跳到主要內容

Tenstorrent評測:拒絕英特爾收購,改融資15億美元 | Tenstorrent Review: Turns Down Intel, Raises $1.5B

By Kit 小克 | AI Tool Observer | 2026-10-21

🇹🇼 Tenstorrent評測:拒絕英特爾收購,改融資15億美元

Tenstorrent評測來了。這家由晶片傳奇工程師Jim Keller領軍的AI晶片新創,最近做了一個讓矽谷跌破眼鏡的決定:拒絕英特爾(Intel)與高通(Qualcomm)開出的收購邀約,轉頭自己對外募了15億美元,把估值推上62億美元。這篇文章帶你搞懂Tenstorrent在賭什麼,以及這對想繞過Nvidia蓋AI算力的團隊意味著什麼。

Tenstorrent是什麼公司?

Tenstorrent是一家開發AI訓練與推論晶片的新創,核心賣點是用開放架構(RISC-V)打造能跟Nvidia GPU打對台的方案,主力產品包括Wormhole、Blackhole晶片,以及把多顆晶片串成機櫃等級系統的Galaxy伺服器。公司執行長Jim Keller曾在AMD、Tesla、Apple主導過關鍵晶片架構,業界份量不小。

為什麼拒絕英特爾、高通的收購?

外電報導,英特爾與高通先前都對Tenstorrent釋出收購興趣,市場一度傳出估值喊到50億美元左右。但Keller最後選擇不賣,改為對外募資。關鍵轉折點據報是今年5月的一場內部demo:Tenstorrent的Galaxy伺服器在生產規模下,成功完整跑完DeepSeek的全尺寸模型,證明硬體不只是實驗室跑分道具,是真的能上線扛流量。

這次募資規模多大?

Tenstorrent這輪募得15億美元,新估值來到62億美元,資金將用於擴大AI晶片生產,並持續開發開放權重(open-weight)的高階可擴展AI技術。

小克的實測心得

老實說,AI晶片新創喊「挑戰Nvidia」的故事聽太多了,大部分最後都死在PPT階段。Tenstorrent這次比較有說服力的地方,是它不是靠合成benchmark跑分吹噓,而是真的把DeepSeek這種主流大模型,在自家Galaxy伺服器上完整跑起來。對開發者來說,這代表雲端供應商名單上未來可能真的會多一個Nvidia以外的選項,但離「一般團隊也能買到、好裝好用」還有一段距離,值不值得等,目前還是個問號。

好不好用,試了才知道。


🇺🇸 Tenstorrent Review: Turns Down Intel, Raises $1.5B

Tenstorrent just made a move that surprised a lot of people in the chip world: the AI chip startup turned down acquisition interest from both Intel and Qualcomm, and instead raised $1.5 billion on its own at a $6.2 billion valuation. Here's what's actually going on, and why it matters if you care about where AI compute is headed beyond Nvidia.

What Is Tenstorrent?

Tenstorrent is an AI chip startup building training and inference hardware around an open RISC-V architecture, positioning itself as an alternative to Nvidia GPUs. Its lineup includes the Wormhole and Blackhole chips, plus Galaxy servers that chain multiple chips into rack-scale systems. CEO Jim Keller is the veteran chip architect behind major designs at AMD, Tesla, and Apple.

Why Did Tenstorrent Turn Down Intel and Qualcomm?

Reports say both Intel and Qualcomm expressed acquisition interest, with some estimates valuing Tenstorrent at around $5 billion. Keller chose to stay independent instead. The reported turning point came in May, when Tenstorrent's Galaxy servers successfully ran the full-size DeepSeek model at production scale — proof the hardware works beyond lab benchmarks.

How Big Was the New Funding Round?

Tenstorrent raised $1.5 billion, pushing its valuation to $6.2 billion. The company says the money will go toward scaling AI chip manufacturing and continuing development of its open-weight, high-end scalable AI technology.

Kit's Honest Take

We've heard the "next Nvidia challenger" pitch a hundred times, and most of those startups die somewhere around the slide deck. What makes Tenstorrent's story slightly more credible is that it didn't lean on synthetic benchmarks — it actually ran a mainstream large model like DeepSeek in production on its own hardware. That's a real signal for developers hoping for a non-Nvidia option in the cloud someday. But "proven in a demo" is still a long way from "easy to buy and deploy," so whether it's worth betting on is still an open question.

好不好用,試了才知道。(You only know if it's good after you've actually tried it.)

Sources / 資料來源

常見問題 FAQ

Tenstorrent是做什麼的公司?

Tenstorrent是一家AI晶片新創,開發以開放RISC-V架構為基礎的AI訓練與推論晶片,執行長是知名晶片架構師Jim Keller。

Tenstorrent這次估值多少?

這輪募資15億美元後,Tenstorrent估值來到62億美元,較先前的收購傳聞估值明顯提升。

為什麼Intel和Qualcomm想收購Tenstorrent?

外媒報導顯示兩家公司都對其AI晶片技術感興趣,但具體收購條件未公開,最終Tenstorrent選擇不賣。

一般開發者現在能用Tenstorrent的晶片嗎?

目前Tenstorrent的Galaxy伺服器與晶片主要面向企業與雲端供應商,一般開發者的取得管道與生態系仍在建立中。

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