<p>在 2024 年，<strong>英偉達</strong>(NVDA -1.14%) 在人工智能（AI）基礎設施方面佔據了主導地位。然而，<strong>博通</strong>(AVGO -6.91%) 將在 2025 年尋求挑戰這家芯片製造商。</p>
<p>這兩隻股票在 2024 年表現強勁，截至目前，英偉達的股價上漲超過 170%，而博通上漲約 107%。讓我們考慮一下在 2025 年即將到來之際，哪隻半導體股票看起來更值得購買。</p>
<h2>AI 基礎設施增長</h2>
<p>自 AI 熱潮開始以來，英偉達是最大的贏家。作為圖形處理單元（GPU）的製造商，該公司的芯片成為 AI 基礎設施的支柱。訓練大型語言模型（LLM）和運行 AI 推理需要大量計算能力，而 GPU 已被證明是執行這些任務的最佳芯片，因為它們能夠同時進行許多計算，這被稱為並行處理。</p>
<p>與此同時，英偉達因其 CUDA 軟件平台成為 GPU 市場的領導者。該公司最初開發了這款免費軟件，以便開發者能夠將其芯片編程用於超越加速圖形渲染（如視頻遊戲）的任務，從而銷售更多芯片。因此，CUDA 成為開發者學習訓練 GPU 的程序，這幫助公司建立了今天所看到的巨大護城河。結果，它在 GPU 市場的份額約為 90%。</p>
<p>強大的 GPU 與 CUDA 帶來的廣泛護城河的結合，使得公司在大型科技公司競相開發更強大的 AI 模型時實現了驚人的增長。在截至 2025 年 1 月的 2025 財年前九個月，英偉達的收入激增 135%，達到 912 億美元，而上個季度其收入飆升 94%，達到 351 億美元。</p>
<p>隨着 AI 模型在變得更加先進時需要指數級增加的計算能力，英偉達的未來增長前景也看起來樂觀。事實上，<strong>Alphabet</strong> 的 Llama AI 模型和 xAI 的 Grok 模型的最新版本正在使用比其前身多出 10 倍的 GPU 進行訓練。同時，英偉達的超大規模客户（擁有大型數據中心的客户）表示，他們計劃在明年增加對 AI 基礎設施的支出，因為他們將 AI 視為一項世代機遇。分析師預計該公司明年的收入將增長超過 50%。</p>
<p><img src="https://imageproxy.pbkrs.com/https://g.foolcdn.com/image//query-b3A9cmVzaXplJnVybD1odHRwczovL2cuZm9vbGNkbi5jb20vZWRpdG9yaWFsL2ltYWdlcy84MDEyMDUvZ2V0dHlpbWFnZXMtMTQwMjQ1Mjg3Ni5qcGcmdz03MDA?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://g.foolcdn.com/image//query-b3A9cmVzaXplJnVybD1odHRwczovL2cuZm9vbGNkbi5jb20vZWRpdG9yaWFsL2ltYWdlcy84MDEyMDUvZ2V0dHlpbWFnZXMtMTQwMjQ1Mjg3Ni5qcGcmdz03MDA"/></p>
<p>圖片來源：Getty Images。</p>
<p>雖然英偉達在 AI 芯片領域佔據主導地位，但博通通過幫助客户開發定製 AI 芯片正在取得進展。其應用特定集成電路（ASIC）是專門為客户的精確需求設計的，可以幫助提高性能和更高效的電力消耗。</p>
<p>Alphabet 是第一家利用博通的技術和專業知識開發自有定製 AI 芯片的公司。結果是一個名為 Trillium 的張量處理單元（TPU），專門設計用於在谷歌的 TensorFlow（一個用於 AI 和機器學習的軟件庫）中工作。Alphabet 表示，這些 TPU 具有多個特性，如矩陣乘法單元（MXU）和專有互連拓撲，使其與大規模商用 GPU 有所不同，並使其在加速 AI 訓練和推理方面理想。</p>
<p>儘管有一些報道表明 Alphabet 希望獨立於博通，但博通贏得了 Alphabet 下一代 TPU 的設計合同。同時，該公司獲得了四個定製 AI 芯片的新客户。<strong>Meta Platforms</strong> 和抖音母公司字節跳動被廣泛認為是公司的既有客户，而 OpenAI 和最近的 <strong>蘋果</strong> 則被認為是新獲得的客户。</p>
<p>在最新的財報電話會議上，博通表示其三大定製 AI 芯片客户在 2027 財年的可尋址市場在 600 億美元到 900 億美元之間，並計劃到 2027 年部署 100 萬個定製 AI 芯片。同時，它表示其兩個新客户可能會增加這一總數。作為參考，Alphabet 花了 15 個月時間設計並在其數據中心內部署其定製 TPU。</p>
<p>雖然博通在 AI 領域的機會很大，但可尋址市場與預期收入之間存在很大差異。該公司還涉及許多與 AI 無關的半導體和軟件業務，這些業務的增長速度並不快。該公司在排除收購 VMware 後，上個季度的有機收入僅增長了 11%，分析師目前預計該公司在截至 2025 年 10 月 31 日的財年內收入將增長 18%，而在之後的年份增長 14%。</p>
<h2>估值與結論</h2>
<p>目前，英偉達的估值低於博通，前瞻市盈率（P/E）約為 30，而博通則超過 33。同時，英偉達的收入增長速度遠快於博通，預計這種趨勢將在 2025 年繼續。此外，英偉達持有約 300 億美元的淨現金，而博通則有 483 億美元的淨債務。</p>
<p><img src="https://imageproxy.pbkrs.com/https://media.ycharts.com/charts/9415ff63bfa57e90f4443a7b63b1bbe9.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://media.ycharts.com/charts/9415ff63bfa57e90f4443a7b63b1bbe9.png"/></p>
<p>NVDA 市盈率（前瞻 1 年）數據來源於 YCharts。</p>
<p>博通通過其對 AI 可尋址市場的評論製造了很多炒作，這推動了其在 12 月的股價。同時，英偉達的股價在年末時表現不佳。但儘管博通獲得了動力，英偉達現在是更便宜的股票，並且預計在短期內其收入增長速度將更快。</p>
<p>博通管理層無疑在投資者心中對未來幾年誰將成為 AI 芯片贏家產生了一些懷疑。但博通的收益不一定會以英偉達的損失為代價。GPU 仍然比定製芯片更具多樣性，並且仍被視為標準。然而，公司也希望有一個除了英偉達之外的選擇，以確保其不會變得過於強大。</p>
<p>我認為這兩隻股票在 2025 年都有成為贏家的潛力，但考慮到目前英偉達的增長優勢和更低的估值，我更傾向於選擇英偉達。</p>

英偉達

博通

<p>英偉達和博通正在人工智能芯片市場競爭，英偉達目前由於其主導的 GPU 市場份額和強勁的收入增長而處於領先地位。英偉達在 2025 財年的收入激增了 135%，這得益於對人工智能基礎設施需求的增加。相比之下，博通正在開發定製的人工智能芯片，並已與谷歌和 Meta 等主要公司簽訂合同。儘管博通的人工智能機會相當可觀，但其整體增長速度相比英偉達較慢，預計英偉達明年的收入將增長超過 50%。英偉達的估值更為有利，市盈率低於博通</p>

<p>When it came to artificial intelligence (AI) infrastructure in 2024, <strong>Nvidia</strong>(NVDA -1.14%) reigned supreme. However, <strong>Broadcom</strong>(AVGO -6.91%) will be looking to challenge the chipmaker in 2025.</p>
<div class="lb-trans"><p>在 2024 年，<strong>英偉達</strong>(NVDA -1.14%) 在人工智能（AI）基礎設施方面佔據了主導地位。然而，<strong>博通</strong>(AVGO -6.91%) 將在 2025 年尋求挑戰這家芯片製造商。</p>
</div><p>Both stocks have had strong runs in 2024, with Nvidia's stock up over 170% year to date as of this writing and Broadcom up around 107%. Let's consider which semiconductor stock looks like the better buy heading into 2025.</p>
<div class="lb-trans"><p>這兩隻股票在 2024 年表現強勁，截至目前，英偉達的股價上漲超過 170%，而博通上漲約 107%。讓我們考慮一下在 2025 年即將到來之際，哪隻半導體股票看起來更值得購買。</p>
</div><h2>AI infrastructure growth</h2>
<div class="lb-trans"><h2>AI 基礎設施增長</h2>
</div><p>Since the start of the AI boom, Nvidia has been the biggest winner. As the maker of graphics processing units (GPUs) the company's chips became the backbone of AI infrastructure. Training large language models (LLMs) and running AI inference take a lot of computing power, and GPUs have proven to be the best chips for these tasks as they are able to perform many calculations at the same time in what is called parallel processing.</p>
<div class="lb-trans"><p>自 AI 熱潮開始以來，英偉達是最大的贏家。作為圖形處理單元（GPU）的製造商，該公司的芯片成為 AI 基礎設施的支柱。訓練大型語言模型（LLM）和運行 AI 推理需要大量計算能力，而 GPU 已被證明是執行這些任務的最佳芯片，因為它們能夠同時進行許多計算，這被稱為並行處理。</p>
</div><p>Nvidia, meanwhile, became the market leader in GPUs due to its CUDA software platform. The company originally developed the free software to allow developers to program its chips for tasks beyond speeding up the graphics rendering in things like video games, the job for which they were initially created, in order to sell more chips. As a result, CUDA became the program on which developers learned to train GPUs, which is what has helped create the large moat the company sees today. It currently has about a 90% market share in GPUs as a result.</p>
<div class="lb-trans"><p>與此同時，英偉達因其 CUDA 軟件平台成為 GPU 市場的領導者。該公司最初開發了這款免費軟件，以便開發者能夠將其芯片編程用於超越加速圖形渲染（如視頻遊戲）的任務，從而銷售更多芯片。因此，CUDA 成為開發者學習訓練 GPU 的程序，這幫助公司建立了今天所看到的巨大護城河。結果，它在 GPU 市場的份額約為 90%。</p>
</div><p>The combination of its powerful GPUs and wide moat due to CUDA led the company to see astronomical growth as large tech companies race to develop more powerful AI models. Through the first nine months of its fiscal 2025 ending in January 2025, Nvidia's revenue surged 135% to $91.2 billion, while last quarter its revenue soared 94% to $35.1 billion.</p>
<div class="lb-trans"><p>強大的 GPU 與 CUDA 帶來的廣泛護城河的結合，使得公司在大型科技公司競相開發更強大的 AI 模型時實現了驚人的增長。在截至 2025 年 1 月的 2025 財年前九個月，英偉達的收入激增 135%，達到 912 億美元，而上個季度其收入飆升 94%，達到 351 億美元。</p>
</div><p>With AI models needing exponentially more computing power as they become more advanced, Nvidia's future growth prospects also look promising. In fact, recent iterations of <strong>Alphabet</strong>'s Llama AI model and xAI's Grok model are being trained using up to 10 times as many GPUs as their predecessors. At the same time, Nvidia's hyperscale customers (those that own huge data centers) have indicated they plan to increase their spending on AI infrastructure next year as they see AI as a generational opportunity. For their part, analysts are projecting the company will grow its revenue by just over 50% next year.</p>
<div class="lb-trans"><p>隨着 AI 模型在變得更加先進時需要指數級增加的計算能力，英偉達的未來增長前景也看起來樂觀。事實上，<strong>Alphabet</strong> 的 Llama AI 模型和 xAI 的 Grok 模型的最新版本正在使用比其前身多出 10 倍的 GPU 進行訓練。同時，英偉達的超大規模客户（擁有大型數據中心的客户）表示，他們計劃在明年增加對 AI 基礎設施的支出，因為他們將 AI 視為一項世代機遇。分析師預計該公司明年的收入將增長超過 50%。</p>
</div><p><img src="https://imageproxy.pbkrs.com/https://g.foolcdn.com/image//query-b3A9cmVzaXplJnVybD1odHRwczovL2cuZm9vbGNkbi5jb20vZWRpdG9yaWFsL2ltYWdlcy84MDEyMDUvZ2V0dHlpbWFnZXMtMTQwMjQ1Mjg3Ni5qcGcmdz03MDA" alt="" original-src="https://g.foolcdn.com/image/?url=https%3A%2F%2Fg.foolcdn.com%2Feditorial%2Fimages%2F801205%2Fgettyimages-1402452876.jpg&amp;op=resize&amp;w=700"/></p>
<p>Image source: Getty Images.</p>
<div class="lb-trans"><p>圖片來源：Getty Images。</p>
</div><p>While Nvidia is dominating the AI chip space, Broadcom is making inroads by helping customers develop custom AI chips. Its application-specific integrated circuits, or ASICs, are designed specifically for a customer's precise needs and can help with improved performance and more efficient power consumption.</p>
<div class="lb-trans"><p>雖然英偉達在 AI 芯片領域佔據主導地位，但博通通過幫助客户開發定製 AI 芯片正在取得進展。其應用特定集成電路（ASIC）是專門為客户的精確需求設計的，可以幫助提高性能和更高效的電力消耗。</p>
</div><p>Alphabet was the first company to use Broadcom's technology and expertise to develop its own custom AI chip. The result is a tensor-processing unit (TPU) called Trillium designed specifically to work within Google's TensorFlow (a software library for AI and machine learning). Alphabet said these TPUs have several features, such as a matrix multiply unit (MXU) and proprietary interconnect topology, that differentiate them from mass-merchant GPUs and make them ideal for accelerating AI training and inference.</p>
<div class="lb-trans"><p>Alphabet 是第一家利用博通的技術和專業知識開發自有定製 AI 芯片的公司。結果是一個名為 Trillium 的張量處理單元（TPU），專門設計用於在谷歌的 TensorFlow（一個用於 AI 和機器學習的軟件庫）中工作。Alphabet 表示，這些 TPU 具有多個特性，如矩陣乘法單元（MXU）和專有互連拓撲，使其與大規模商用 GPU 有所不同，並使其在加速 AI 訓練和推理方面理想。</p>
</div><p>While there were some reports that Alphabet was looking to just go on its own without Broadcom's help, Broadcom won the design contract for the next generation of Alphabet's TPUs. Meanwhile, the company gained four more custom AI chip customers. <strong>Meta Platforms</strong> and TikTok owner ByteDance are widely believed to be the company's established customers, while OpenAI and more recently <strong>Apple</strong> are thought to be recent customer wins.</p>
<div class="lb-trans"><p>儘管有一些報道表明 Alphabet 希望獨立於博通，但博通贏得了 Alphabet 下一代 TPU 的設計合同。同時，該公司獲得了四個定製 AI 芯片的新客户。<strong>Meta Platforms</strong> 和抖音母公司字節跳動被廣泛認為是公司的既有客户，而 OpenAI 和最近的 <strong>蘋果</strong> 則被認為是新獲得的客户。</p>
</div><p>On its latest earnings call, Broadcom said its three largest custom AI chip customers had an addressable market of between $60 billion to $90 billion in fiscal 2027 alone and that they were planning on deploying 1 million of its custom AI chips by 2027. Meanwhile, it said its two newer AI chip customers could add to that total. As a reference, it took Alphabet 15 months to have its custom TPUs designed and deployed within its data centers.</p>
<div class="lb-trans"><p>在最新的財報電話會議上，博通表示其三大定製 AI 芯片客户在 2027 財年的可尋址市場在 600 億美元到 900 億美元之間，並計劃到 2027 年部署 100 萬個定製 AI 芯片。同時，它表示其兩個新客户可能會增加這一總數。作為參考，Alphabet 花了 15 個月時間設計並在其數據中心內部署其定製 TPU。</p>
</div><p>While the AI opportunity for Broadcom is large, there is a big difference between an addressable market and expected revenue. The company is also involved in a lot of semiconductor and software businesses outside of AI, which aren't growing as quickly. The company grew organic revenue just 11% last quarter when excluding its acquisition of VMware, and analysts are currently looking for the company to increase its revenue by 18% this fiscal year, ending Oct. 31, 2025, and 14% the year after.</p>
<div class="lb-trans"><p>雖然博通在 AI 領域的機會很大，但可尋址市場與預期收入之間存在很大差異。該公司還涉及許多與 AI 無關的半導體和軟件業務，這些業務的增長速度並不快。該公司在排除收購 VMware 後，上個季度的有機收入僅增長了 11%，分析師目前預計該公司在截至 2025 年 10 月 31 日的財年內收入將增長 18%，而在之後的年份增長 14%。</p>
</div><h2>Valuation and verdict</h2>
<div class="lb-trans"><h2>估值與結論</h2>
</div><p>Nvidia currently trades at a cheaper valuation than Broadcom with a forward price-to-earnings (P/E) ratio of about 30 compared to over 33 for Broadcom. Meanwhile, Nvidia is currently growing its revenue much faster, which is expected to continue in 2025. In addition, Nvidia holds about $30 billion in net cash, while Broadcom has $48.3 billion in net debt.</p>
<div class="lb-trans"><p>目前，英偉達的估值低於博通，前瞻市盈率（P/E）約為 30，而博通則超過 33。同時，英偉達的收入增長速度遠快於博通，預計這種趨勢將在 2025 年繼續。此外，英偉達持有約 300 億美元的淨現金，而博通則有 483 億美元的淨債務。</p>
</div><p><img src="https://imageproxy.pbkrs.com/https://media.ycharts.com/charts/9415ff63bfa57e90f4443a7b63b1bbe9.png" alt="" original-src="https://media.ycharts.com/charts/9415ff63bfa57e90f4443a7b63b1bbe9.png"/></p>
<p>NVDA PE Ratio (Forward 1y) data by YCharts.</p>
<div class="lb-trans"><p>NVDA 市盈率（前瞻 1 年）數據來源於 YCharts。</p>
</div><p>Broadcom created a lot of hype with its comments on its AI addressable market, which have powered its stock price in December. Meanwhile, Nvidia's stock struggled to close out the year. But while Broadcom grabbed the momentum, Nvidia is now the cheaper stock and it's still expected to grow its revenue much more quickly in the near term.</p>
<div class="lb-trans"><p>博通通過其對 AI 可尋址市場的評論製造了很多炒作，這推動了其在 12 月的股價。同時，英偉達的股價在年末時表現不佳。但儘管博通獲得了動力，英偉達現在是更便宜的股票，並且預計在短期內其收入增長速度將更快。</p>
</div><p>Broadcom management has certainly put some doubt in investors' minds when it comes to who will be the big AI chip winner in the coming years. But Broadcom's gains won't necessarily come at Nvidia's expense. GPUs still have more versatility than custom chips and are still considered the standard. However, companies also want an option other than Nvidia to make sure it doesn't become too powerful.</p>
<div class="lb-trans"><p>博通管理層無疑在投資者心中對未來幾年誰將成為 AI 芯片贏家產生了一些懷疑。但博通的收益不一定會以英偉達的損失為代價。GPU 仍然比定製芯片更具多樣性，並且仍被視為標準。然而，公司也希望有一個除了英偉達之外的選擇，以確保其不會變得過於強大。</p>
</div><p>I think both stocks have the potential to be winners in 2025, but I prefer Nvidia at this point given its current superior growth and cheaper valuation.</p>
<div class="lb-trans"><p>我認為這兩隻股票在 2025 年都有成為贏家的潛力，但考慮到目前英偉達的增長優勢和更低的估值，我更傾向於選擇英偉達。</p>
</div>

Nvidia vs. Broadcom: Which Is the Better AI Chip Stock to Own in 2025?

- 2024 年，Nvidia 在人工智能基礎設施領域佔據主導地位。  
- Broadcom 計劃在 2025 年挑戰 Nvidia，已獲得多個定製 AI 芯片客户。  
- Nvidia 預計 2025 年收入將增長超過 50%，而 Broadcom 的增長較慢。  

英偉達與博通：哪隻 AI 芯片股票在 2025 年更值得持有？