<p>每个月似乎都有一家公司宣布增加其人工智能（AI）数据中心的预算。上个月，<strong>谷歌母公司</strong> 公布了今年 850 亿美元的资本支出计划，较其原先 2025 年的 750 亿美元预期有所上调。一周后，<strong>亚马逊</strong> 表示将其资本支出预算提高至 1180 亿美元，较 1000 亿美元有所增加。</p>
<p>大型科技公司的支出在未来几年只会继续上升。市场研究公司 Dell'Oro Group 预测，到 2028 年，前 10 大科技公司将在 AI 基础设施上花费超过 1 万亿美元。这一数字较今年的约 5930 亿美元有所增加。</p>
<p><strong>英伟达</strong> 至今是这些公司支出狂潮中最大的受益者之一。其 GPU 在 AI 训练和推理服务器中发挥着关键作用。对其芯片的需求推动了其销量和价格的上涨，这在过去三年的惊人收入和利润增长中得到了体现。在生成 AI 革命进行约两年半的时间里，该公司的收入同比增长了 69%，最近一个季度的调整后利润增长了 57%。</p>
<p>但英伟达过去三年的强劲业绩并不一定意味着该股票在未来三年内会成为一个绝佳投资。事实上，以下三只半导体股票目前看起来更适合投资，以利用大型科技公司持续增加的 AI 支出。</p>
<p><img src="https://imageproxy.pbkrs.com/https://g.foolcdn.com/image//query-b3A9cmVzaXplJnVybD1odHRwczovL2cuZm9vbGNkbi5jb20vZWRpdG9yaWFsL2ltYWdlcy84Mjg2NzAvZ2V0dHlpbWFnZXMtYXJ0aWZpY2lhbC1pbnRlbGxnZW5jZS1jaGlwLmpwZWcmdz03MDA?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" width="700" height="393" original-src="https://imageproxy.pbkrs.com/https://g.foolcdn.com/image//query-b3A9cmVzaXplJnVybD1odHRwczovL2cuZm9vbGNkbi5jb20vZWRpdG9yaWFsL2ltYWdlcy84Mjg2NzAvZ2V0dHlpbWFnZXMtYXJ0aWZpY2lhbC1pbnRlbGxnZW5jZS1jaGlwLmpwZWcmdz03MDA"/></p>
<p>图片来源：Getty Images。</p>
<h2>1. 马威尔科技</h2>
<p><strong>马威尔科技</strong>（MRVL 0.63%）是亚马逊和 <strong>微软</strong> 定制 AI 芯片设计的芯片制造商。这些定制芯片被称为 XPU，与 GPU 不同，XPU 是根据更精确的规格设计的，以优化大型语言模型训练或推理的性价比。</p>
<p>XPU 的机会正在迅速增长，因为大型科技公司希望使用更具成本效益的芯片，减少对英伟达的依赖。<strong>博通</strong>，另一家 XPU 制造商，预计到 2027 年，其三家定制硅客户将在芯片上花费 600 亿到 900 亿美元。马威尔预计到 2028 年，其客户将面临 550 亿美元的机会。为了更好地理解，这一数字与其过去四个季度的收入总额 65 亿美元相比。</p>
<p>马威尔似乎准备好抓住这一机会。根据富邦研究的报告，微软最近升级了其下一代 Maia300 芯片的规格。尽管这将推迟生产时间表至明年年底，但这也表明微软将迅速增加生产。</p>
<p>富邦预计 2026 年第四季度将订购 30 万到 40 万台，2027 年将保持类似的支出速度。这可能意味着单一芯片设计的年收入可达 100 亿到 120 亿美元。</p>
<p>马威尔还设计用于充分利用 GPU 和 XPU 的网络芯片。网络设备确保数据快速高效地从一个服务器移动到另一个服务器。这对于充分发挥用于训练和运行 AI 系统的昂贵服务器的处理能力至关重要。</p>
<p>如前所述，它并不是唯一一家与大型科技公司合作定制硅解决方案的公司。博通也有多项高调合同，并且与马威尔一样有许多扩展市场份额的机会。</p>
<p>但目前博通的股票对新投资者来说要贵得多。马威尔的交易价格为预期收益的 27 倍，而博通的倍数为 45 倍。这使得马威尔成为希望利用定制硅崛起的投资者的更好选择。</p>
<h2>2. 美光科技</h2>
<p><strong>美光科技</strong>（MU 3.26%）是少数几家内存芯片供应商之一。对于生成 AI 来说，最重要的内存芯片类型是称为高带宽内存（HBM）的芯片。这些芯片附加在 GPU 上，使处理器能够以最小的延迟访问数据。</p>
<p>美光在开发 HBM 芯片方面稍显滞后，但它迅速赶上了竞争，并赢得了一些大合同。<strong>AMD</strong> 选择美光为其 MI355X GPU 提供 HBM3E 芯片，这些芯片表现出良好的结果。AMD 最新的芯片可能会通过提供具有竞争力的性价比来夺取英伟达的 GPU 市场份额。</p>
<p>美光计划在明年开始生产 HBM4 芯片，此前在今年上半年向潜在客户交付了样品。管理层表示，新芯片在性能上比当前一代芯片提高了 60%，同时功耗降低了 20%。</p>
<p>HBM 芯片最近是美光业绩的最大驱动力。HBM 的销售环比增长了 50%，推动整体收入较上个季度增长了 15%，较去年增长了 51%。</p>
<p>HBM 的持续增长应推动利润率扩大。管理层的展望预计下个季度的毛利率为 42%，高于最近一个季度的 39% 和一年前的 28%。</p>
<p>公司预计在不久的将来将实现强劲的利润增长，因为 GPU 和 XPU 制造商将继续购买其内存芯片，但它面临重大风险。由于美光自己制造芯片，因此在需求增加时在工具和产能上投入巨大。但如果需求下降，可能会导致盈利能力急剧下降。这一点因内存芯片的差异化程度不高而加剧，客户可能很容易更换供应商。</p>
<p>尽管如此，该股票目前的交易价格仅为预期收益的 14 倍。在这个价格下，投资者获得了很大的潜力。如果 AI 支出如预期继续上升，这一盈利周期可能会延续到未来，使得美光科技的当前价格成为投资者的绝佳机会。</p>
<h2>3. 台湾半导体制造公司</h2>
<p><strong>台湾半导体制造公司</strong>（TSM 0.77%），简称 TSMC，是全球最大的芯片制造商。如果 Nvidia、博通、迈威尔科技或其他任何公司想要生产领先的芯片设计，他们很可能会与 TSMC 合作。这是因为它拥有世界上最先进的技术。因此，约三分之二的硅制造支出都流向了这家台湾公司。</p>
<p>TSMC 的技术领先得益于良性循环。由于该公司占据了所有支出的巨大份额，它有更多资金再投资于研发和扩建产能。随着研发支出的增加，它能够更快地开发和完善下一代工艺，并且有能力满足新芯片的需求，从而带来更多收入。</p>
<p>该公司最近收到了几条好消息，进一步提升了其在行业中的竞争地位。首先，<strong>英特尔</strong> 表示不会为其最新一代工艺提供制造服务。如果它没有为下一代工艺获得主要客户的合同，它将完全停止开发。这使得 TSMC 在提升其下一代 2nm 工艺时减少了一个竞争对手。</p>
<p>此外，TSMC 还获得了特朗普政府对外国半导体 100% 关税的豁免。这一豁免可能是由于该公司在亚利桑那州建设工厂的巨大承诺。</p>
<p>这两项措施为 TSMC 加大 2nm 工艺的生产铺平了道路，目前已有客户排队等候。该公司据报道对上一代 3nm 芯片收取了显著的溢价。结合快速的生产提升，这应该能支持强劲的毛利率，因为它迅速过渡到下一代节点。</p>
<p>尽管近期因好消息而股价上涨，TSMC 的股票目前交易价格不到未来收益的 25 倍。考虑到管理层预计基于强劲的 AI 芯片增长，从现在到 2029 年平均收入将增长 20%，这对于一家拥有巨大竞争优势的公司来说，仍然是一个很好的价格。</p>

博通

迈威尔科技

谷歌-A

亚马逊

半导体 3 倍做多 - Direxion

<p>大型科技公司预计到 2028 年将在人工智能基础设施上花费超过 1 万亿美元，较今年的 5930 亿美元大幅增加。像 Alphabet 和亚马逊这样的公司正在显著增加其资本支出。虽然英伟达从这一趋势中受益，但三只半导体股票——迈威尔科技、美光科技和博通——被认为是潜在的更好投资。迈威尔专注于定制人工智能芯片，而美光则专注于对人工智能性能至关重要的高带宽内存芯片。这两家公司都处于有利位置，可以利用对人工智能相关技术日益增长的需求</p>

<p>Every month it seems like one company or another announces it is increasing its budget for artificial intelligence (AI) data centers. Last month, <strong>Alphabet</strong> released plans for $85 billion in capital expenditures (capex) this year, up from its original 2025 outlook of $75 billion. Not to be outdone, <strong>Amazon,</strong> a week later, said it increased its capex budget to $118 billion, up from $100 billion.</p>
<div class="lb-trans"><p>每个月似乎都有一家公司宣布增加其人工智能（AI）数据中心的预算。上个月，<strong>谷歌母公司</strong> 公布了今年 850 亿美元的资本支出计划，较其原先 2025 年的 750 亿美元预期有所上调。一周后，<strong>亚马逊</strong> 表示将其资本支出预算提高至 1180 亿美元，较 1000 亿美元有所增加。</p>
</div><p>Big tech's spending is only set to climb higher over the next few years. Market researcher Dell'Oro Group forecasts the top 10 big tech companies will spend over $1 trillion on AI infrastructure in 2028. That's up from about $593 billion this year.</p>
<div class="lb-trans"><p>大型科技公司的支出在未来几年只会继续上升。市场研究公司 Dell'Oro Group 预测，到 2028 年，前 10 大科技公司将在 AI 基础设施上花费超过 1 万亿美元。这一数字较今年的约 5930 亿美元有所增加。</p>
</div><p><strong>Nvidia</strong> has been one of the biggest winners of these companies' spending spree so far. Its GPUs play a crucial role in AI training and inference servers. Demand for its chips has pushed both its volume and pricing higher, and that's reflected in its incredible revenue and profit growth over the last three years. Roughly two and a half years into the generative AI revolution, the company saw its revenue climb another 69% year over year, with profits climbing 57% on an adjusted basis during its most recent quarter.</p>
<div class="lb-trans"><p><strong>英伟达</strong> 至今是这些公司支出狂潮中最大的受益者之一。其 GPU 在 AI 训练和推理服务器中发挥着关键作用。对其芯片的需求推动了其销量和价格的上涨，这在过去三年的惊人收入和利润增长中得到了体现。在生成 AI 革命进行约两年半的时间里，该公司的收入同比增长了 69%，最近一个季度的调整后利润增长了 57%。</p>
</div><p>But Nvidia's strong results of the last three years don't necessarily make the stock a shoo-in to be a great investment over the next three years. In fact, the following three semiconductor stocks look like better investments right now to capitalize on the continued increase in AI spending by big tech.</p>
<div class="lb-trans"><p>但英伟达过去三年的强劲业绩并不一定意味着该股票在未来三年内会成为一个绝佳投资。事实上，以下三只半导体股票目前看起来更适合投资，以利用大型科技公司持续增加的 AI 支出。</p>
</div><p><img src="https://imageproxy.pbkrs.com/https://g.foolcdn.com/image//query-b3A9cmVzaXplJnVybD1odHRwczovL2cuZm9vbGNkbi5jb20vZWRpdG9yaWFsL2ltYWdlcy84Mjg2NzAvZ2V0dHlpbWFnZXMtYXJ0aWZpY2lhbC1pbnRlbGxnZW5jZS1jaGlwLmpwZWcmdz03MDA" alt="" original-src="https://g.foolcdn.com/image/?url=https%3A%2F%2Fg.foolcdn.com%2Feditorial%2Fimages%2F828670%2Fgettyimages-artificial-intellgence-chip.jpeg&amp;op=resize&amp;w=700"/></p>
<p>Image source: Getty Images.</p>
<div class="lb-trans"><p>图片来源：Getty Images。</p>
</div><h2>1. Marvell Technology</h2>
<div class="lb-trans"><h2>1. 马威尔科技</h2>
</div><p><strong>Marvell Technology</strong>(MRVL 0.63%) is the chipmaker behind custom AI chip designs from Amazon and <strong>Microsoft</strong>. Instead of GPUs, these custom chips, called XPUs, are designed to more exact specifications to optimize price performance for large language model training or inference.</p>
<div class="lb-trans"><p><strong>马威尔科技</strong>（MRVL 0.63%）是亚马逊和 <strong>微软</strong> 定制 AI 芯片设计的芯片制造商。这些定制芯片被称为 XPU，与 GPU 不同，XPU 是根据更精确的规格设计的，以优化大型语言模型训练或推理的性价比。</p>
</div><p>The XPU opportunity is growing quickly, as big tech companies look to use more cost-efficient chips and reduce their reliance on Nvidia. <strong>Broadcom</strong>, another XPU maker, expects its three custom silicon customers to spend $60 billion to $90 billion on chips by 2027. Marvell sees a $55 billion opportunity for its customers by 2028. To put that in perspective, its revenue over the last four quarters totals $6.5 billion.</p>
<div class="lb-trans"><p>XPU 的机会正在迅速增长，因为大型科技公司希望使用更具成本效益的芯片，减少对英伟达的依赖。<strong>博通</strong>，另一家 XPU 制造商，预计到 2027 年，其三家定制硅客户将在芯片上花费 600 亿到 900 亿美元。马威尔预计到 2028 年，其客户将面临 550 亿美元的机会。为了更好地理解，这一数字与其过去四个季度的收入总额 65 亿美元相比。</p>
</div><p>And Marvell looks poised to capitalize on that opportunity. Microsoft recently upgraded the specifications of its next-generation Maia300 chip, according to a report from Fubon Research. And while that will push out the production timeline until late next year, it also suggests Microsoft will ramp up its production quickly.</p>
<div class="lb-trans"><p>马威尔似乎准备好抓住这一机会。根据富邦研究的报告，微软最近升级了其下一代 Maia300 芯片的规格。尽管这将推迟生产时间表至明年年底，但这也表明微软将迅速增加生产。</p>
</div><p>Fubon expects 300,000 to 400,000 units ordered for the fourth quarter of 2026, with a similar pace of spending throughout 2027. That could mean $10 billion to $12 billion in annual revenue from a single chip design.</p>
<div class="lb-trans"><p>富邦预计 2026 年第四季度将订购 30 万到 40 万台，2027 年将保持类似的支出速度。这可能意味着单一芯片设计的年收入可达 100 亿到 120 亿美元。</p>
</div><p>Marvell also designs networking chips for getting the most out of GPUs and XPUs. Networking equipment ensures data moves quickly and efficiently from one server to another. That's essential for getting the most out of the processing capabilities of the expensive servers used for training and running AI systems.</p>
<div class="lb-trans"><p>马威尔还设计用于充分利用 GPU 和 XPU 的网络芯片。网络设备确保数据快速高效地从一个服务器移动到另一个服务器。这对于充分发挥用于训练和运行 AI 系统的昂贵服务器的处理能力至关重要。</p>
</div><p>As mentioned, it isn't the only company working with big tech on custom silicon solutions. Broadcom has several high-profile contracts as well, and just as many opportunities to expand its market share as Marvell.</p>
<div class="lb-trans"><p>如前所述，它并不是唯一一家与大型科技公司合作定制硅解决方案的公司。博通也有多项高调合同，并且与马威尔一样有许多扩展市场份额的机会。</p>
</div><p>But Broadcom's stock is far more expensive for new investors at the moment. Marvell trades for 27 times forward earnings estimates compared to a multiple of 45 for Broadcom. That makes Marvell a much better option for investors looking to capitalize on the rise of custom silicon.</p>
<div class="lb-trans"><p>但目前博通的股票对新投资者来说要贵得多。马威尔的交易价格为预期收益的 27 倍，而博通的倍数为 45 倍。这使得马威尔成为希望利用定制硅崛起的投资者的更好选择。</p>
</div><h2>2. Micron Technology</h2>
<div class="lb-trans"><h2>2. 美光科技</h2>
</div><p><strong>Micron Technology</strong>(MU 3.26%) is one of a handful of suppliers of memory chips. Perhaps the most important type of memory chip for generative AI is called high-bandwidth memory, or HBM. These chips are attached to GPUs, giving the processors access to data with minimal latency.</p>
<div class="lb-trans"><p><strong>美光科技</strong>（MU 3.26%）是少数几家内存芯片供应商之一。对于生成 AI 来说，最重要的内存芯片类型是称为高带宽内存（HBM）的芯片。这些芯片附加在 GPU 上，使处理器能够以最小的延迟访问数据。</p>
</div><p>Micron was a bit late to develop its HBM chips, but it has quickly caught up to the competition, and it's winning some big contracts. <strong>AMD</strong> tapped Micron to supply the HBM3E chips in its MI355X GPUs, which have shown promising results. AMD's newest chips could be poised to take share of the GPU market from Nvidia by offering competitive price performance.</p>
<div class="lb-trans"><p>美光在开发 HBM 芯片方面稍显滞后，但它迅速赶上了竞争，并赢得了一些大合同。<strong>AMD</strong> 选择美光为其 MI355X GPU 提供 HBM3E 芯片，这些芯片表现出良好的结果。AMD 最新的芯片可能会通过提供具有竞争力的性价比来夺取英伟达的 GPU 市场份额。</p>
</div><p>Micron plans to start production on its HBM4 chips next year, after delivering samples to potential customers in the first half of this year. Management says the new chips offer a 60% performance improvement over its current generation of chips with a 20% reduction in power consumption.</p>
<div class="lb-trans"><p>美光计划在明年开始生产 HBM4 芯片，此前在今年上半年向潜在客户交付了样品。管理层表示，新芯片在性能上比当前一代芯片提高了 60%，同时功耗降低了 20%。</p>
</div><p>HBM chips have been the biggest driver of Micron's results recently. Sales of HBM increased 50% sequentially, driving overall revenue up by 15% from its previous quarter and 51% from the previous year.</p>
<div class="lb-trans"><p>HBM 芯片最近是美光业绩的最大驱动力。HBM 的销售环比增长了 50%，推动整体收入较上个季度增长了 15%，较去年增长了 51%。</p>
</div><p>And the continued ramp-up in HBM should drive expanding margins. Management's outlook calls for a 42% gross margin next quarter, up from 39% in its most recent quarter and 28% a year ago.</p>
<div class="lb-trans"><p>HBM 的持续增长应推动利润率扩大。管理层的展望预计下个季度的毛利率为 42%，高于最近一个季度的 39% 和一年前的 28%。</p>
</div><p>The company sees very strong profit growth in the near future as GPU and XPU makers continue to buy up its memory chips, but it faces significant risks. Since Micron manufactures its own chips, it spends heavily on tooling and capacity while demand increases. But if demand falls, it could see a sharp drop in profitability. That's exacerbated by the fact that memory chips aren't well differentiated and a customer could easily switch suppliers.</p>
<div class="lb-trans"><p>公司预计在不久的将来将实现强劲的利润增长，因为 GPU 和 XPU 制造商将继续购买其内存芯片，但它面临重大风险。由于美光自己制造芯片，因此在需求增加时在工具和产能上投入巨大。但如果需求下降，可能会导致盈利能力急剧下降。这一点因内存芯片的差异化程度不高而加剧，客户可能很容易更换供应商。</p>
</div><p>Still, the stock currently trades for just 14 times earnings estimates. At that price, investors are getting a lot of potential. And if AI spending continues to rise as expected, this earnings cycle could extend well into the future, making the current price for Micron Technology a great opportunity for investors.</p>
<div class="lb-trans"><p>尽管如此，该股票目前的交易价格仅为预期收益的 14 倍。在这个价格下，投资者获得了很大的潜力。如果 AI 支出如预期继续上升，这一盈利周期可能会延续到未来，使得美光科技的当前价格成为投资者的绝佳机会。</p>
</div><h2>3. Taiwan Semiconductor Manufacturing</h2>
<div class="lb-trans"><h2>3. 台湾半导体制造公司</h2>
</div><p><strong>Taiwan Semiconductor Manufacturing</strong>(TSM 0.77%), known as TSMC for short, is the largest chip fabricator in the world. If Nvidia, Broadcom, Marvell, or anyone else wants a leading-edge chip design produced, they're probably going to work with TSMC. That's because it has some of the best technology in the world. As a result, roughly two-thirds of all silicon manufacturing spending goes to the Taiwanese company.</p>
<div class="lb-trans"><p><strong>台湾半导体制造公司</strong>（TSM 0.77%），简称 TSMC，是全球最大的芯片制造商。如果 Nvidia、博通、迈威尔科技或其他任何公司想要生产领先的芯片设计，他们很可能会与 TSMC 合作。这是因为它拥有世界上最先进的技术。因此，约三分之二的硅制造支出都流向了这家台湾公司。</p>
</div><p>TSMC's technology lead is supported by a virtuous cycle. Since the company commands a huge share of all spending, it has more money to reinvest in research and development (R&amp;D) and building extra capacity. With more R&amp;D spending, it's able to develop and perfect the next-generation processes faster, and it has the capacity to meet demand for new chips, bringing in even more revenue.</p>
<div class="lb-trans"><p>TSMC 的技术领先得益于良性循环。由于该公司占据了所有支出的巨大份额，它有更多资金再投资于研发和扩建产能。随着研发支出的增加，它能够更快地开发和完善下一代工艺，并且有能力满足新芯片的需求，从而带来更多收入。</p>
</div><p>The company recently received several pieces of good news, further improving its competitive position in the industry. First, <strong>Intel</strong> said it wouldn't offer fabrication services for its newest-generation process. And if it didn't get a major customer under contract for its next-generation process, it would stop developing it entirely. That leaves one less competitor for TSMC as it ramps up its next-generation 2nm process.</p>
<div class="lb-trans"><p>该公司最近收到了几条好消息，进一步提升了其在行业中的竞争地位。首先，<strong>英特尔</strong> 表示不会为其最新一代工艺提供制造服务。如果它没有为下一代工艺获得主要客户的合同，它将完全停止开发。这使得 TSMC 在提升其下一代 2nm 工艺时减少了一个竞争对手。</p>
</div><p>And TSMC received an exemption from the Trump administration's 100% tariff on foreign-sourced semiconductors. That exemption was likely due to huge commitments from the company to build factories in Arizona.</p>
<div class="lb-trans"><p>此外，TSMC 还获得了特朗普政府对外国半导体 100% 关税的豁免。这一豁免可能是由于该公司在亚利桑那州建设工厂的巨大承诺。</p>
</div><p>Both clear the way for TSMC to ramp up production of its 2nm process, which already has customers lined up. The company is reportedly charging a significant premium over the previous generation 3nm chips. Combined with a quick ramp up in production, it should support strong gross margins as it quickly transitions to the next-generation node.</p>
<div class="lb-trans"><p>这两项措施为 TSMC 加大 2nm 工艺的生产铺平了道路，目前已有客户排队等候。该公司据报道对上一代 3nm 芯片收取了显著的溢价。结合快速的生产提升，这应该能支持强劲的毛利率，因为它迅速过渡到下一代节点。</p>
</div><p>Despite the recent price gains in the stock on the good news, TSMC shares currently trade for less than 25 times forward earnings. Considering management expects average revenue increases of 20% per year between now and 2029 based on strong AI chip growth, that's still a great price to pay for a company with huge competitive advantages.</p>
<div class="lb-trans"><p>尽管近期因好消息而股价上涨，TSMC 的股票目前交易价格不到未来收益的 25 倍。考虑到管理层预计基于强劲的 AI 芯片增长，从现在到 2029 年平均收入将增长 20%，这对于一家拥有巨大竞争优势的公司来说，仍然是一个很好的价格。</p>
</div>

Big Tech Is on Track to Spend Over $1 Trillion on AI Infrastructure by 2028. These 3 Semiconductor Stocks Could Be the Biggest Winners (Hint: Not Nvidia)

- 大型科技公司每月增加人工智能数据中心预算，Alphabet 和亚马逊分别提高至 850 亿美元和 1180 亿美元。  
- 市场研究机构预测，2028 年前十大科技公司在 AI 基础设施上的支出将超过 1 万亿美元。  
- Nvidia、Marvell、Micron 和台积电等公司在 AI 支出增长中受益，未来前景乐观。  

大型科技公司预计到 2028 年将在人工智能基础设施上投入超过 1 万亿美元。这三只半导体股票可能成为最大的赢家（提示：不是英伟达）