AI推理雲估值大戰：Fireworks 175億、Baseten 130億同月爆發 | AI Inference Cloud Valuations Explode: Fireworks $17.5B, Baseten $13B

7月 28, 2026

By Kit 小克 | AI Tool Observer | 2026-07-29

🇹🇼 AI推理雲估值大戰：Fireworks 175億、Baseten 130億同月爆發

AI推理雲市場在2026年7月迎來罕見的同月雙爆發：Fireworks AI 拿下15億美元D輪，估值衝上175億美元；短短三週前，Baseten 才剛拿下同樣15億美元的F輪，估值來到130億美元。這不是巧合，而是「跑模型比訓模型更賺錢」這件事，終於被創投用真金白銀確認了。

什麼是AI推理雲？為什麼突然值這麼多錢

推理（inference）指的是模型訓練完成後，實際回應使用者請求的運算過程——每次你跟 ChatGPT 對話、Cursor 幫你補完程式碼，背後都是一次推理。Fireworks AI 和 Baseten 做的事，是幫企業把開源或客製化模型，用比 OpenAI、Anthropic 官方 API 更便宜、更快的方式跑起來。

這次融資最值得留意的不是估值數字，而是背後的營收成長曲線：

Fireworks AI：年化營收突破10億美元，年增5倍；平台每日處理超過40兆 token，一年前只有15兆。客戶名單包括 Uber、Shopify、Cursor、Harvey 等重量級企業與AI原生新創。
Baseten：過去一年營收成長20倍，推理量成長40倍。投資方包括 Altimeter、Conviction、Spark Capital 等。

對開發者跟企業意味著什麼

這場AI推理雲大戰的實際意義,是開發者現在有真正能打的替代方案——不必事事仰賴 OpenAI 或 Anthropic 的官方 API,可以把開源模型或微調過的客製模型部署在 Fireworks、Baseten 這類專業推理平台上,壓低單次呼叫成本。這也解釋了為什麼 Nvidia 會親自下場投資 Fireworks：晶片賣得再好,終究要有人把運算用起來變現。

但誠實講,15-17倍營收的估值不便宜,而且推理這一層正被兩面夾擊——上游是自建推理服務的模型廠(OpenAI、Anthropic都在擴大自家API的性價比),下游是雲端三巨頭隨時可能砸錢自建競品。這些公司的護城河,現階段還是「比別人快、比別人便宜」的執行力,不是技術門檻。市場夠不夠大,養得起這麼多家175億美元等級的公司,還有待觀察。

好不好用，試了才知道。

🇺🇸 AI Inference Cloud Valuations Explode: Fireworks $17.5B, Baseten $13B

The AI inference cloud market just had a rare double-explosion in the same month: Fireworks AI closed a $1.5B Series D at a $17.5 billion valuation, just three weeks after Baseten raised the same $1.5B in a Series F at a $13 billion valuation. This isn't coincidence — it's venture capital finally putting real money behind the thesis that running models profitably beats training them.

What Is AI Inference Cloud, and Why the Sudden Valuations

Inference is the computation that happens after a model is trained, every time it actually answers a request — every ChatGPT reply, every Cursor autocomplete. Fireworks AI and Baseten specialize in running open-source or custom-fine-tuned models cheaper and faster than going straight through OpenAI's or Anthropic's official APIs.

The valuation headline matters less than the revenue curve underneath it:

Fireworks AI: crossed $1B in annualized revenue, up 5x year-over-year. The platform now processes over 40 trillion tokens per day, up from 15 trillion a year ago. Customers include Uber, Shopify, Cursor, and Harvey.
Baseten: revenue grew 20x and inference volume grew 40x over the past year, backed by Altimeter, Conviction, and Spark Capital.

What This Means for Developers and Companies

The practical upshot of this AI inference cloud race is that developers now have real alternatives to defaulting to OpenAI or Anthropic's official APIs — open-weight or fine-tuned custom models can run on specialized platforms like Fireworks or Baseten at a fraction of the per-call cost. It also explains why Nvidia personally invested in Fireworks: selling chips is only half the business: someone has to turn that compute into revenue.

Honestly though, valuations at 15-17x revenue aren't cheap, and the inference layer is being squeezed from both sides — model labs like OpenAI and Anthropic are aggressively improving their own API pricing, while the big three clouds could build competing services whenever they decide to. Right now the moat for these companies is pure execution — faster, cheaper — not a technical barrier. Whether the market is big enough to sustain multiple $17.5B-class companies here is still an open question.

You won't know until you try 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

搜尋此網誌

AI小貼士