PTT職涯區 / Stock (股票)

[新聞] OpenAI 首款自研晶片預計交由台積電生產

看板Stock (股票)作者 (My broken dreams)時間13小時前 (2025/02/10 19:14), 編輯推噓62(63推 1噓 53→)
留言117則, 81人參與, 1小時前最新討論串1/1
原文標題：Exclusive-OpenAI set to finalize first custom chip design this year 原文連結：https://reurl.cc/M6xa8L 發布時間：2025-2-10 記者署名：Anna Tong, Max A. Cherney and Krystal Hu 原文內容： OpenAI is pushing ahead on its plan to reduce its reliance on Nvidia for its chip supply by developing its first generation of in-house artificial-intelligence silicon. The ChatGPT maker is finalizing the design for its first in-house chip in the next few months and plans to send it for fabrication at Taiwan Semiconductor Manufacturing Co, sources told Reuters. The process of sending a first design through a chip factory is called "taping out." OpenAI and TSMC declined to comment. The update shows that OpenAI is on track to meet its ambitious goal of mass production at TSMC in 2026. A typical tape-out costs tens of millions of dollars and will take roughly six months to produce a finished chip, unless OpenAI pays substantially more for expedited manufacturing. There is no guarantee the silicon will function on the first tape out and a failure would require the company to diagnose the problem and repeat the tape-out step. Inside OpenAI, the training-focused chip is viewed as a strategic tool to strengthen OpenAI's negotiating leverage with other chip suppliers, the sources said. After the initial chip, OpenAI's engineers plan to develop increasingly advanced processors with broader capabilities with each new iteration. If the initial tape out goes smoothly, it would enable the ChatGPT maker to mass-produce its first in-house AI chip and potentially test an alternative to Nvidia's chips later this year. OpenAI’s plan to send its design to TSMC this year demonstrates the startup has made speedy progress on its first design, a process that can take other chip designers years longer. Big tech companies such as Microsoft and Meta have struggled to produce satisfactory chips despite years of effort. The recent market rout triggered by Chinese AI startup DeepSeek has also raised questions about whether fewer chips will be needed in developing powerful models in the future. The chip is being designed by OpenAI’s in-house team led by Richard Ho, which had doubled in the past months to 40 people, in collaboration with Broadcom. Ho joined OpenAI more than a year ago from Alphabet's Google where he helped lead the search giant's custom AI chip program. Reuters first reported OpenAI’s plans with Broadcom last year. Ho's team is smaller than the large-scale efforts at tech giants such as Google or Amazon. A new chip design for an ambitious, large-scale program could cost $500 million for a single version of a chip, according to industry sources with knowledge of chip design budgets. Those costs could double to build the necessary software and peripherals around it. Generative AI model makers like OpenAI, Google and Meta have demonstrated that ever-larger numbers of chips strung together in data centers make models smarter, and as a result, they have an insatiable demand for the chips. Meta has said it will spend $60 billion on AI infrastructure in the next year and Microsoft has said it will spend $80 billion in 2025. Currently, Nvidia’ s chips are the most popular and hold a market share of roughly 80%. OpenAI is itself participating in the $500 billion Stargate infrastructure program announced by U.S. President Donald Trump last month. But rising costs and dependence on a single supplier have led major customers such as Microsoft, Meta and now OpenAI to explore in-house or external alternatives to Nvidia’s chips. OpenAI's in-house AI chip, while capable of both training and running AI models, will initially be deployed on a limited scale, and primarily for running AI models, the sources said. The chip will have a limited role within the company's infrastructure. To build out an effort as comprehensive as Google or Amazon's AI chip program, OpenAI would have to hire hundreds of engineers. TSMC is manufacturing OpenAI's AI chip using its advanced 3-nanometer process technology. The chip features a commonly used systolic array architecture with high-bandwidth memory (HBM) - also used by Nvidia for its chips - and extensive networking capabilities, sources said. OpenAI 正加速自研 AI 晶片，以降低對 NVIDIA 的依賴，預計今年內完成設計並送交台 積電（TSMC）生產，目標 2026 年量產。該晶片主要用於 AI 推論，採 3 奈米製程、高 頻寬記憶體（HBM）與 systolic array 架構，具備強大網路能力。開發團隊由前 Google AI 晶片專家 Richard Ho 領導，並與 Broadcom 合作，團隊已擴增至 40 人。開 發一款高階 AI 晶片成本可能達 5 億美元。由於 AI 晶片需求激增，Microsoft、Meta 及 OpenAI 等企業正尋求 NVIDIA 以外的解決方案。若首款晶片 tape-out 成功， OpenAI 可能於今年內測試其作為 NVIDIA 替代方案。此舉不僅提升議價能力，也有助於 降低 AI 運算成本，強化 OpenAI 的技術自主權。 心得/評論： OpenAI預計晶片交給台積電代工 新聞消息一出 夜盤直接嘎上天 世界的G怎麼輸？ 空蛙表示： -- ※ 發信站: 批踢踢實業坊(ptt.cc), 來自: 218.32.44.219 (臺灣) ※ 文章網址: https://www.ptt.cc/bbs/Stock/M.1739186057.A.A4C.html
→
02/10 19:15, 13小時前 , 1F
不然要給誰
02/10 19:15, 1F
推
02/10 19:15, 13小時前 , 2F
不然你要給INTEL嗎?
02/10 19:15, 2F
推
02/10 19:16, 13小時前 , 3F
真的只需要買2330就夠了
02/10 19:16, 3F
推
02/10 19:16, 13小時前 , 4F
不然還有誰能做
02/10 19:16, 4F
推
02/10 19:17, 13小時前 , 5F
AVGO設計？
02/10 19:17, 5F
推
02/10 19:17, 13小時前 , 6F
要前往STARGATE了
02/10 19:17, 6F
→
02/10 19:17, 13小時前 , 7F
deepseek:
02/10 19:17, 7F
推
02/10 19:17, 13小時前 , 8F
資本支出沒有調升的話這邊多別的地方就會少吧
02/10 19:17, 8F
推
02/10 19:18, 13小時前 , 9F
怎麼不早說?
02/10 19:18, 9F
→
02/10 19:18, 13小時前 , 10F
02/10 19:18, 10F
→
02/10 19:18, 13小時前 , 11F
其實不說也知道
02/10 19:18, 11F
推
02/10 19:19, 13小時前 , 12F
輝達崩
02/10 19:19, 12F
推
02/10 19:19, 13小時前 , 13F
02/10 19:19, 13F
推
02/10 19:19, 13小時前 , 14F
真的有多蛙被這種白癡東西騙進去接盤？
02/10 19:19, 14F
→
02/10 19:19, 13小時前 , 15F
AVGO未來趨勢大贏家
02/10 19:19, 15F
→
02/10 19:20, 13小時前 , 16F
世界的大GG，傳家寶
02/10 19:20, 16F
→
02/10 19:21, 13小時前 , 17F
阿不給台積電就剩三星給intl啦
02/10 19:21, 17F
→
02/10 19:23, 13小時前 , 18F
要收關稅了喔
02/10 19:23, 18F
→
02/10 19:23, 13小時前 , 19F
只要台積解決了關稅股價就水到渠成
02/10 19:23, 19F
推
02/10 19:23, 13小時前 , 20F
AZ AZ AZ
02/10 19:23, 20F
→
02/10 19:25, 13小時前 , 21F
蛋糕沒有變大啊，是在嗨啥
02/10 19:25, 21F
推
02/10 19:25, 13小時前 , 22F
師傅：沒什麼
02/10 19:25, 22F
推
02/10 19:26, 12小時前 , 23F
不然要給誰？
02/10 19:26, 23F
推
02/10 19:26, 12小時前 , 24F
直接找工廠拿貨比較便宜的概念
02/10 19:26, 24F
推
02/10 19:26, 12小時前 , 25F
蛋糕一樣大，但少了一個人在中間賺一手啊
02/10 19:26, 25F
推
02/10 19:29, 12小時前 , 26F
少了一個仲介，賣房的賺到，買房也賺到
02/10 19:29, 26F
推
02/10 19:30, 12小時前 , 27F
亞歷桑那亞歷桑那亞歷桑那
02/10 19:30, 27F
推
02/10 19:31, 12小時前 , 28F
那就是創意受惠了
02/10 19:31, 28F
推
02/10 19:31, 12小時前 , 29F
剛台積博通一起噴
02/10 19:31, 29F
→
02/10 19:32, 12小時前 , 30F
還在創意 剛博通已經先動
02/10 19:32, 30F
推
02/10 19:32, 12小時前 , 31F
不然給誰做？記得後面乖乖排隊
02/10 19:32, 31F
推
02/10 19:32, 12小時前 , 32F
（請問價格）（貴或便宜？）（嘻嘻。）
02/10 19:32, 32F
推
02/10 19:33, 12小時前 , 33F
當初不是聽說要用A16 太貴了？
02/10 19:33, 33F
→
02/10 19:33, 12小時前 , 34F
每個巨頭一定都想過跳過nv找台積 但nv現在還是龍頭
02/10 19:33, 34F
推
02/10 19:33, 12小時前 , 35F
創意明天鎖死
02/10 19:33, 35F
推
02/10 19:34, 12小時前 , 36F
NV晶片有那麼好做 AMD早就做出來跟NV競爭了
02/10 19:34, 36F
推
02/10 19:34, 12小時前 , 37F
當NV AMD這麼多年的發展是吃素的？最後還是灰頭土
02/10 19:34, 37F
→
02/10 19:34, 12小時前 , 38F
臉回去找NV
02/10 19:34, 38F
→
02/10 19:34, 12小時前 , 39F
創意要噴爛了嗎
02/10 19:34, 39F
還有 38 則推文
→
02/10 21:25, 11小時前 , 78F
要付台積電5成毛利 但外購要先付設計商5成毛利 他
02/10 21:25, 78F
→
02/10 21:25, 11小時前 , 79F
們再付GG5成毛利 等於一個晶片成本如果是100鎂 我
02/10 21:25, 79F
→
02/10 21:25, 11小時前 , 80F
自製的可以200買到 跟AMD買要400 跟NV買可能要500
02/10 21:25, 80F
→
02/10 21:25, 11小時前 , 81F
如果自製的效率不要輸NV太多就有搞頭(簡單講就是如
02/10 21:25, 81F
→
02/10 21:25, 11小時前 , 82F
果能做到AMD水平就可以考慮自製 但如果要往外買AMD
02/10 21:25, 82F
→
02/10 21:25, 11小時前 , 83F
水平是不夠的) 但現在先進製程的入門門檻超高 只有
02/10 21:25, 83F
→
02/10 21:25, 11小時前 , 84F
用量超大的巨頭公司有機會自己搞 其他規模小一點的
02/10 21:25, 84F
→
02/10 21:25, 11小時前 , 85F
還是只能乖乖買NV就是了
02/10 21:25, 85F
→
02/10 21:28, 10小時前 , 86F
現在就沒有其它家可以做 外星人放棄了INTEL 可惜啊
02/10 21:28, 86F
噓
02/10 21:29, 10小時前 , 87F
這根本廢話 不然誰還能做？
02/10 21:29, 87F
→
02/10 21:30, 10小時前 , 88F
自研是不容易啊 但你不投資 就要面對老黃漲價的臉色
02/10 21:30, 88F
推
02/10 21:35, 10小時前 , 89F
推論 有機會用低價的硬體 自己設計也不一定比量產
02/10 21:35, 89F
→
02/10 21:35, 10小時前 , 90F
划算 現在是nv利潤高 自己做可能比較便宜也避免被
02/10 21:35, 90F
→
02/10 21:35, 10小時前 , 91F
卡技術 改天說不定被搞到更低價 推論用成熟製程ed
02/10 21:35, 91F
→
02/10 21:35, 10小時前 , 92F
ge就處理完了 外接pcie就處理好了
02/10 21:35, 92F
→
02/10 21:35, 10小時前 , 93F
賣鏟子的根本不擔心誰成為AI霸主
02/10 21:35, 93F
→
02/10 21:44, 10小時前 , 94F
大不了不放米國
02/10 21:44, 94F
→
02/10 21:47, 10小時前 , 95F
訂單多 股東會宣布蓋美國二廠
02/10 21:47, 95F
推
02/10 21:48, 10小時前 , 96F
沒用了
02/10 21:48, 96F
推
02/10 21:49, 10小時前 , 97F
自製只要給台積電五成毛利，那asic廠商免費？
02/10 21:49, 97F
推
02/10 21:55, 10小時前 , 98F
問OPENAI，XAI 如何投資，小韮們 唯一選擇私募公募
02/10 21:55, 98F
→
02/10 21:55, 10小時前 , 99F
兼得的Cross ETF - XOVR。
02/10 21:55, 99F
推
02/10 22:03, 10小時前 , 100F
GG這麼保密 怎會流出這消息
02/10 22:03, 100F
推
02/10 22:03, 10小時前 , 101F
英特爾：美國同胞救救我
02/10 22:03, 101F
推
02/10 22:15, 10小時前 , 102F
Open 要降低 NV需求，自研晶片，這2家都要找gg，除
02/10 22:15, 102F
→
02/10 22:15, 10小時前 , 103F
了一開始要試研期間對gg總投片數有增加，可是最後gg
02/10 22:15, 103F
→
02/10 22:15, 10小時前 , 104F
總量也沒增加多少吧
02/10 22:15, 104F
推
02/10 22:17, 10小時前 , 105F
這個新聞要是再以往發佈是利多，可是在現時期，變
02/10 22:17, 105F
→
02/10 22:17, 10小時前 , 106F
成又給三普勒索的藉口
02/10 22:17, 106F
推
02/10 22:35, 9小時前 , 107F
不管哪一家都要找GG GG怎麼輸？
02/10 22:35, 107F
推
02/10 23:06, 9小時前 , 108F
有些“工作者”真是說鬼話不嫌累喔 ＧＧ最擔心的
02/10 23:06, 108F
→
02/10 23:07, 9小時前 , 109F
並非川普 而是太過單一的客戶分佈 有人來跟ＮＶ搶
02/10 23:07, 109F
→
02/10 23:07, 9小時前 , 110F
ＡＩ晶片市場 這對ＧＧ是再好不過的事情
02/10 23:07, 110F
→
02/10 23:08, 9小時前 , 111F
因為 這就讓晶片製造的供需更往賣方市場推了
02/10 23:08, 111F
推
02/10 23:23, 9小時前 , 112F
也就是說 不管川普政府要玩什麼花樣 ＧＧ的籌碼只會
02/10 23:23, 112F
→
02/10 23:23, 9小時前 , 113F
因其於市場更全面的重要性提升而更多
02/10 23:23, 113F
推
02/10 23:24, 9小時前 , 114F
說好救救intel呢？
02/10 23:24, 114F
推
02/11 03:02, 5小時前 , 115F
好喔~噴
02/11 03:02, 115F
→
02/11 06:59, 1小時前 , 116F
印tel就扶不起的阿斗 想救就要花大錢 馬監國跟川普
02/11 06:59, 116F
→
02/11 06:59, 1小時前 , 117F
怎麼可能多拿錢給它？
02/11 06:59, 117F
‣ 更多 DrowningPool 的文章
文章代碼(AID): #1dgT-9fC (Stock)
Stock 近期熱門文章
更多 近期熱門文章 >>
PTT職涯區 即時熱門文章
更多 即時熱門文章 >>
‣ 更多 DrowningPool 的文章
文章代碼(AID): #1dgT-9fC (Stock)