OpenAI營收評測：700億變500億，Nvidia應聲下跌 | OpenAI Revenue Review: $70B Revised to $50B, Markets Slide

10月 10, 2026

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

🇹🇼 OpenAI營收評測：700億變500億，Nvidia應聲下跌

OpenAI營收這週爆出大新聞：原本市場傳的年化營收700億美元，被《金融時報》證實其實只有約500億美元，整整少了200億。消息一出，Nvidia、Oracle、CoreWeave等AI概念股全部應聲下跌。但真相沒有想像中戲劇化——問題不是客戶變少了，而是算法不一樣。

關鍵不是客戶變少，是算法打架

這次OpenAI營收落差的根源，出在OpenAI跟Anthropic計算營收的方式完全不同。假設客戶透過Microsoft Azure花1美元買OpenAI的token用量，OpenAI只會把自己分到的那約20美分算進營收，剩下的80美分算是Azure的收入。

但Anthropic不是這樣算的：它會把整整1美元都列成營收,再把雲端夥伴分走的部分記成「銷售成本」。同樣一筆生意，兩家公司帳上看起來差了5倍。市場原本用Anthropic的算法去推算OpenAI的700億營收,等於是拿兩套不同的尺在比身高。

股價為什麼還是跌了

既然是算法問題不是需求問題,股價為什麼還是跌？答案很直接：AI泡沫的敘事這一年一直很脆弱,華爾街已經把700億這個數字寫進估值模型。一旦發現基準數字錯了,不管背後原因是什麼,市場的第一反應就是賣壓先出再問問題——這也側面說明投資人對AI基礎設施支出的信心,其實比表面上更緊繃。

對一般用戶跟開發者的影響

ChatGPT訂閱價格短期不會變,這次是會計口徑問題,不是現金流出問題
企業客戶若透過Azure等雲端夥伴採購,帳面營收歸屬可能比你想的複雜,簽約前問清楚SLA比糾結新聞數字更實際
這次事件再次提醒:AI公司的營收數字橫向比較時,務必先確認計算基準,700億跟500億差的不是業績,是記帳規則

OpenAI仍對外表示,年底前年化營收有信心衝到700億以上,主要靠企業市場成長。但這次插曲說明,在AI狂熱期,連「營收」這個最基本的數字都可能因為算法差異被誤讀,投資人跟媒體下次引用數字前,最好先問一句:算的方式一樣嗎?

好不好用,試了才知道。

🇺🇸 OpenAI Revenue Review: $70B Revised to $50B, Markets Slide

OpenAI revenue made headlines this week when the Financial Times revealed the company's actual annualized revenue sits around $50 billion — roughly $20 billion below the $70 billion figure that had been circulating among investors. Nvidia, Oracle, and CoreWeave shares all dropped on the news. But the real story is less dramatic than the headline: this isn't about fewer customers. It's about accounting methodology.

Not a Demand Problem — an Accounting Mismatch

The gap in OpenAI revenue reporting comes down to how OpenAI and Anthropic count sales made through cloud partners. When a customer spends $1 on OpenAI tokens via Microsoft Azure, OpenAI only books its roughly 20-cent cut as revenue — Azure keeps the rest.

Anthropic does it differently: it records the full $1 as top-line revenue, then lists the cloud partner's share as a cost of sales. Same transaction, two very different-looking numbers. Investors had been using Anthropic's method to estimate OpenAI's run rate — effectively comparing two companies with two different rulers.

So Why Did Stocks Still Fall?

If it's just an accounting difference, why the selloff? Because the AI bubble narrative has been fragile all year, and Wall Street had already baked the $70B figure into valuation models. Once the baseline number turned out to be wrong — regardless of the reason — the market's first instinct was to sell first and ask questions later. That reflex says more about how nervous investors already are about AI infrastructure spending than about OpenAI's actual business.

What This Means for Users and Developers

ChatGPT subscription pricing isn't affected short-term — this is a reporting-standard issue, not a cash-flow issue
If your company buys OpenAI capacity through a cloud partner like Azure, the revenue attribution is more complicated than it looks — worth asking about before you lock in a contract, more than worrying about the headline number
The bigger lesson: when comparing AI company revenue figures, always check the accounting basis first. The $20B gap here is a bookkeeping rule, not a performance gap

OpenAI still tells investors it expects to hit or exceed $70 billion in annualized revenue by year-end, driven mainly by enterprise growth. But this episode is a reminder that even the most basic number in AI — revenue — can be misread during a hype cycle if nobody checks whether two companies are counting the same way.

好不好用,試了才知道 — the only way to know if it holds up is to actually test 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