原文標題： Nvidia’s new ‘robot brain’ goes on sale for $3,499 as company targets robotics for growth 原文連結：https://tinyurl.com/2pnvcd2h 發布時間： Published Mon, Aug 25 202511:00 AM EDTUpdated Mon, Aug 25 202512:11 PM EDT 記者署名：Kif Leswing 原文內容： Nvidia announced Monday that its latest robotics chip module, the Jetson AGX Thor, is now on sale for $3,499 as a developer kit. 輝達週一宣布，其最新的機器人晶片模組 Jetson AGX Thor 已開售，開發套件價格為 3, 499 美元。 The company calls the chip a “robot brain.” The first kits ship next month, Nvidia said last week, and the chips will allow customers to create robots. 公司將這款晶片稱為「機器人大腦」。首批套件將於下個月出貨，這些晶片將允許客戶開 發並打造機器人。 After a company uses the developer kit to prototype their robot, Nvidia will sell Thor T5000 modules that can be installed in production-ready robots. If a company needs more than 1,000 Thor chips, Nvidia will charge $2,999 per module. 企業在使用開發套件打造原型機後，輝達將出售 Thor T5000 模組，可安裝於量產機器人 。若企業採購超過 1,000 顆 Thor 晶片，單價將降至 2,999 美元。 CEO Jensen Huang has said robotics is the company’s largest growth opportunity outside of artificial intelligence, which has led to Nvidia’s overall sales more than tripling in the past two years. CEO 黃仁勳表示，除了人工智慧之外，機器人是公司最大的成長機會。過去兩年，AI 已 帶動輝達總營收 翻了三倍以上。 “We do not build robots, we do not build cars, but we enable the whole industry with our infrastructure computers and the associated software,” said Deepu Talla, Nvidia’s vice president of robotics and edge AI. 輝達副總裁（負責機器人與邊緣 AI）Deepu Talla 表示：「我們不造機器人，也不造汽 車，但我們提供整個產業所需的基礎電腦與軟體。」 The Jetson Thor chips are based on a Blackwell graphics processor, which is Nvidia’s current generation of technology used in its AI chips, as well as its chips for computer games. Jetson Thor 晶片基於 Blackwell GPU 架構，這是輝達現行的最新一代技術，用於 AI 晶片與遊戲顯卡。 Nvidia said that its Jetson Thor chips are 7.5 times faster than its previous generation. That allows them to run generative AI models, including large language models and visual models that can interpret the world around them, which is essential for humanoid robots. The Jetson Thor chips are equipped with 128GB of memory, which is essential for big AI models. 輝達表示，Jetson Thor 晶片速度比前一代快 7.5 倍，能運行生成式 AI 模型，包括 大 型語言模型（LLMs） 與 電腦視覺模型，能夠解讀周遭環境，這對 人形機器人 至關重要 。晶片內建 128GB 記憶體，以支援大型 AI 模型。 Companies including Agility Robotics, Amazon, Meta and Boston Dynamics are using its Jetson chips, Nvidia said. Nvidia has also invested in robotics companies such as Field AI. 輝達指出，包括 Agility Robotics、亞馬遜、Meta、Boston Dynamics 都在使用其 Jetson 晶片。輝達同時投資了 Field AI 等機器人公司。 However, robotics remains a small business for Nvidia, accounting for about 1% of the company’s total revenue, despite the fact that it has launched several new robot chips since 2014. But it’s growing fast. 然而，機器人業務在輝達整體營收中仍僅佔 約 1%，即使自 2014 年以來已推出多款機器 人晶片。不過，其成長速度非常快。 Nvidia recently combined its business units to group its automotive and robotics divisions into the same line item. That unit reported $567 million in quarterly sales in May, which represented a 72% increase on an annual basis. 輝達最近將 汽車業務與機器人業務 合併為同一部門。今年 5 月該部門季度銷售額達 5. 67 億美元，年增 72%。 The company said its Jetson Thor chips can be used for self-driving cars as well, especially from Chinese brands. Nvidia calls its car chips Drive AGX, and while they are similar to its robotics chips, they run an operating system called Drive OS that’s been tuned for automotive purposes. 公司表示，Jetson Thor 晶片也可用於 自駕車，特別是中國品牌。輝達的汽車晶片稱為 Drive AGX，雖然與機器人晶片相似，但搭載專為汽車調校的 Drive OS 作業系統。 心得/評論： 全球機器人市場規模 根據多家研調（MarketsandMarkets、Allied Market Research 等），全球機器人市場 2024 約 500~600 億美元，預計到 2030 可達 2000~2500 億美元（CAGR 20%+）。 其中 人形機器人 + 服務型機器人 預期增速最快，因 AI + 自動化需求爆發。 AI 晶片在機器人領域的需求 機器人需要 高性能 AI 晶片，尤其是能支援 LLM、視覺、語音控制的邊緣運算。 這意味著單台機器人可能需要一顆甚至多顆高端模組（如 Thor T5000）。 NVIDIA 的優勢 目前已經有 Amazon、Meta、Boston Dynamics 這類龍頭採用，形成生態優勢。 Thor 晶片與自駕車晶片共享架構，跨領域規模效應顯著。 軟硬體整合（CUDA、Isaac、Omniverse 模擬平台），形成高門檻。 NVIDIA 可能市佔率 短期（2025~2027）：NVIDIA 在機器人 AI 晶片市場有望拿下 50%+，因缺乏直接對手（ AMD、Intel、Qualcomm 在高階 AI 應用仍落後）。 中期（2028~2030）：隨著中國本土 GPU / AI 晶片崛起，NVIDIA 全球市佔率可能回落到 35~45%，但在歐美與高階應用仍保持領先。 -- ※ 發信站: 批踢踢實業坊(ptt.cc), 來自: 150.117.197.145 (臺灣) ※ 文章網址: https://webptt.cc/bbs/Stock/M.1756166927.A.CC3.html