VC囤货GPU供给“氧气”,创业者真的买账吗?
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">文 | 甲子光年科技产业智库,作者|田思奇</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">文 | 甲子光年科技产业智库,作者|田思奇</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">氧气——人类生命的<span style="color: black;">基本</span>,灵感的源泉。而在AI时代,GPU<span style="color: black;">便是</span>供养每一位创业者的氧气。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">据美国科技<span style="color: black;">媒介</span>The Information在7月中旬的<span style="color: black;">报告</span>,管理着420亿美元资产的头部风投<span style="color: black;">机构</span>Andreessen Horowitz(简<span style="color: black;">叫作</span>a16z),已斥资数亿美元购买了数千块英伟达H100芯片。<span style="color: black;">得到</span>a16z投资的初创企业<span style="color: black;">能够</span><span style="color: black;">得到</span>这些芯片的<span style="color: black;">运用</span>权。该计划被a16z<span style="color: black;">叫作</span>作“氧气计划”,凸显了GPU<span style="color: black;">做为</span>核心计算资源的重要性。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">GPU的稀缺性已无需赘言,a16z<span style="color: black;">亦</span>不是<span style="color: black;">独一</span>的囤货商。自2023年<span style="color: black;">败兴</span>,VC界兴起打造GPU“蓄水池”的风潮。知名投资搭档奈特·弗里德曼(Nat Friedman)和丹尼尔·格罗斯(Daniel Gross),以及Conviction等新兴投资<span style="color: black;">公司</span>都采取了类似的做法。一场围绕计算资源的战略博弈,正在VC届悄然上演。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">市场供需的天平,仍然<span style="color: black;">处在</span>倾斜的状态。H100<span style="color: black;">这般</span>的旗舰级GPU,大多早已被科技巨头以<span style="color: black;">长时间</span>巨额订单锁定,初创企业<span style="color: black;">常常</span>要排队许久。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">另一方面,资金充裕的大型科技<span style="color: black;">机构</span>逐步深入AI投资<span style="color: black;">行业</span>,投资总额<span style="color: black;">乃至</span>超越传统风投。<span style="color: black;">这里</span>背景下,VC们<span style="color: black;">经过</span>囤积稀缺资源,<span style="color: black;">加强</span>对<span style="color: black;">优秀</span>投资对象的吸引力,<span style="color: black;">显出</span>顺理成章。毕竟,<span style="color: black;">倘若</span>既能<span style="color: black;">得到</span>充足的资金支持,又能<span style="color: black;">持有</span>数以万计的H100<span style="color: black;">做为</span>算力支撑,初创企业似乎<span style="color: black;">无</span>太多拒绝科技巨头橄榄枝的理由。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">然而「甲子光年」对话的投资人均对a16z这一模式的招揽效果提出质疑。<span style="color: black;">她们</span>更相信,无论VC规模<span style="color: black;">怎样</span>,囤卡<span style="color: black;">行径</span>都属于锦上添花,<span style="color: black;">没法</span><span style="color: black;">作为</span>决定性<span style="color: black;">原因</span>。<strong style="color: blue;">这似乎更像是a16z的<span style="color: black;">媒介</span>宣传策略,<span style="color: black;">实质</span>功效顶<span style="color: black;">大都是</span>一张不痛不痒的优惠券。</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">a16z的囤货策略</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">a16z的雄心壮志,在其对GPU集群的大规模投资中显露无疑。据<span style="color: black;">媒介</span><span style="color: black;">报告</span>,这家<span style="color: black;">危害</span>投资巨头<span style="color: black;">将来</span>会将其GPU集群扩展至2万块以上,这一规模与埃隆·马斯克的xAI<span style="color: black;">机构</span>用于训练Grok大模型的GPU数量相当,预计耗资高达50亿美元。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">Luma AI是首批<span style="color: black;">运用</span>a16z GPU资源训练模型的<span style="color: black;">机构</span>之一,这家文生视频企业今年1月刚<span style="color: black;">得到</span>a16z领投的4300万美元B轮融资。该<span style="color: black;">机构</span>计划动用约3000块GPU来训练新模型。Luma首席执行官Amit Jaan在<span style="color: black;">报告</span>中<span style="color: black;">暗示</span>,<strong style="color: blue;">尽管其他<span style="color: black;">危害</span>投资<span style="color: black;">机构</span>对Luma的估值更高,但他更看重a16z<span style="color: black;">供给</span>的GPU资源。</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">国内某VC投资人Fiona在接受「甲子光年」<span style="color: black;">拜访</span>时<span style="color: black;">暗示</span>,a16z囤积GPU的<span style="color: black;">行径</span>无疑是有益的策略。“上一个互联网时代,<span style="color: black;">有些</span>VC会<span style="color: black;">供给</span>共享办公空间<span style="color: black;">做为</span>增值服务(value add),” Fiona总结<span style="color: black;">叫作</span>,“但在当前的AI时代,团队最需要的已然不是办公场所。投资者自然要使出浑身解数,拿出新的本领继续吸引可遇不可求的优秀团队。”</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">Fiona还指出,在创业者眼中,基金的<span style="color: black;">优良</span>各有<span style="color: black;">区别</span>。<span style="color: black;">有些</span>创业者更看重基金创始人的资质,如美国红杉(Sequoia)的创始人唐·瓦伦丁(Don Valentine)在硅谷享有极高的声誉;另<span style="color: black;">有些</span>人可能更看重附加价值。“<strong style="color: blue;">GPU</strong><strong style="color: blue;">是大模型<span style="color: black;">关联</span><span style="color: black;">行业</span>创业者最需要的东西之一。</strong>面对两家同等<span style="color: black;">要求</span>的<span style="color: black;">公司</span>,其中一家说我<span style="color: black;">能够</span>把卡借给你,它的吸引力自然会更大。”</p>
<div style="color: black; text-align: left; margin-bottom: 10px;">
<a style="color: black;">
展开全文
</a>
</div>
<div style="color: black; text-align: left; margin-bottom: 10px;">
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">更重要的是,<strong style="color: blue;">囤积</strong><strong style="color: blue;">GPU</strong><strong style="color: blue;">是一种引人注目的宣传方式,能够在<span style="color: black;">媒介</span>上吸引眼球。a16z的公共宣传本身就在VC行业首屈一指</strong>,开创了VC合伙人频繁写博客、做播客的范式,并在其中引入<span style="color: black;">海量</span>对<span style="color: black;">将来</span>的分析,强调拥抱AI时代的重要性。“<span style="color: black;">她们</span><span style="color: black;">亦</span>想借此向创业者宣传,请<span style="color: black;">安心</span>拿着<span style="color: black;">咱们</span>的钱去创业吧,” Fiona认为。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">基于这些好处,a16z在GPU<span style="color: black;">行业</span>的押注显然并非独一无二。2023年,奈特·弗里德曼和丹尼尔·格罗斯就宣布<span style="color: black;">创立</span>了自己的计算集群,名叫仙女座(Andromeda),由超过4000个GPU<span style="color: black;">构成</span>。这对搭档会以<strong style="color: blue;"><span style="color: black;">小于</span>市场价格的<span style="color: black;">花费</span></strong>向其投资组合中的<span style="color: black;">机构</span><span style="color: black;">供给</span><span style="color: black;">运用</span>权。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">然而运行一个</strong><strong style="color: blue;">GPU</strong><strong style="color: blue;">集群绝非易事</strong><strong style="color: blue;">。</strong>弗里德曼和格罗斯支付了高达九位数美元的前期<span style="color: black;">花费</span>来租赁芯片,两人还亲自承担了<span style="color: black;">守护</span>的职责。“这是一项工业规模的运营。它需要熬夜、早起,周末不眠不休,”格罗斯在谈到该集群时说道,“<span style="color: black;">倘若</span>奈特和我<span style="color: black;">无</span>十年的软件和硬件项目运营经验,运营起来会很棘手。”</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">弗里德曼则对《福布斯》<span style="color: black;">暗示</span>,他<span style="color: black;">已然</span><span style="color: black;">作为</span><strong style="color: blue;">AI</strong><strong style="color: blue;">初创企业的全职芯片经纪人</strong>。“有几周我把大部分时间都花在为<span style="color: black;">大众</span>找 GPU上。”</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">国内某VC投资人Chris便从这个<span style="color: black;">方向</span>提出质疑:“<span style="color: black;">咱们</span>接触的许多创业者都<span style="color: black;">暗示</span>,<span style="color: black;">倘若</span>想用OpenAI的API,<span style="color: black;">她们</span>不直接从OpenAI接入,而是用微软Azure上的API。<strong style="color: blue;">微软自己的计算中心运行<span style="color: black;">数年</span>,能耗成本、调度优化,稳定性方面都很卓越,这才是创业者看重的。a16z的</strong><strong style="color: blue;">GPU</strong><strong style="color: blue;">集群能有多少稳定性呢?</strong>它的两位创始人<span style="color: black;">无</span>硬件背景。重新打造一个计算中心,比大厂的云<span style="color: black;">设备</span>效果还要好,这不是VC从业者<span style="color: black;">能够</span>实现的。”</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">值得一提的是,a16z在GPU<span style="color: black;">行业</span>的野心不仅限于自己购买或租赁,还<span style="color: black;">触及</span>与芯片供应商的深度合作。据《福布斯》<span style="color: black;">报告</span>,a16z正在与芯片商洽谈共建计算项目,<span style="color: black;">加强</span>其在人工智能<span style="color: black;">行业</span>的竞争力。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">然而靠囤积</strong><strong style="color: blue;">GPU</strong><strong style="color: blue;"><span style="color: black;">或</span>潜在的共建计算中心来<span style="color: black;">得到</span>投资上的竞争<span style="color: black;">优良</span>,似乎不是a16z<span style="color: black;">这般</span><span style="color: black;">持有</span>几百亿美元的头部VC需要<span style="color: black;">思虑</span>的。</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">“a16z是业内品牌效应最好的VC之一,能够投进几乎所有头部大模型<span style="color: black;">机构</span>,” Chris对「甲子光年」<span style="color: black;">暗示</span>,“<strong style="color: blue;">创业者非常<span style="color: black;">留意</span>自己拿到了谁的钱。能<span style="color: black;">得到</span>a16z<span style="color: black;">或</span>红杉资本的投资,本身就比<span style="color: black;">得到</span>算力资源重要得多。</strong><span style="color: black;">这般</span>的投资背书<span style="color: black;">能够</span><span style="color: black;">极重</span>地<span style="color: black;">提高</span><span style="color: black;">机构</span>的认可度。”</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">至于规模较小的VC,似乎更<span style="color: black;">无</span>囤卡的<span style="color: black;">道理</span></strong>:“一亿美金的小型VC,即<span style="color: black;">运用</span>管理费买到少量GPU资源,<span style="color: black;">亦</span><span style="color: black;">没法</span>在竞争中占优,” Chris<span style="color: black;">暗示</span>,“创业者不会嫌弃资源多,但<span style="color: black;">倘若</span>我能从另一家<span style="color: black;">危害</span>投资<span style="color: black;">机构</span>多融到100万美元,我当然<span style="color: black;">能够</span>不要这些硬件资源。”</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">GPU稀缺的真伪</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">倘若</span><span style="color: black;">供给</span>算力资源比给予创业者资金更重要,其<span style="color: black;">背面</span>的<span style="color: black;">规律</span>是<strong style="color: blue;">GPU</strong><strong style="color: blue;">的稀缺程度远超等值的现金</strong>。<span style="color: black;">那样</span>在2024年下半年,GPU的稀缺还是一个真命题吗?</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">自去年<span style="color: black;">败兴</span>,有些创业者把它看作稀土金属,有些则<span style="color: black;">暗示</span>获取它的难度堪比毒品——正如AI软件Hebbia首席执行官 George Sivulka所说,行业里的人们会交头接耳:“嗯,我认识一个有H100的人。”</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">亦</span><span style="color: black;">便是</span>从2023年<span style="color: black;">起始</span>,风投<span style="color: black;">公司</span>Index Ventures与甲骨文开展合作,免费向初创企业<span style="color: black;">供给</span>英伟达的芯片。该<span style="color: black;">机构</span>湾区合伙人艾琳·普赖斯-赖特(Erin Price-Wright)<span style="color: black;">亦</span><span style="color: black;">暗示</span>,获取算力资源是AI<span style="color: black;">机构</span>最大的挑战之一,<strong style="color: blue;"><span style="color: black;">针对</span>尚处<span style="color: black;">初期</span>的<span style="color: black;">机构</span><span style="color: black;">来讲</span>,<span style="color: black;">得到</span></strong><strong style="color: blue;">GPU</strong><strong style="color: blue;">尤其困难。成本不是最大的问题,但市面上超过95%的GPU容量<span style="color: black;">已然</span>分给了科技巨头。</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">这<span style="color: black;">明显</span>加快了这些<span style="color: black;">机构</span>的发展,格罗斯<span style="color: black;">叫作</span>之为“时间膨胀机制”。“你最需要的<span style="color: black;">便是</span>时间。计算集群是一种加速时间的方式,你<span style="color: black;">能够</span>将一个<span style="color: black;">本来</span>需要一年时间训练的模型在<span style="color: black;">咱们</span>的集群上用<span style="color: black;">1星期</span>时间完成。”格罗斯说。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">“<span style="color: black;">需要</span>非常高,远远超过<span style="color: black;">咱们</span>的供应。<span style="color: black;">咱们</span><span style="color: black;">每日</span>都在争分夺秒,”英伟达CEO黄仁勋在今年5月的财报会议上<span style="color: black;">暗示</span>。该<span style="color: black;">机构</span>一季度净利润同比增长628% 。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">不外</span>随着GPU短缺形势的缓解,<span style="color: black;">有些</span>投资者<span style="color: black;">起始</span><span style="color: black;">调节</span>策略。美国红杉资本并<span style="color: black;">无</span>急于买入GPU,而是在博客上<span style="color: black;">发帖</span><span style="color: black;">叫作</span>GPU短缺的高峰<span style="color: black;">已然</span>过去。<strong style="color: blue;"><span style="color: black;">倘若</span>将<span style="color: black;">报答</span>定义为赚取足够的利润来抵消投资,红杉估计生成式</strong><strong style="color: blue;">AI</strong><strong style="color: blue;">理应创造6000亿美元的收入,而要实现这种<span style="color: black;">报答</span>还需要几十年的时间。</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">文案</span>作者,身为红杉资本合伙人的戴维·卡恩 (David Cahn) 并不相信购买GPU就像修建铁路——它们会贬值,<span style="color: black;">火速</span>会过时,不会像修铁路那样带来太多的垄<span style="color: black;">判断</span>价权。他认为一种危险的“错觉”<span style="color: black;">已然</span>占据了上风:“<span style="color: black;">咱们</span>都会<span style="color: black;">火速</span>致富,<span style="color: black;">由于</span>AGI明天就要来了,我们都需要<span style="color: black;">贮存</span><span style="color: black;">独一</span>有价值的资源,那<span style="color: black;">便是</span>GPU。”</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">如下图所示,该<span style="color: black;">文案</span>的推断<span style="color: black;">规律</span>是:英伟达芯片运营的收入应为数据中心总成本的50%(另一半是能源和建筑成本);而数据中心的成本应该占AI企业为硬件<span style="color: black;">设备</span>付出的成本的50%。<span style="color: black;">亦</span><span style="color: black;">便是</span>说,AI行业的年产值应为英伟达营收的4倍。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">按照</span>英伟达最新营收来推算,AI行业的年产值应超过6000亿美元,才够覆盖数据中心等<span style="color: black;">基本</span><span style="color: black;">设备</span>的开销。但<span style="color: black;">日前</span>的<span style="color: black;">状况</span>并非如此,<span style="color: black;">因此</span><span style="color: black;">文案</span>结论是AI行业<span style="color: black;">泡泡</span>加剧。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><img src="https://q0.itc.cn/q_70/images03/20240729/ede7e109ddbf4eeeae5b845496d8c22f.png" style="width: 50%; margin-bottom: 20px;"></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">表格单位:十亿美元(billion),<span style="color: black;">照片</span><span style="color: black;">源自</span>:红杉资本</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">Databricks首席执行官阿里·戈德斯(Ali Ghodsi)<span style="color: black;">亦</span>对GPU火热的<span style="color: black;">需要</span>泼了一盆冷水。他在今年2月预测明年AI芯片的价格就会大幅下跌,重新实现供需平衡。正如人们对互联网带宽的担忧悄然殆尽,“<span style="color: black;">一样</span>的事情<span style="color: black;">亦</span>会<span style="color: black;">出现</span>在 GPU 上”。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">新兴风投Conviction去年<span style="color: black;">亦</span><span style="color: black;">起始</span>从云厂商商那里租赁GPU,以成本价转供给初创<span style="color: black;">机构</span>。但今年该风投<span style="color: black;">已然</span>减少了订单量,并将部分服务器投放到市场上出售。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">“据<span style="color: black;">咱们</span><span style="color: black;">认识</span>,<span style="color: black;">咱们</span>投资的美国创业<span style="color: black;">机构</span>在算力层面并<span style="color: black;">无</span>遇到太多问题,美国的算力稀缺似乎不在于缺乏创业<span style="color: black;">机构</span>,尤其是应用层<span style="color: black;">机构</span>所需要的少量算力。” Chris<span style="color: black;">暗示</span>,<strong style="color: blue;">“<span style="color: black;">有些</span>云厂商,<span style="color: black;">包含</span>AWS和谷歌都主动<span style="color: black;">暗示</span>可<span style="color: black;">以避免</span>费赠送<span style="color: black;">有些</span>算力试用,用完了再花钱租。</strong>租金成本不会对<span style="color: black;">机构</span><span style="color: black;">导致</span><span style="color: black;">明显</span>影响。<strong style="color: blue;">”</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">然则</span><span style="color: black;">针对</span>Luma<span style="color: black;">这般</span>急需算力的模型层<span style="color: black;">机构</span><span style="color: black;">来讲</span>,Chris认为<span style="color: black;">她们</span>的确可能正在排队,<span style="color: black;">因此</span>才需要a16z的直接支持:<strong style="color: blue;">“这就像餐厅里的大桌和小桌排队<span style="color: black;">同样</span>,模型层<span style="color: black;">机构</span>相当于在排大桌,而应用层<span style="color: black;">机构</span>排小桌,两者的排队速度<span style="color: black;">区别</span>。</strong>与大<span style="color: black;">机构</span>合作的流程<span style="color: black;">亦</span>比较长,而a16z能直接<span style="color: black;">供给</span>卡,这就比等待投资款项到账后再租用大<span style="color: black;">机构</span>的云要快得多。<strong style="color: blue;">”</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">这<span style="color: black;">亦</span>印证了a16z囤积GPU的相对<span style="color: black;">恰当</span>性:“a16z<span style="color: black;">做为</span>大体量的基金,投资标的<span style="color: black;">亦</span>都是比<span style="color: black;">很强</span>的模型层<span style="color: black;">机构</span>,确实很容易触及GPU短缺的临界点。”</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">大<span style="color: black;">都数</span>VC不会像a16z<span style="color: black;">这般</span>,直接购买<span style="color: black;">海量</span>终将过时的H100,并静静等待它烂在自己手里。<span style="color: black;">例如</span><span style="color: black;">以上</span>Index Ventures不会直接购买任何芯片,与Oracle的合作只<span style="color: black;">寓意</span>着它<span style="color: black;">表率</span>被投<span style="color: black;">机构</span>作出承诺并支付云的<span style="color: black;">花费</span>,<span style="color: black;">保证</span>创业者“在正确的时间与正确的人<span style="color: black;">创立</span>联系,以便<span style="color: black;">她们</span><span style="color: black;">得到</span>所需的资源。”</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">Fiona<span style="color: black;">亦</span><span style="color: black;">暗示</span>,她<span style="color: black;">无</span>听说国内哪家基金会主动买卡,<span style="color: black;">更加多</span>地<span style="color: black;">装扮</span>“算力<span style="color: black;">中间商</span>”的角色。“a16z<span style="color: black;">得到</span>GPU肯定要比国内的资本方便。<span style="color: black;">咱们</span>和同行<span style="color: black;">此刻</span>的策略都是和云厂商,<span style="color: black;">包含</span>亚马逊云、Google Cloud等加强联系,这些都是更方便的共享算力。”</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">OpenAI的CEO萨姆·奥尔特曼的弟弟杰克·奥特曼 (Jack Altman) <span style="color: black;">亦</span>在近期<span style="color: black;">起始</span>担任算力<span style="color: black;">中间商</span>,推出了名叫Generate的服务。<strong style="color: blue;">他掌舵的风投<span style="color: black;">机构</span> Alt Capital 会帮初创企业免费<span style="color: black;">得到</span>微软<span style="color: black;">基本</span><span style="color: black;">设备</span>的资源。</strong>据微软<span style="color: black;">机构</span>增长与生态系统副总裁安妮·珀尔(Annie Pearl)介绍,每家初创企业大约会<span style="color: black;">运用</span>两周微软的<span style="color: black;">设备</span>来训练自己的大模型,<span style="color: black;">另外</span>微软还会给每家<span style="color: black;">机构</span><span style="color: black;">供给</span>35万美元的额度来<span style="color: black;">运用</span>Azure的云服务。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">与此<span style="color: black;">同期</span>,Alt Capital将向每家初创<span style="color: black;">机构</span>投资25万美元。值得一提的是,Alt Capital的Generate服务<span style="color: black;">重点</span>和较成熟的初创<span style="color: black;">机构</span>合作,投资对象<span style="color: black;">一般</span><span style="color: black;">已然</span>完成A轮融资,金额超过1000万美元以上。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">同期</span>,VC看待创业者的方式<span style="color: black;">亦</span>有所<span style="color: black;">区别</span>。 Fiona强调,“<span style="color: black;">咱们</span><span style="color: black;">做为</span><span style="color: black;">初期</span><span style="color: black;">危害</span>投资者,还是期待创业者<span style="color: black;">能够</span>独立<span style="color: black;">有些</span>,<span style="color: black;">尽可能</span>不要<span style="color: black;">显现</span>要做这件事的核心元素不<span style="color: black;">晓得</span>怎么搞定就<span style="color: black;">起始</span>创业的<span style="color: black;">状况</span>。”</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">这不<span style="color: black;">寓意</span>着Fiona所在的<span style="color: black;">公司</span>会拒绝创业者的求助。“<span style="color: black;">咱们</span>会<span style="color: black;">回复</span>创业者提出的<span style="color: black;">详细</span><span style="color: black;">需要</span>,但<span style="color: black;">咱们</span><span style="color: black;">亦</span>会倾向于认为,<span style="color: black;">倘若</span>创业者自己<span style="color: black;">无</span>什么算力资源的<span style="color: black;">途径</span>,且算力不到位业务就<span style="color: black;">没法</span>开展,<strong style="color: blue;">那他是不是应该留在算力资源比较多的大<span style="color: black;">机构</span>内部创业,等到自己足够成熟以后再出来?</strong>”</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">Fiona还<span style="color: black;">暗示</span>,她<span style="color: black;">无</span>碰到过<span style="color: black;">必定</span>需要VC<span style="color: black;">供给</span>算力的创业者,<span style="color: black;">亦</span><span style="color: black;">无</span>创业者会仅仅<span style="color: black;">由于</span>她所在的基金不承诺帮忙找算力就拒绝她。“这多半不是他们最<span style="color: black;">期盼</span><span style="color: black;">咱们</span><span style="color: black;">帮忙</span>的方面。<span style="color: black;">咱们</span>更常遇到的<span style="color: black;">状况</span>是,<strong style="color: blue;">优秀的创业者会<span style="color: black;">期盼</span><span style="color: black;">咱们</span>这些平时见人比较多的<span style="color: black;">公司</span><span style="color: black;">帮忙</span>他找到优秀的人才。</strong>”</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">然而人才<span style="color: black;">亦</span>会<span style="color: black;">留意</span>一家创业<span style="color: black;">机构</span>的算力水平。最著名的例子之一是AI搜索引擎Perplexity创始人阿拉丁·斯里尼瓦斯(Aravind Srinivas)在今年早些时候的<span style="color: black;">抱怨</span>:“我曾试图从Meta聘请一位资深<span style="color: black;">科研</span>员,结果他说等我有了1万块H100再去找他。<span style="color: black;">思虑</span>到AI发展这么快,即使 Perplexity拿到<span style="color: black;">有些</span>英伟达的芯片,世界<span style="color: black;">亦</span>已进入下一个时代。到时候<span style="color: black;">她们</span>又会说等我有2万块H100再找他。”</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">因此呢</span>,这反映了AI投资界了另一种暗涌的趋势:既有钱<span style="color: black;">亦</span>有充足算力的大厂,<span style="color: black;">是不是</span>正逐步取代传统VC在投资界的地位?</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">大厂的FOMO</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">在微软投资OpenAI,亚马逊投资Anthropic等令人瞩目的投资<span style="color: black;">背面</span>,科技巨头在生成式AI企业中的投资,正在超过传统VC<span style="color: black;">公司</span>的投资额。外界分析认为,这些<span style="color: black;">措施</span><span style="color: black;">重点</span>出于FOMO的<span style="color: black;">心情</span>(fear of missing out,指跟不上潮流<span style="color: black;">引起</span>落后所带来的恐惧)。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">据<span style="color: black;">科研</span><span style="color: black;">机构</span>PitchBook数据,微软、谷歌和亚马逊去年达<span style="color: black;">成为了</span>一系列重磅交易,占 2023 年新兴AI<span style="color: black;">机构</span>270亿美元总融资的三分之二。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><img src="https://q9.itc.cn/q_70/images03/20240729/6402e42dec2941479b07385bc5b4167b.png" style="width: 50%; margin-bottom: 20px;"></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">2023年,科技巨头的投资把生成式人工智能融资额推至历史新高,<span style="color: black;">照片</span><span style="color: black;">源自</span>:PitchBook、金融时报</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">与此<span style="color: black;">同期</span>,佛罗里达大学金融学教授亚历杭德罗·洛佩兹-利拉(Alejandro Lopez-Lira)<span style="color: black;">暗示</span>,<strong style="color: blue;">这<span style="color: black;">一样</span>是一种阻碍初创企业竞争的手段。“这些<span style="color: black;">机构</span>能够接触到如此多的创业<span style="color: black;">机构</span>,这很可能减少<span style="color: black;">将来</span>竞争对手的数量,”洛佩兹-利拉说道。“<span style="color: black;">无</span>人想重复雅虎未能收购谷歌的错误。”</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">另一方面,构建和训练生成式人工智能工具需要巨大的算力和资金。<span style="color: black;">因此呢</span>初创<span style="color: black;">机构</span>更愿意与<span style="color: black;">同期</span><span style="color: black;">供给</span><span style="color: black;">基本</span><span style="color: black;">设备</span>的大型科技<span style="color: black;">机构</span>合作。而这又<span style="color: black;">快速</span>推高了初创<span style="color: black;">机构</span>的估值,使得传统VC更难押注<span style="color: black;">处在</span>技术前沿的标的。《金融时报》2023年底<span style="color: black;">报告</span><span style="color: black;">叫作</span>,<strong style="color: blue;"><span style="color: black;">危害</span>投资<span style="color: black;">机构</span>被迫放慢支出速度,只为适应更高的利率和投资组合估值下降的困境</strong>。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">初期</span><span style="color: black;">危害</span>投资<span style="color: black;">机构</span>Tapestry VC的创始合伙人帕特里克·墨菲 (Patrick Murphy)<span style="color: black;">暗示</span>:“即使是管理着数百亿美元资金的世界顶级<span style="color: black;">危害</span>投资<span style="color: black;">机构</span><span style="color: black;">亦</span>难以竞争。在这次AI平台的转变中,<span style="color: black;">仅有</span>百万分之一概率<span style="color: black;">显现</span>的<span style="color: black;">潜能</span><span style="color: black;">机构</span>,大部分都<span style="color: black;">已然</span>被大型科技<span style="color: black;">机构</span><span style="color: black;">霸占</span>。”</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">面对VC和大厂之间的<span style="color: black;">选取</span>,Fiona认为<span style="color: black;">详细</span>要看大厂的投资条款。“大厂的<span style="color: black;">优良</span>在于生态资源的倾斜,现成的应用场景接入,<span style="color: black;">包含</span>算力资源的优惠,这些都是大厂用来吸引优秀创业团队的常规筹码。但<span style="color: black;">咱们</span>VC<span style="color: black;">做为</span>财务投资人,<span style="color: black;">一般</span>能给创业团队非常高的自主权。”</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">更何况,大厂的战略投资可能<span style="color: black;">包含</span><span style="color: black;">有些</span>有利于大厂生态的<span style="color: black;">需要</span>。“有些团队看中大厂生态的帮衬,有些创业者相信自己不需要依附大厂,足以独当一面。这就将考验创始人<span style="color: black;">怎样</span>判断自己做的业务在生态中跟大厂的相对位置,再<span style="color: black;">思虑</span><span style="color: black;">是不是</span>拿大厂的投资,” Fiona<span style="color: black;">暗示</span>。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">Chris<span style="color: black;">亦</span>认为,大厂的投资带有战略诉求,会对创业者<span style="color: black;">导致</span><span style="color: black;">必定</span>限制。这些限制并不在于强迫创业<span style="color: black;">机构</span>做某些事情,而是可能会减少<span style="color: black;">她们</span>的部分<span style="color: black;">选取</span>。“假设今天字节跳动的豆包API价格大幅下降,<span style="color: black;">作为</span>国内价格最低的API服务。<span style="color: black;">那样</span>接受了其他大厂投资的创业<span style="color: black;">机构</span><span style="color: black;">亦</span>许就会被限制不要<span style="color: black;">运用</span>豆包API。”</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">至于大厂投资<span style="color: black;">是不是</span>旨在遏制潜在的竞争对手,Chris<span style="color: black;">无</span>给出肯定的回答,<span style="color: black;">然则</span>指出一个特征:“<strong style="color: blue;">大厂的战略投资所持有的创业<span style="color: black;">机构</span>股份比例会更大。例如阿里巴巴投资月之暗面占股份36%,而普通VC可能只会占十几个百分点,<span style="color: black;">一般</span>会分散投资物来谋求更好的财务<span style="color: black;">报答</span>。大厂的投资<span style="color: black;">能够</span>分散,<span style="color: black;">亦</span><span style="color: black;">能够</span>不分散,它不完全是为了挣钱,而是出于一个<span style="color: black;">此刻</span>可能还不<span style="color: black;">知道</span>的目的。”</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">另一个好处是,大厂投资OpenAI<span style="color: black;">这般</span>的模型层企业的<span style="color: black;">同期</span>,<span style="color: black;">亦</span><span style="color: black;">能够</span>继续加强与电力<span style="color: black;">机构</span>的合作,在美国多地<span style="color: black;">创立</span>电站和<span style="color: black;">基本</span><span style="color: black;">设备</span>,<span style="color: black;">从而</span>构建<span style="color: black;">持有</span>数十万显卡的超大规模集群。这些更加强大的计算中心<span style="color: black;">能够</span><span style="color: black;">帮忙</span>被投<span style="color: black;">机构</span>训练更强大的模型,<span style="color: black;">做为</span>配套的数据中心和计算资源<span style="color: black;">亦</span>能得到稳定地<span style="color: black;">运用</span>,从而<span style="color: black;">保证</span>投资<span style="color: black;">报答</span>,<span style="color: black;">提高</span>财务表现。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">相对<span style="color: black;">来讲</span>,VC的主营业务是赚取财务<span style="color: black;">报答</span>,而不需要关心创业<span style="color: black;">机构</span>的<span style="color: black;">详细</span>业务,<span style="color: black;">亦</span>不会期待自己<span style="color: black;">供给</span>的GPU能比得过大<span style="color: black;">机构</span>的云<span style="color: black;">设备</span>。<span style="color: black;">因此呢</span>,<span style="color: black;">危害</span>投资的条款<span style="color: black;">一般</span>比大<span style="color: black;">机构</span>的条款简单。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">Chris强调:“<strong style="color: blue;">VC<span style="color: black;">供给</span>的最有价值的东西是无形资产,<span style="color: black;">包含</span>公共关系、运营经验,营销节奏等等。</strong>难道红杉投资一个新能源<span style="color: black;">机构</span>还要给它搭配一个汽车生产线,做<span style="color: black;">一起</span>电池吗?产业链是有分工的,被投企业肯定会用宁德时代的电池。”</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">“<span style="color: black;">针对</span>创业者<span style="color: black;">来讲</span>,这些无形资产,不会<span style="color: black;">无</span>硬件资源重要。而硬件资源相<span style="color: black;">针对</span>风投的品牌影响力<span style="color: black;">来讲</span>,就更不是稀缺的资源了,” Chris总结道。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">*应受访者<span style="color: black;">需求</span>,Fiona和Chris为化名。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">参考资料:</strong><a style="color: black;"><span style="color: black;">返回<span style="color: black;">外链论坛: http://www.fok120.com</span>,查看<span style="color: black;">更加多</span></span></a></p>《Andreessen Horowitz is building a stash of more than 20,000 GPUs to win AI deals》,
The Information《AI investors are wooing startups with massive computing clusters》,
Forbes《The Desperate Hunt for the A.I. Boom’s Most Indispensable Prize》,
The New York Times《Jack Altman’s Venture Fund Launches Accelerator for Enterprise AI Startups》,
Wall Street Journal《Big Tech outspends venture capital firms in AI investment frenzy》,Financial
Times《Why Generative AI Startups Are So Dependent on Big Tech》,Inc.
</div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">责任编辑:网友投稿</span></p>
页:
[1]