<p>几乎所有全球大型科技公司都在竞相建立比其他公司更多的计算能力，以满足大型语言模型训练和人工智能推理的需求。仅四大支出公司今年的资本支出就预计超过 3200 亿美元，其中大部分将用于数据中心。</p>
<p>少数几家公司从这一切支出中受益匪浅，其中最重要的就是 <strong>Nvidia</strong>（NVDA -0.74%）。这家芯片制造商的数据显示，其数据中心收入在三年内增长了超过 10 倍，因为大型科技公司以最快的速度购买其图形处理单元（GPU）。这为公司及其投资者带来了惊人的盈利增长。</p>
<p>需求继续推动 Nvidia 的增长，预计第二季度收入将增长 50%。但竞争开始追赶市场领导者，另一家人工智能巨头可能在未来三年内以更快的速度增长其人工智能芯片收入。此外，其股票的估值更具吸引力，使其成为希望利用人工智能下一个增长周期的投资者的绝佳选择。</p>
<h2>竞争正在取得进展</h2>
<p>Nvidia 面临来自两个主要来源的竞争：其他 GPU 制造商和定制芯片设计。</p>
<p>在其他 GPU 来源方面，<strong>Advanced Micro Devices</strong>（AMD -2.31%）是 Nvidia 唯一的实际竞争对手。尽管 Nvidia 的芯片成本仍然很高，但它仍然是超大规模云计算公司的首选，提供更好的性能。Nvidia 的专有软件 CUDA 为其锁定客户提供了另一种方式。此外，Nvidia 还提供更好的扩展解决方案。</p>
<p>但 AMD 上个月展示了新芯片，旨在消除许多这些优势。其 MI400 芯片预计将在 2026 年推出，提供一种 “机架规模” 系统，使一整机架的 AMD 芯片能够像单一计算引擎一样运行。这可以提供比 Nvidia 更好的性价比，为大型数据中心提供了一个真正的竞争者。它已经获得了 OpenAI、<strong>Meta Platforms</strong> 和 <strong>Microsoft</strong> 等公司的新系列芯片的承诺。</p>
<p>竞争的另一个来源是来自 <strong>Meta</strong>、<strong>Microsoft</strong>、<strong>Alphabet</strong> 的谷歌和 <strong>Amazon</strong>（AMZN -0.07%）等公司的定制硅设计。这些公司是 Nvidia 最大的客户，但它们都希望用更多的定制设计芯片填满自己的数据中心，这些芯片是与 <strong>Broadcom</strong>（AVGO -0.37%）和 <strong>Marvell Technology</strong> 合作制造的。</p>
<p>这些应用特定集成电路（ASIC）在生成性人工智能方面不如 GPU 灵活。然而，上述四家公司在其定制解决方案的人工智能训练和推理方面都取得了出色的成果。它们都在努力扩大这些芯片在未来迭代中的使用案例。对 <strong>Microsoft</strong>、<strong>Google</strong> 和 <strong>Amazon</strong> 来说，重要的是，云服务提供商表示，他们的客户能够通过自己的芯片设计获得比 Nvidia 更好的性价比。而且它们各自都有激励去向客户推销定制硅，因为这将使客户锁定在各自的生态系统中。</p>
<p>但上述公司中没有一家处于与另一家重要的人工智能基础设施提供商同样良好的位置。无论客户最终决定使用哪种芯片，它都可以利用不断增长的支出。</p>
<p><img src="https://imageproxy.pbkrs.com/https://g.foolcdn.com/image//query-b3A9cmVzaXplJnVybD1odHRwczovL2cuZm9vbGNkbi5jb20vZWRpdG9yaWFsL2ltYWdlcy84MjM5ODUvZ2V0dHlpbWFnZXMtYXJ0aWZpY2lhbC1pbnRlbGxnZW5jZS1jaGlwLmpwZWcmdz03MDA?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-b3A9cmVzaXplJnVybD1odHRwczovL2cuZm9vbGNkbi5jb20vZWRpdG9yaWFsL2ltYWdlcy84MjM5ODUvZ2V0dHlpbWFnZXMtYXJ0aWZpY2lhbC1pbnRlbGxnZW5jZS1jaGlwLmpwZWcmdz03MDA"/></p>
<p>图片来源：Getty Images。</p>
<h2>预计在三年内将人工智能收入增加三倍的公司</h2>
<p>当 Nvidia、AMD、Broadcom 或 Marvell 希望将其设计实际印刷在硅晶圆上时，他们需要与制造商签约。而他们都选择了同一家制造商：<strong>Taiwan Semiconductor Manufacturing</strong>（TSM -2.38%），也称为台积电。</p>
<p>台积电凭借两个关键优势在半导体制造中成为主导力量：其规模和技术。这两者相辅相成。台积电的领先技术使其赢得了来自最大芯片设计公司的最大合同。因此，它扩大了规模以满足这种需求。它还将这些收入再投资于研发，以确保保持技术领先。随着对更先进芯片的需求增加，台积电是唯一能够满足这种需求的公司。</p>
<p>在当前需求激增的市场中，台积电可以在其资本支出上产生可观的回报。尽管需求的下行周期会对其利润造成压力，但管理层在预测未来需求、适当预算和为下一个周期做好准备方面历来表现良好。</p>
<p>不过，目前台积电预计其与人工智能相关的收入将在 2025 年至 2027 年间增长三倍。管理层表示，今年它的人工智能相关收入有望从 2024 年翻倍，达到约 260 亿美元。要实现三倍增长，它只需在 2026 年和 2027 年每年增长略超过 20%。</p>
<p>但管理层预计在 2025 年至 2029 年的五年期间，年均增长率将达到 40%，这意味着在今年翻倍后，未来四年平均增长 28%。因此，即使管理层高估了其长期前景，到 2027 年其与人工智能相关的收入仍有可能至少增长三倍。</p>
<p>这一增长的一个关键方面将是其 2nm 和 1.6nm 工艺的推出。台积电计划快速推出这两个制造节点，2nm 将在今年晚些时候到来，下一代将在 2026 年底推出。据报道，台积电对 2nm 芯片的收费为每片 3 万美元，而 3nm 芯片的收费约为 2 万美元。台积电已经为 AMD、Microsoft、Amazon、Google 等公司安排了 2nm 合同。</p>
<p>尽管与人工智能相关的收入在台积电总收入中仍占小部分，但根据管理层的长期预测，到本世纪末，这一比例预计将攀升至约 30%。公司预计在未来五年内总收入将增长约 20%，并且在此期间随着 2nm 和 1.6nm 生产的提升，能够产生强劲的利润率。总体而言，盈利增长应与收入增长保持同步，得益于对人工智能芯片的强劲需求。</p>
<p>与此同时，截至目前，该股票的前瞻市盈率低于 25。这使其相较于其他人工智能股票如 Nvidia 或 AMD（市盈率约为 35 倍）而言，具有极大的价值。考虑到台积电目前的强势地位，值得将其纳入你的投资组合。</p>

台积电

半导体 3 倍做多 - Direxion

<p>人工智能芯片市场正在快速发展，主要科技公司在数据中心方面进行了大量投资。英伟达（Nvidia）实现了显著增长，但来自 AMD 以及微软和谷歌等公司的定制芯片设计的竞争正在上升。台湾半导体制造公司（台积电）预计到 2027 年其人工智能芯片收入将增长超过三倍，受益于其规模和技术领先地位。台积电预计今年其与人工智能相关的收入将翻倍，达到约 260 亿美元，抓住对先进芯片日益增长的需求</p>

<p>Just about every big tech company in the world is in a race to build more computing capacity than anyone else in order to meet the demands for large language model training and AI inference. The four biggest spenders alone are set to invest over $320 billion in capital expenditures this year, most of which will go toward data centers.</p>
<div class="lb-trans"><p>几乎所有全球大型科技公司都在竞相建立比其他公司更多的计算能力，以满足大型语言模型训练和人工智能推理的需求。仅四大支出公司今年的资本支出就预计超过 3200 亿美元，其中大部分将用于数据中心。</p>
</div><p>A handful of companies have been big beneficiaries of all that spending, not least of which is <strong>Nvidia</strong>(NVDA -0.74%). The chipmaker has seen its data center revenue climb more than 10-fold in three years as big tech buys up its graphics processing units (GPUs) as fast as it can make them. That's led to phenomenal earnings growth for the company and its investors.</p>
<div class="lb-trans"><p>少数几家公司从这一切支出中受益匪浅，其中最重要的就是 <strong>Nvidia</strong>（NVDA -0.74%）。这家芯片制造商的数据显示，其数据中心收入在三年内增长了超过 10 倍，因为大型科技公司以最快的速度购买其图形处理单元（GPU）。这为公司及其投资者带来了惊人的盈利增长。</p>
</div><p>Demand continues to fuel growth at Nvidia, with expectations for 50% revenue growth in the second quarter.But competition is starting to catch up with the market leader and another AI giant could grow its AI chip revenue faster than Nvidia over the next three years. What's more, its stock trades at a much more attractive valuation, making it a great buy for investors looking to capitalize on the next round of growth in AI.</p>
<div class="lb-trans"><p>需求继续推动 Nvidia 的增长，预计第二季度收入将增长 50%。但竞争开始追赶市场领导者，另一家人工智能巨头可能在未来三年内以更快的速度增长其人工智能芯片收入。此外，其股票的估值更具吸引力，使其成为希望利用人工智能下一个增长周期的投资者的绝佳选择。</p>
</div><h2>The competition is making progress</h2>
<div class="lb-trans"><h2>竞争正在取得进展</h2>
</div><p>Nvidia faces competition from two main source: other GPU makers and custom chip designs.</p>
<div class="lb-trans"><p>Nvidia 面临来自两个主要来源的竞争：其他 GPU 制造商和定制芯片设计。</p>
</div><p>When it comes to other sources for GPUs, <strong>Advanced Micro Devices</strong>(AMD -2.31%) is the only practical competitor for Nvidia. Nvidia has remained a top choice for hyperscalers, offering better performance even though the costs of its chips remain high. Nvidia's proprietary software, CUDA, gives it another way to lock in customers. And Nvidia offers better scaling solutions as well.</p>
<div class="lb-trans"><p>在其他 GPU 来源方面，<strong>Advanced Micro Devices</strong>（AMD -2.31%）是 Nvidia 唯一的实际竞争对手。尽管 Nvidia 的芯片成本仍然很高，但它仍然是超大规模云计算公司的首选，提供更好的性能。Nvidia 的专有软件 CUDA 为其锁定客户提供了另一种方式。此外，Nvidia 还提供更好的扩展解决方案。</p>
</div><p>But AMD showed off new chips last month that aim to put many of those advantages to bed. Its MI400 chips, coming in 2026, offer a &#34;rack-scale&#34; system that allows a full rack of AMD chips to function like a single compute engine. That can offer better price performance than Nvidia, providing a legitimate contender for large data centers. It's already received commitments from OpenAI, <strong>Meta Platforms</strong>, and <strong>Microsoft</strong>, among others, for its newest series of chips.</p>
<div class="lb-trans"><p>但 AMD 上个月展示了新芯片，旨在消除许多这些优势。其 MI400 芯片预计将在 2026 年推出，提供一种 “机架规模” 系统，使一整机架的 AMD 芯片能够像单一计算引擎一样运行。这可以提供比 Nvidia 更好的性价比，为大型数据中心提供了一个真正的竞争者。它已经获得了 OpenAI、<strong>Meta Platforms</strong> 和 <strong>Microsoft</strong> 等公司的新系列芯片的承诺。</p>
</div><p>The other source of competition is custom silicon designs from companies like Meta, Microsoft, <strong>Alphabet</strong>'s Google, and <strong>Amazon</strong>(AMZN -0.07%). Those represent Nvidia's biggest customers, but they're all looking to fill their data centers with more of their custom-designed chips made in partnership with <strong>Broadcom</strong>(AVGO -0.37%) and <strong>Marvell Technology</strong>.</p>
<div class="lb-trans"><p>竞争的另一个来源是来自 <strong>Meta</strong>、<strong>Microsoft</strong>、<strong>Alphabet</strong> 的谷歌和 <strong>Amazon</strong>（AMZN -0.07%）等公司的定制硅设计。这些公司是 Nvidia 最大的客户，但它们都希望用更多的定制设计芯片填满自己的数据中心，这些芯片是与 <strong>Broadcom</strong>（AVGO -0.37%）和 <strong>Marvell Technology</strong> 合作制造的。</p>
</div><p>These application-specific integrated circuits (ASICs) aren't as flexible as GPUs for generative AI purposes. However, all four of the above have seen excellent results with AI training and inference on their custom solutions. And they're all trying to expand the use cases of those chips in future iterations. Importantly for Microsoft, Google, and Amazon, the cloud providers say their customers are able to generate better price performance with their own chip designs than Nvidia's. And they each have an incentive to sell their customers on custom silicon, since it will lock those customers into their respective ecosystems.</p>
<div class="lb-trans"><p>这些应用特定集成电路（ASIC）在生成性人工智能方面不如 GPU 灵活。然而，上述四家公司在其定制解决方案的人工智能训练和推理方面都取得了出色的成果。它们都在努力扩大这些芯片在未来迭代中的使用案例。对 <strong>Microsoft</strong>、<strong>Google</strong> 和 <strong>Amazon</strong> 来说，重要的是，云服务提供商表示，他们的客户能够通过自己的芯片设计获得比 Nvidia 更好的性价比。而且它们各自都有激励去向客户推销定制硅，因为这将使客户锁定在各自的生态系统中。</p>
</div><p>But none of the above-mentioned companies are in as good a position as another important AI infrastructure provider. And it can capitalize on the growing spending, no matter which chips customers ultimately decide to build with.</p>
<div class="lb-trans"><p>但上述公司中没有一家处于与另一家重要的人工智能基础设施提供商同样良好的位置。无论客户最终决定使用哪种芯片，它都可以利用不断增长的支出。</p>
</div><p><img src="https://imageproxy.pbkrs.com/https://g.foolcdn.com/image//query-b3A9cmVzaXplJnVybD1odHRwczovL2cuZm9vbGNkbi5jb20vZWRpdG9yaWFsL2ltYWdlcy84MjM5ODUvZ2V0dHlpbWFnZXMtYXJ0aWZpY2lhbC1pbnRlbGxnZW5jZS1jaGlwLmpwZWcmdz03MDA" alt="" original-src="https://g.foolcdn.com/image/?url=https%3A%2F%2Fg.foolcdn.com%2Feditorial%2Fimages%2F823985%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>The company set to more than triple AI revenue in three years</h2>
<div class="lb-trans"><h2>预计在三年内将人工智能收入增加三倍的公司</h2>
</div><p>When Nvidia, AMD, Broadcom, or Marvell want to actually take their designs and get them printed on silicon wafers, they need to contract with a manufacturer. And they all choose the same manufacturer for their needs: <strong>Taiwan Semiconductor Manufacturing</strong>(TSM -2.38%), otherwise known as TSMC.</p>
<div class="lb-trans"><p>当 Nvidia、AMD、Broadcom 或 Marvell 希望将其设计实际印刷在硅晶圆上时，他们需要与制造商签约。而他们都选择了同一家制造商：<strong>Taiwan Semiconductor Manufacturing</strong>（TSM -2.38%），也称为台积电。</p>
</div><p>TSMC has become a dominant force in semiconductor fabrication due to two key advantages: its scale and its technology. Both serve to feed one another. TSMC's leading technology means it wins the biggest contracts from the biggest chip designers. As a result, it builds out scale to meet that demand. It also takes that revenue and reinvests it into R&amp;D to ensure it maintains its technology lead. As demand for more advanced chips increases, it's the only company that has the scale that can meet that demand.</p>
<div class="lb-trans"><p>台积电凭借两个关键优势在半导体制造中成为主导力量：其规模和技术。这两者相辅相成。台积电的领先技术使其赢得了来自最大芯片设计公司的最大合同。因此，它扩大了规模以满足这种需求。它还将这些收入再投资于研发，以确保保持技术领先。随着对更先进芯片的需求增加，台积电是唯一能够满足这种需求的公司。</p>
</div><p>In a market with booming demand, like we're seeing right now, TSMC can produce considerable returns on its capital spending. And while a down-cycle in demand will put a strain on its profits, management has historically done a good job at forecasting future demand, budgeting appropriately, and positioning for the next cycle.</p>
<div class="lb-trans"><p>在当前需求激增的市场中，台积电可以在其资本支出上产生可观的回报。尽管需求的下行周期会对其利润造成压力，但管理层在预测未来需求、适当预算和为下一个周期做好准备方面历来表现良好。</p>
</div><p>Right now, though, TSMC could see its AI-related revenue more than triple from 2025 through 2027. Management said it's on track to double its AI-related revenue from 2024 this year alone, bringing it to around $26 billion. To triple, it would only need to grow a bit over 20% per year for 2026 and 2027.</p>
<div class="lb-trans"><p>不过，目前台积电预计其与人工智能相关的收入将在 2025 年至 2027 年间增长三倍。管理层表示，今年它的人工智能相关收入有望从 2024 年翻倍，达到约 260 亿美元。要实现三倍增长，它只需在 2026 年和 2027 年每年增长略超过 20%。</p>
</div><p>But management expects 40% average annual growth over the five-year period from 2025 through 2029, implying an average of 28% growth in the next four years after doubling this year. So, even if management is overestimating its long-term prospects, it's still likely to at least triple AI-related revenue by 2027.</p>
<div class="lb-trans"><p>但管理层预计在 2025 年至 2029 年的五年期间，年均增长率将达到 40%，这意味着在今年翻倍后，未来四年平均增长 28%。因此，即使管理层高估了其长期前景，到 2027 年其与人工智能相关的收入仍有可能至少增长三倍。</p>
</div><p>A key aspect of that growth will be the introduction of its 2nm and 1.6nm processes. TSMC is set to release both manufacturing nodes in quick succession, with the 2nm arriving later this year and the next generation arriving in late 2026. The company is reportedly charging $30,000 per wafer for 2nm chips compared to about $20,000 for 3nm chips. TSMC already has 2nm contracts lined up for AMD, Microsoft, Amazon, Google, and others.</p>
<div class="lb-trans"><p>这一增长的一个关键方面将是其 2nm 和 1.6nm 工艺的推出。台积电计划快速推出这两个制造节点，2nm 将在今年晚些时候到来，下一代将在 2026 年底推出。据报道，台积电对 2nm 芯片的收费为每片 3 万美元，而 3nm 芯片的收费约为 2 万美元。台积电已经为 AMD、Microsoft、Amazon、Google 等公司安排了 2nm 合同。</p>
</div><p>While AI-related revenue remains a small portion of TSMC's total revenue, it should climb to about 30% of total revenue by the end of the decade based on management's long-term forecast. The company expects about 20% total revenue growth over the next five years, and it should be able to produce strong margins in that period as it ramps up 2nm and 1.6nm production. Overall, earnings growth should keep pace with revenue growth, bolstered by strong demand for AI chips.</p>
<div class="lb-trans"><p>尽管与人工智能相关的收入在台积电总收入中仍占小部分，但根据管理层的长期预测，到本世纪末，这一比例预计将攀升至约 30%。公司预计在未来五年内总收入将增长约 20%，并且在此期间随着 2nm 和 1.6nm 生产的提升，能够产生强劲的利润率。总体而言，盈利增长应与收入增长保持同步，得益于对人工智能芯片的强劲需求。</p>
</div><p>Meanwhile, the stock trades for a forward PE ratio of less than 25 as of this writing. That makes it an incredible value compared to other AI stocks like Nvidia or AMD, which trade around 35 times earnings. With the strong position TSMC is in right now, it's worth adding to your portfolio.</p>
<div class="lb-trans"><p>与此同时，截至目前，该股票的前瞻市盈率低于 25。这使其相较于其他人工智能股票如 Nvidia 或 AMD（市盈率约为 35 倍）而言，具有极大的价值。考虑到台积电目前的强势地位，值得将其纳入你的投资组合。</p>
</div>

Prediction: This Artificial Intelligence (AI) Giant Will More Than Triple Its AI Chip Revenue in 3 Years. (Hint: Not Nvidia)

- 全球科技公司竞相提升计算能力以满足 AI 需求。  
- 四大巨头预计今年资本支出超过 3200 亿美元，主要用于数据中心。  
- TSMC 预计 AI 收入在 2025 至 2027 年间将增长三倍。  

预测：这家人工智能（AI）巨头将在三年内将其 AI 芯片收入增长超过三倍。（提示：不是 Nvidia）