跳到主要內容

Nvidia併購Hugging Face評測:129億美元買下開源AI門面 | Nvidia Hugging Face Acquisition Review: $12.9B Buys Open-Source AI Hub

By Kit 小克 | AI Tool Observer | 2026-08-28

🇹🇼 Nvidia併購Hugging Face評測:129億美元買下開源AI門面

Nvidia併購Hugging Face的消息在2026年8月27日曝光:這家GPU巨頭同意以約129億美元買下有「AI界GitHub」之稱的開源模型平台Hugging Face。這筆交易金額約是該公司2023年估值(45億美元)的三倍,也是年營收(約1.5億美元)的80倍,換算下來是一筆極高溢價的收購案,也是本週AI圈最熱門的話題。

從拒絕5億美元投資到接受129億收購

最耐人尋味的轉折是:Hugging Face九個月前才拒絕Nvidia的5億美元投資(當時估值僅70億美元),理由是不想讓單一晶片廠商擁有過大話語權、影響平台中立性。如今卻同意以近三倍估值直接被整碗端走,落差高達59億美元。這中間發生了什麼,外界只能猜測是資金壓力、股東變現需求,或是Nvidia開出了無法拒絕的條件。

為什麼Nvidia要買Hugging Face?

Hugging Face本身賺不了多少錢,但它掌握的是整個開源AI生態系的入口——全球開發者上傳模型、資料集、跑基準測試都在這裡進行。對Nvidia來說,晶片賣得再好,終究要有人在上面跑模型;買下這個「開發者聚集地」等於買下了未來的用戶黏著度與資料流向,這是光靠晶圓廠做不出來的資產。

開源社群最擔心什麼

  • 平台中立性:Hugging Face過去標榜對所有硬體廠商(AMD、Google TPU、Nvidia等)一視同仁,被單一晶片商收購後能否維持中立,是社群最大疑慮
  • 競爭對手觀望:AMD、Intel等競品陣營是否還願意讓自家最佳化模型放上「敵營旗下」的平台,值得觀察
  • 商業化壓力:129億美元的收購成本,Nvidia勢必要想辦法從Hugging Face身上榨出更多獲利,免費開源精神能撐多久是個問號

對開發者實際有什麼影響?

短期內,日常使用Hugging Face下載模型、跑Transformers函式庫應該不會有立即變化——交易還在協議階段,尚未正式完成,後續也可能面臨反壟斷審查。但中長期來看,如果你的工作流高度依賴Hugging Face Hub做模型托管或資料集分享,現在是時候思考「單一供應商風險」:留意是否有替代託管方案(如自架GitLab、私有S3+Git LFS),避免未來平台政策轉向時措手不及。

這起Nvidia併購Hugging Face案,某種程度上重演了微軟收購GitHub的劇本——一家基礎設施巨頭買下開發者社群的「客廳」。GitHub最終保住了獨立運營,但社群信任花了好幾年才慢慢重建。這次會不會有同樣結局,現在下定論還太早。

好不好用,試了才知道。


🇺🇸 Nvidia Hugging Face Acquisition Review: $12.9B Buys Open-Source AI Hub

The Nvidia Hugging Face acquisition broke on August 27, 2026: the GPU giant agreed to buy the open-source model platform known as the "GitHub of AI" for roughly $12.9 billion. That price is nearly triple Hugging Face's 2023 valuation of $4.5 billion, and about 80x its estimated $150 million annual revenue — a steep premium that made this the biggest AI story of the week.

From Rejecting $500M to Accepting $12.9B

The strangest part of this story: Hugging Face turned down a $500 million Nvidia investment just nine months ago, when it was valued at $7 billion, specifically to avoid letting a single chipmaker gain outsized influence over the platform's neutrality. Now it's agreeing to be bought outright at nearly triple that valuation — a $5.9 billion swing. What changed in between is anyone's guess: cash pressure, investor liquidity needs, or simply an offer too large to refuse.

Why Does Nvidia Want Hugging Face?

Hugging Face itself doesn't make much money, but it controls something more valuable: the front door to the entire open-source AI ecosystem. Every developer uploading models, sharing datasets, or running benchmarks passes through it. For Nvidia, selling chips is only half the equation — someone still has to run models on top of them. Owning the developer hub means owning future stickiness and data flow, something no fab can manufacture on its own.

What the Open-Source Community Is Worried About

  • Platform neutrality: Hugging Face has long positioned itself as vendor-neutral across AMD, Google TPU, and Nvidia hardware. Whether that holds under single-chipmaker ownership is the community's top concern
  • Rival hesitation: Will AMD, Intel, and other competitors still want their optimized models hosted on a platform now owned by a rival
  • Monetization pressure: At a $12.9B price tag, Nvidia will need to extract real revenue from Hugging Face — how long the free, open-source ethos survives that pressure is an open question

What This Actually Means for Developers

In the short term, day-to-day usage — downloading models, running the Transformers library — shouldn't change immediately. The deal is still at the agreement stage, not closed, and will likely face antitrust scrutiny. But longer term, if your workflow depends heavily on Hugging Face Hub for model hosting or dataset sharing, now's a reasonable time to think about single-vendor risk: check whether you have a fallback (self-hosted GitLab, private S3 + Git LFS) in case platform policy shifts later.

This Nvidia Hugging Face acquisition echoes Microsoft's purchase of GitHub in some ways — an infrastructure giant buying the developer community's living room. GitHub eventually preserved its independence, but it took years to rebuild trust. Whether this plays out the same way is too early to call.

好不好用,試了才知道 — you won't know until you've actually used it.

Sources / 資料來源

延伸閱讀 / Related Articles


AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends

留言

這個網誌中的熱門文章

Google Ironwood TPU v7 推理專用晶片解析:效能追平 NVIDIA、成本低 44%,AI 晶片戰爭正式開打 | Google Ironwood TPU v7 Explained: Matching NVIDIA Performance at 44% Lower Cost — The AI Chip War Heats Up

Claude Code 實測:AI 幫你寫程式到底行不行? | Claude Code Review: Can AI Really Code for You?

Cursor vs GitHub Copilot vs Claude Code:AI 程式助手大比拼 | AI Coding Assistants Compared: Cursor vs GitHub Copilot vs Claude Code