Nvidia GB300 結果共20筆

While the silicon vendors captured the keynote headlines, Taiwan’s hardware manufacturers drew the biggest crowds on the show floor. Their expansive booths featured server racks, advanced cooling systems, robots, and complete computing platforms. The attention was well deserved. Taiwan assembled roughly 90 percent of the world’s AI servers in 2025, and total server shipments are forecast to rise 19.2 percent in 2026, marking the strongest growth in years. Global shipments in the second quarter alone are projected to exceed five million units for the first time. These manufacturers turned Computex into a vast showroom for the physical machines powering the AI boom.

Most of the attention in the AI server boom goes to the companies building the servers. Delta Electronics makes the power and thermal layer they all depend on.The Taoyuan-based firm heads into Computex 2026 with its strongest first-quarter results ever and near-record April revenue. Chairman Ping Cheng recently told investors that customer demand is “very, very large” and current capacity “absolutely insufficient” – language Delta has rarely used in any prior cycle. Its booth theme, “Superior Efficiency, Shaping Sustainable AI,” reflects two decades of quiet mastery over the technologies now most critical to AI: high-efficiency power conversion, advanced cooling, and their tight integration.

Computex 2026 opens on June 2 in Taipei under the banner “AI Together,” and for Taiwan’s server makers the timing is close to perfect. The first quarter delivered record results across the entire group. Hyperscaler orders for Nvidia’s next-generation GB300 rack systems are accelerating into the second half of the year. And a new category of compute that the industry has started calling Physical AI, is opening a fresh growth lane alongside the familiar workloads of training and inference.

When the AI server industry began standardizing on liquid cooling in 2025, Sunonwealth Electric Machine Industry, the Kaohsiung-based fan and motor specialist better known as Sunon, looked like the cooling supply chain’s most exposed name. A year later, that reading has aged badly. Founded in 1980 and listed on the Taiwan Stock Exchange under ticker 2421, Sunon has used its fan and motor heritage as a launchpad rather than a millstone, building out a credible liquid-cooling portfolio while its legacy business continues to grow.

AI伺服器大廠美超微（Supermicro）傳丟失大單！根據Bluefin研究公司最新報告指出，雲端巨頭甲骨文（Oracle）已取消與美超微合作，撤回價值高達14億美元（約新台幣450億元）的輝達（NVIDIA）GB300 NVL72機櫃訂單。市場消息傳出，這筆龐大訂單將由台廠緯穎（6669）接手，消息一出引發產業高度關注。

One of the least remarked features of Taiwan’s AI hardware boom is that it has not been led by the kind of companies the global business press usually celebrates. There are no garage startups here, no twenty-something founders, no venture capital origin myths. The firms riding the AI infrastructure wave, supplying cold plates to GB300 racks and vapor chambers to hyperscaler servers, are for the most part middle-aged industrial companies that have been grinding away in the component trenches for three decades or more.

Wistron Corporation was once best known as one of the world’s largest notebook assemblers. Today, it is one of the biggest stories in the AI hardware industry. In the first quarter of 2026, the Taiwan ODM posted the highest revenue growth rate among the island’s five major server makers, cementing a transformation that has reshaped the company from top to bottom. Wistron reported first-quarter 2026 consolidated revenue of NT$846.3 billion (US$26.5 billion), up an extraordinary 144.3% year-over-year. March alone set a new monthly record at NT$333.0 billion (US$10.4 billion), up 117.7% from a year earlier. February was equally striking, with revenue of NT$284.9 billion (US$8.9 billion), a 177.4% year-over-year jump.

台灣的AI伺服器製造商剛繳出了一份在兩年前看來不可能的第一季成績單。五大代工製造商（ODM）合計創造超過新台幣2兆元（約630億美元）的營收，動能來自輝達（Nvidia）GB300機架平台及美國主要超大規模業者客製化ASIC伺服器的強勁需求浪潮。邁入第二季，問題已不再是需求是否真實，而是台灣能否快速擴產以滿足需求。

Quanta Computer (2382.TW) built its reputation as the world’s largest notebook contract manufacturer. That identity is changing fast. AI server revenue has grown so quickly that notebooks now account for less than 20% of total sales. Through its cloud subsidiary QCT (Quanta Cloud Technology), Quanta has become one of the two largest AI server builders in the world alongside Foxconn, assembling rack systems for the hyperscale data centers powering the global AI buildout.

Taiwan’s AI server makers just delivered a first quarter that would have seemed impossible two years ago. The five leading ODMs generated more than NT$2 trillion (roughly US$63 billion) in revenue, driven by Nvidia’s GB300 rack platform and a surging wave of custom ASIC servers from the major US hyperscalers. The question heading into Q2 is no longer whether demand is real. It is whether Taiwan can ramp up production fast enough to meet it.

Taiwan Semiconductor Manufacturing Company has opened 2026 with a record-shattering first quarter. Revenue for January through March reached NT$1.134 trillion (US$35.9 billion), a 35.1 percent increase over the same period in 2025 and comfortably above the top end of the company’s own guidance of NT$1.13 trillion (US$35.8 billion). The figure eclipses the previous quarterly high of NT$1.05 trillion (US$33.2 billion) set in the fourth quarter of 2025 and extends a run of uninterrupted growth driven almost entirely by AI chip demand.

Foxconn enters the second quarter of 2026 riding its strongest growth in years, powered by explosive demand for AI servers. Yet the company still faces meaningful challenges in its traditional consumer electronics business, which continues to represent a substantial portion of its revenue. Its Q1 results make it clear that AI infrastructure has become the company’s primary growth engine. Whether that engine can fully offset softening demand elsewhere will shape the quarters ahead.

Taiwan’s AI server makers delivered extraordinary first-quarter 2026 results, drive by Nvidia GB300 server shipments to hyperscale cloud customers including Amazon, Google, Meta, Microsoft, and Oracle. The numbers confirm that AI server assembly has become the single strongest growth engine in Taiwan’s technology sector.

鴻海董事長劉揚偉親征輝達主辦的GTC大會，現場不僅展示最新AI基礎架構「NVIDIA GB300 NVL72」，並首次公開美國廠區智慧製造技術，因應AI伺服器需求持續增長，鴻海過去一年擴大相關投資，並宣布將於德克薩斯州、威斯康辛州、加州等地持續擴產，深化與輝達合作。劉揚偉強調，這將協助頂尖客戶於AI競爭中領先，展現台灣在全球AI供應鏈的關鍵地位。

鴻海（2317）於開放運算計劃全球峰會宣布，將攜手NVIDIA導入800 VDC電力架構，首度應用於高雄K-1專案的AI資料中心。這項合作目標推動AI產業升級，提升資料中心能源效率與運算效能，並展現台灣於全球AI供應鏈中的關鍵地位。而鴻海旗下鴻佰科技也展示最新NVIDIA GB300 NVL72及創新水冷解決方案，強化AI基礎設施整合能力，預期將為產業帶來高密度、高效能的AI工廠新時代。

黃仁勳在主題演講中，宣布NVIDIA台灣新總部「星鏈園區」將落腳台北北士科，新總部將成為AI研發與供應鏈整合，結合全球最先進的AI基礎建設，打造台灣第一台巨型AI超級電腦。而黃仁勳也揭揭曉最新的GB300，推理效能提升1.5倍，預計今年第三季量產，供應鏈再度由台灣主導，專家分析，AI技術不斷突破，台灣不只是製造基地，更是未來AI創新的戰略核心。

輝達（NVIDIA）執行長黃仁勳今天在年度GTC技術大會發表主題演講，發表新一代繪圖處理器（GPU）Blackwell Ultra及人工智慧（AI）伺服器GB300，主要合作夥伴鴻海擴大參展，展示GB300 NVL72平台、人形機器人以及數位孿生技術實例。

輝達（NVIDIA）執行長黃仁勳今天在GTC技術大會主題演講指出，現在是加速運算的臨界點，並發表新一代繪圖處理器（GPU）Blackwell Ultra及人工智慧（AI）伺服器GB300，他也重現自己的名言「買越多、省越多」，並形容今年GTC是「AI的超級盃」。

輝達（NVIDIA）GTC大會，即將於本週隆重登場。本次大會備受市場矚目，外界預期輝達將發表下一代AI運算晶片GB300，並揭示最新的技術藍圖與市場策略。根據歷年數據，輝達GTC大會通常為市場帶來強勁動能，歷屆發布的技術與產品皆成為業界關注焦點。

輝達（NVIDIA）近期儘管繳出亮眼的財報與財測，但股價持續遭遇賣壓，3日更是一度重挫9.7%，收盤跌幅達8.69%，以每股114.06美元作收，創下近6個月以來的新低點。據統計，輝達股價在過去一週內下跌了18%，目前市值與1月6日3.66兆美元（約新台幣120兆元）的巔峰相比下跌逾20%，已經蒸發了約9000億美元（約新台幣29兆元）。