<p><img src="https://imageproxy.pbkrs.com/https://invezz-wp-media.lon1.digitaloceanspaces.com/2026/04/image-1-1777446995.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://invezz-wp-media.lon1.digitaloceanspaces.com/2026/04/image-1-1777446995.png"/></p>
<p>根據路透社的報道，華為的 Ascend 950 人工智能芯片的需求在 DeepSeek 的 V4 模型發佈後激增，該模型運行在這家總部位於深圳的公司的硬件上。</p>
<p>包括字節跳動、騰訊和阿里巴巴在內的主要中國互聯網公司正在積極聯繫華為，以確保獲得新的芯片訂單，消息人士告訴路透社。</p>
<p>這些討論與模型發佈後的採購努力有關。</p>
<p>專注於雲計算和 GPU 租賃服務的公司也在迅速下訂單，路透社引用的兩位消息人士表示，儘管沒有具體公司被確認。</p>
<h2>性能差距與監管環境</h2>
<p>華為的 Ascend 950PR 芯片據稱在性能上顯著優於英偉達的 H20 芯片，後者是英偉達在去年被北京禁止進口之前允許在中國銷售的最先進處理器。</p>
<p>然而，它仍然落後於英偉達更先進的 H200 芯片。</p>
<p>儘管美國和中國都已批准監管，但 H200 芯片的對華運輸尚未開始。</p>
<p>北京和華盛頓之間關於銷售條件的持續分歧延遲了部署，為華為在半導體市場的擴展創造了機會。</p>
<p>950PR 標誌着華為在多年努力爭取國內科技公司大規模訂單後的一個轉折點。</p>
<p>早期客户測試成功，字節跳動和阿里巴巴等公司在一月份收到樣品後開始規劃訂單。</p>
<h2>DeepSeek V4 推動國內 AI 生態系統</h2>
<p>DeepSeek V4 的發佈顯著提升了對國內 AI 硬件的需求，反映出美國出口限制對獲取英偉達頂級芯片的影響。</p>
<p>DeepSeek 決定專門為華為硬件優化其模型，標誌着向中國本土半導體生態系統的轉變。</p>
<p>上週，華為表示，其基於 950 系列芯片構建的 Ascend SuperNode 基礎設施將全面支持 DeepSeek V4 模型。</p>
<p>該公司補充説，其整個產品線已針對 V4 推理進行了調整，推理是指使用訓練模型執行 AI 任務的過程。</p>
<p>在中國芯片製造商中，華為的 Ascend 950 系列目前是唯一能夠支持壓縮數值處理技術的國內解決方案，允許在較低成本下實現更高的計算效率，路透社的報道強調了這一點。</p>
<p>雲服務提供商迅速部署了該模型。</p>
<p>阿里雲的百聯平台在發佈當天就提供了 DeepSeek V4，官方定價提供 V4-Pro 和 V4-Flash 兩個版本。</p>
<p>騰訊雲也在同一天在其 TokenHub 平台上推出了預覽服務，將該模型部署到國內基礎設施及其新加坡網關。</p>
<h2>供應限制可能持續</h2>
<p>雲平台的快速推廣擴大了數百萬用户和開發者的訪問，增加了對 AI 處理能力及其所需芯片的需求。</p>
<p>DeepSeek 為開發者提供 V4 的 75% 折扣，直到 5 月 5 日，並表示 V4-Pro 的定價可能在 2026 年下半年隨着華為超級節點的大規模出貨而下降。</p>
<p>然而，該公司也承認，供應限制將在生產提升之前持續存在。</p>
<p>V4 模型包括兩個版本，V4-Pro 具有 1.6 萬億參數，V4-Flash 具有 2840 億參數，均支持一百萬令牌的上下文窗口。</p>
<p>這些模型在 MIT 開源許可證下發布，允許廣泛的商業使用。</p>
<p>華為計劃在今年出貨約 750,000 個 950PR 芯片，4 月開始大規模生產，預計在 2026 年下半年實現全面出貨，路透社的報道提到。</p>
<p>然而，由於美國對先進芯片製造設備的限制，產量可能仍低於需求，這限制了中國的製造能力。</p>
<p>華為 Ascend 950 芯片需求在 DeepSeek V4 發佈後激增的文章首發於 Invezz。</p>

英偉達

2 倍做多英偉達 ETF - GraniteShares

2 倍做多 BABA ETF - GraniteShares

2 倍做多阿里巴巴 ETF - KraneShares

<p>華為的 Ascend 950 AI 芯片需求激增，原因是 DeepSeek 推出了 V4 型號，字節跳動、騰訊和阿里巴巴等主要中國公司紛紛尋求新訂單。Ascend 950PR 芯片的性能超過了英偉達的 H20，但仍落後於 H200，而 H200 因美中緊張關係面臨發貨延遲。DeepSeek 的 V4 型號推動了國內 AI 生態系統的發展，華為的基礎設施對此提供了全面支持。儘管計劃今年出貨 75 萬個單位，但由於美國對芯片製造設備的限制，供應瓶頸預計將持續</p>

<p><img src="https://imageproxy.pbkrs.com/https://invezz-wp-media.lon1.digitaloceanspaces.com/2026/04/image-1-1777446995.png" alt="" original-src="https://invezz-wp-media.lon1.digitaloceanspaces.com/2026/04/image-1-1777446995.png"/></p>
<p>Demand for Huawei’s Ascend 950 artificial intelligence chips has surged following the release of DeepSeek’s V4 model, which runs on the Shenzhen-based company’s hardware, as per a Reuters report.</p>
<div class="lb-trans"><p>根據路透社的報道，華為的 Ascend 950 人工智能芯片的需求在 DeepSeek 的 V4 模型發佈後激增，該模型運行在這家總部位於深圳的公司的硬件上。</p>
</div><p>Major Chinese internet companies, including ByteDance, Tencent, and Alibaba, are actively reaching out to Huawei to secure new chip orders, the sources told Reuters.</p>
<div class="lb-trans"><p>包括字節跳動、騰訊和阿里巴巴在內的主要中國互聯網公司正在積極聯繫華為，以確保獲得新的芯片訂單，消息人士告訴路透社。</p>
</div><p>The discussions are linked to procurement efforts following the model’s release.</p>
<div class="lb-trans"><p>這些討論與模型發佈後的採購努力有關。</p>
</div><p>Companies focused on cloud computing and GPU rental services are also moving quickly to place orders, according to two sources cited in the Reuters report, although specific firms were not identified.</p>
<div class="lb-trans"><p>專注於雲計算和 GPU 租賃服務的公司也在迅速下訂單，路透社引用的兩位消息人士表示，儘管沒有具體公司被確認。</p>
</div><h2>Performance gap and regulatory landscape</h2>
<div class="lb-trans"><h2>性能差距與監管環境</h2>
</div><p>Huawei’s Ascend 950PR chip is said to significantly outperform Nvidia’s H20 chip, which had been the most advanced processor Nvidia was allowed to sell in China before Beijing blocked its import last year.</p>
<div class="lb-trans"><p>華為的 Ascend 950PR 芯片據稱在性能上顯著優於英偉達的 H20 芯片，後者是英偉達在去年被北京禁止進口之前允許在中國銷售的最先進處理器。</p>
</div><p>However, it still trails Nvidia’s more advanced H200 chip.</p>
<div class="lb-trans"><p>然而，它仍然落後於英偉達更先進的 H200 芯片。</p>
</div><p>Despite regulatory approvals from both the United States and China, shipments of the H200 to China have not begun.</p>
<div class="lb-trans"><p>儘管美國和中國都已批准監管，但 H200 芯片的對華運輸尚未開始。</p>
</div><p>Ongoing disagreements between Beijing and Washington over sales conditions have delayed deployment, creating an opening for Huawei to expand its presence in the semiconductor market.</p>
<div class="lb-trans"><p>北京和華盛頓之間關於銷售條件的持續分歧延遲了部署，為華為在半導體市場的擴展創造了機會。</p>
</div><p>The 950PR marks a turning point for Huawei after years of struggling to secure large-scale orders from domestic technology firms.</p>
<div class="lb-trans"><p>950PR 標誌着華為在多年努力爭取國內科技公司大規模訂單後的一個轉折點。</p>
</div><p>Earlier customer testing was successful, and companies such as ByteDance and Alibaba began planning orders after receiving samples in January,.</p>
<div class="lb-trans"><p>早期客户測試成功，字節跳動和阿里巴巴等公司在一月份收到樣品後開始規劃訂單。</p>
</div><h2>DeepSeek V4 fuels domestic AI ecosystem</h2>
<div class="lb-trans"><h2>DeepSeek V4 推動國內 AI 生態系統</h2>
</div><p>The release of DeepSeek V4 has significantly boosted demand for domestic AI hardware, reflecting the impact of US export restrictions on access to Nvidia’s top-tier chips.</p>
<div class="lb-trans"><p>DeepSeek V4 的發佈顯著提升了對國內 AI 硬件的需求，反映出美國出口限制對獲取英偉達頂級芯片的影響。</p>
</div><p>DeepSeek’s decision to optimise its model specifically for Huawei hardware signals a shift towards China’s homegrown semiconductor ecosystem.</p>
<div class="lb-trans"><p>DeepSeek 決定專門為華為硬件優化其模型，標誌着向中國本土半導體生態系統的轉變。</p>
</div><p>Last week, Huawei stated that its Ascend SuperNode infrastructure, built on the 950 series chips, would fully support DeepSeek V4 models.</p>
<div class="lb-trans"><p>上週，華為表示，其基於 950 系列芯片構建的 Ascend SuperNode 基礎設施將全面支持 DeepSeek V4 模型。</p>
</div><p>The company added that its entire product line had been adapted for V4 inference, referring to the process of executing AI tasks using trained models.</p>
<div class="lb-trans"><p>該公司補充説，其整個產品線已針對 V4 推理進行了調整，推理是指使用訓練模型執行 AI 任務的過程。</p>
</div><p>Among Chinese chipmakers, Huawei’s Ascend 950 series is currently the only domestic solution capable of supporting compressed numerical processing techniques, allowing higher computational efficiency at lower costs, as highlighted in the Reuters report.</p>
<div class="lb-trans"><p>在中國芯片製造商中，華為的 Ascend 950 系列目前是唯一能夠支持壓縮數值處理技術的國內解決方案，允許在較低成本下實現更高的計算效率，路透社的報道強調了這一點。</p>
</div><p>Cloud providers have moved quickly to deploy the model.</p>
<div class="lb-trans"><p>雲服務提供商迅速部署了該模型。</p>
</div><p>Alibaba Cloud’s Bailian platform made DeepSeek V4 available on the day of its release, offering both V4-Pro and V4-Flash versions at official pricing.</p>
<div class="lb-trans"><p>阿里雲的百聯平台在發佈當天就提供了 DeepSeek V4，官方定價提供 V4-Pro 和 V4-Flash 兩個版本。</p>
</div><p>Tencent Cloud also launched preview services on its TokenHub platform the same day, deploying the model across domestic infrastructure and its Singapore gateway.</p>
<div class="lb-trans"><p>騰訊雲也在同一天在其 TokenHub 平台上推出了預覽服務，將該模型部署到國內基礎設施及其新加坡網關。</p>
</div><h2>Supply constraints likely to persist</h2>
<div class="lb-trans"><h2>供應限制可能持續</h2>
</div><p>The rapid rollout across cloud platforms has expanded access to millions of users and developers, increasing demand for AI processing capacity and the chips that power it.</p>
<div class="lb-trans"><p>雲平台的快速推廣擴大了數百萬用户和開發者的訪問，增加了對 AI 處理能力及其所需芯片的需求。</p>
</div><p>DeepSeek is offering developers a 75% discount on V4 until May 5 and indicated that pricing for V4-Pro could decline in the second half of 2026 once Huawei’s supernodes ship at scale.</p>
<div class="lb-trans"><p>DeepSeek 為開發者提供 V4 的 75% 折扣，直到 5 月 5 日，並表示 V4-Pro 的定價可能在 2026 年下半年隨着華為超級節點的大規模出貨而下降。</p>
</div><p>However, the company also acknowledged that supply constraints will persist until production ramps up.</p>
<div class="lb-trans"><p>然而，該公司也承認，供應限制將在生產提升之前持續存在。</p>
</div><p>The V4 models include two versions V4-Pro with 1.6 trillion parameters and V4-Flash with 284 billion parameters both supporting a one-million-token context window.</p>
<div class="lb-trans"><p>V4 模型包括兩個版本，V4-Pro 具有 1.6 萬億參數，V4-Flash 具有 2840 億參數，均支持一百萬令牌的上下文窗口。</p>
</div><p>The models are released under the MIT open-source licence, allowing broad commercial use.</p>
<div class="lb-trans"><p>這些模型在 MIT 開源許可證下發布，允許廣泛的商業使用。</p>
</div><p>Huawei is planning to ship approximately 750,000 units of the 950PR chip this year, with mass production beginning in April and full-scale shipments expected in the second half of 2026, as mentioned in a Reuters report.</p>
<div class="lb-trans"><p>華為計劃在今年出貨約 750,000 個 950PR 芯片，4 月開始大規模生產，預計在 2026 年下半年實現全面出貨，路透社的報道提到。</p>
</div><p>However, output is likely to remain below demand due to US restrictions on advanced chipmaking equipment, which limit China’s manufacturing capabilities.</p>
<div class="lb-trans"><p>然而，由於美國對先進芯片製造設備的限制，產量可能仍低於需求，這限制了中國的製造能力。</p>
</div><p>The post Huawei Ascend 950 chip demand surges after DeepSeek V4 launch appeared first on Invezz</p>
<div class="lb-trans"><p>華為 Ascend 950 芯片需求在 DeepSeek V4 發佈後激增的文章首發於 Invezz。</p>
</div>

Huawei Ascend 950 chip demand surges after DeepSeek V4 launch

- 華為的 Ascend 950 人工智能芯片需求激增，因 DeepSeek V4 模型發佈。  
- 多家中國互聯網公司積極聯繫華為，確保新芯片訂單。  
- 預計華為將於 2026 年下半年全面出貨 750,000 個 950PR 芯片。

在 DeepSeek V4 發佈後，華為昇騰 950 芯片的需求激增