同樣貼著AI標籤，毛利率卻在往三個方向走

發佈日期: 作者:
同樣貼著AI標籤，毛利率卻在往三個方向走

MSFT 的毛利率連四季收窄：69.0%、68.0%、67.6%、67.2%。四個數字排成一列，像一條緩坡往下走的折線圖。

同一時期，META 的毛利率在 81.4% 到 82.0% 之間微幅震盪，幾乎沒有方向，像一條水平線貼在高處發呆。NVIDIA 則反過來，從 72.4% 一路爬到 75.0%，最近一季微收在 74.9%，整體仍是上升趨勢。

三家公司同樣被分析師貼上「AI 概念股」，但毛利率走勢根本沒有在同一個方向。這不是市場情緒的問題，是商業模式本身的問題。

賣工具的 vs. 蓋工廠的

NVIDIA 的邏輯最清楚：晶片供不應求，定價能力極強。買家排隊，賣方訂價，成本沒什麼理由大漲，毛利就跟著往上走。這個結構在短期內很難動搖，除非供給突然暴增或需求急凍，兩件事目前都沒發生。

META 的結構則是另一種乾淨法。廣告業務本來就是高毛利模型，AI 進來之後做的事是讓廣告投放更準、讓用戶停留更久，而不是新增什麼重資產。AI 把廣告效益推高，但 META 自己不用多買一台伺服器就能多收一塊廣告費——邊際成本趨近於零的生意，本來就不怕規模變大。

MSFT 的問題在於選擇做的事本質上很重。Azure 要擴規模，資料中心要蓋，伺服器要買，折舊跟著一起長。Copilot 和 Azure OpenAI 服務帶來的營收在成長，但成本的增速不遑多讓。毛利率收窄不是壞消息，是「燒錢換規模」階段的正常現象——但投資人要看清楚，這跟 META 那種「幾乎不用加碼就能收更多錢」的模型，完全不是同一種 AI 故事。

標籤相同，底下的錢走法不同

廣告公司做久了有個直覺：同一個廣告詞貼在不同產品上，不代表這兩個產品賣的是同一件事。「AI 概念股」這個標籤現在就是這個狀態。

NVIDIA 賣的是算力，AI 熱潮讓定價能力極強，毛利跟著走高。META 賣的是廣告注意力，AI 讓廣告更準，但商業模式的本質沒變，毛利維持高位。MSFT 賣的是雲端基礎設施，AI 讓資料中心要蓋更多、設備要買更多，毛利因此收窄。

三種故事，三種財報結構。季報的數字不說謊，只是懶得翻譯。

毛利率的方向比毛利率的高低更值得看

投資圈有時候太專注在「誰的毛利率比較高」，反而忽略「方向」這件事。MSFT 的 67.2% 聽起來仍然不錯，但方向是向下的，而且伴隨資本支出持續擴大。NVIDIA 的 74.9% 背後是定價能力還在，需求還沒退。META 的 82% 是廣告飛輪自己在轉，不太需要 AI 投資來撐場面。

如果把這三條線畫在同一張圖上，再疊上各家對 AI 的資本支出規模，答案其實很清楚：資本支出越重的公司，短期毛利率的壓力越大，這不是意外，是必然。

問題從來不是「這家公司有沒有 AI」，而是「AI 在這家公司的商業模式裡，是幫它少花錢、還是讓它多花錢」。財報四個季度的數字已經給了答案，剩下的就是願不願意去讀。

— 胡凱翊（Kevin）

本文為個人研究筆記，不構成任何投資建議。

Same AI Label, Three Very Different Margin Stories

Four quarters. Four numbers. Microsoft’s gross margin went 69.0%, 68.0%, 67.6%, 67.2% — a slow, deliberate slide downward. Not a crash. A slope.

Over the same period, Meta held between 81.4% and 82.0%, barely flinching. NVIDIA climbed from 72.4% to 75.0%, dipping only slightly to 74.9% in the most recent quarter. Three companies wearing the same “AI stock” label. Three margin lines pointing in completely different directions.

This isn’t a sentiment story. It’s a business model story.

Who Sells the Picks, Who Digs the Mine

NVIDIA’s math is the simplest. Demand exceeds supply; pricing power stays intact. Buyers queue up, sellers set the price, costs don’t have much reason to spike. Gross margin drifts upward almost on its own. That structure doesn’t break easily unless supply floods the market or demand collapses — neither has happened.

Meta’s structure is a different kind of clean. Advertising was already a high-margin game. AI arrives and makes ads sharper, keeps users on platform longer, but doesn’t require Meta to buy a warehouse full of new hardware to collect the next dollar of revenue. The flywheel spins faster; the incremental cost stays near zero. Margin holds at the high end because the model was already built that way.

Microsoft chose the heavy path. Scaling Azure means building data centers and buying servers. Depreciation grows alongside the footprint. Copilot and Azure AI services are generating revenue, but costs are climbing right alongside them. A narrowing margin here isn’t a crisis signal — it’s what “spend to scale” looks like in the income statement. The question is whether investors are reading it that way, or just seeing the AI headline and stopping there.

Same Tag, Different Cash Flow Logic

Spend enough time in advertising and a pattern emerges: slapping the same tagline on two products doesn’t mean they’re selling the same thing. “AI stock” is in that exact spot right now.

NVIDIA sells compute. The AI boom amplifies pricing power, and gross margin follows. Meta sells attention. AI makes that attention more precise, but the underlying model doesn’t change — margin stays elevated. Microsoft sells cloud infrastructure. AI means building more of everything — and margin compresses as a result.

Three narratives. Three income statement shapes. The quarterly numbers don’t editorialize. They just sit there, waiting.

Direction Matters More Than Level

The habit of comparing gross margin levels — “67% vs. 82%, so Meta wins” — misses something. Direction, paired with capex trajectory, tells the more honest story.

Microsoft at 67.2% with a downward trend and expanding capital expenditure is a different animal from Microsoft at 67.2% with a stable margin. NVIDIA at 74.9% with pricing power intact reads differently from a company at 74.9% running out of room. Meta at 82% generating incremental ad revenue with near-zero marginal cost is, structurally, a different business from the other two.

Put all three lines on one chart and overlay each company’s capital spending commitments. The pattern clarifies fast: heavier infrastructure bets compress near-term margins. That’s not a flaw in the strategy — it might be exactly the right call — but it’s not the same trade as buying a company whose AI investment mostly makes existing revenue run hotter.

The real question was never “does this company have AI.” It’s whether AI, inside this specific model, is a cost reducer or a cost multiplier. Four quarters of data have already answered that. The numbers are readable. The label is just a label.

— 胡凱翊(Kevin)

本文為個人研究筆記,不構成任何投資建議。

延伸閱讀

短網址分享
https://justfly.idv.tw/s/xpWy5CQ