05050A', borderWidth: 2 },
|
||
data: [
|
||
{ name: 'Q1 技术验证', value: 1, label: { show: true, position: 'top', formatter: '谷歌Gemini 1.5\n阶跃星辰Step-2' } },
|
||
{ name: 'Q2 开源与价格战', value: 2, label: { show: true, position: 'top', formatter: 'DeepSeek-V2低价引爆\n阿里Qwen/腾讯混元开源' } },
|
||
{ name: 'Q3-Q4 国产算力突破', value: 3, label: { show: true, position: 'top', formatter: '华为盘古+昇腾闭环\n字节/讯飞/蚂蚁跟进' } }
|
||
]
|
||
}]
|
||
};
|
||
timelineChart.setOption(timelineOption);
|
||
}
|
||
|
||
// Cost Chart
|
||
var costChartDom = document.getElementById('cost-chart');
|
||
if(costChartDom) {
|
||
var costChart = echarts.init(costChartDom, 'dark_sci_fi');
|
||
var costOption = {
|
||
tooltip: {
|
||
trigger: 'axis',
|
||
axisPointer: { type: 'shadow' },
|
||
formatter: '{b}<br/>{c} 元/百万Tokens'
|
||
},
|
||
grid: { left: '3%', right: '4%', bottom: '3%', containLabel: true },
|
||
xAxis: {
|
||
type: 'category',
|
||
data: ['GPT-4-Turbo (输入)', 'DeepSeek-V2 (输入)'],
|
||
axisLabel: { interval: 0, rotate: 15 }
|
||
},
|
||
yAxis: { type: 'value', name: '元/百万Tokens' },
|
||
series: [{
|
||
name: 'API成本',
|
||
type: 'bar',
|
||
barWidth: '40%',
|
||
data: [
|
||
{ value: 72.5, itemStyle: { color: '#FF00FF' }, label: {show: true, position: 'top', formatter: '约100倍'} },
|
||
{ value: 1, itemStyle: { color: '#00FFFF' } }
|
||
],
|
||
label: { show: true, position: 'top', color: '#fff' }
|
||
}]
|
||
};
|
||
costChart.setOption(costOption);
|
||
}
|
||
|
||
// Resize charts on window resize
|
||
window.addEventListener('resize', function() {
|
||
if(timelineChart) timelineChart.resize();
|
||
if(costChart) costChart.resize();
|
||
});
|
||
});
|
||
</script>
|
||
</body>
|
||
</html> |