Nvidia 如何用 AI 智能體檢查自身估值是否偏高：2020 年以來的 P/E

All outputs are real tool calls to the Equibles MCP server from 21 October 2026, trimmed.

"Nvidia looks expensive" usually means the price went up a lot. That's a different claim from the valuation going up, and over the last three years the two went in opposite directions.

Here's how I get an AI agent to check a stock against its own valuation history, without the look-ahead mistake that makes most homemade charts wrong.

Connect it first

In Claude, open Customize → Connectors → Discover, search for Equibles, click Connect to Claude and Authorize, then switch it on in a chat from + → Connectors. In ChatGPT: Plugins, search Equibles, Install plugin. In Claude Code, claude mcp add --transport http equibles https://mcp.equibles.com/mcp and then /mcp to sign in. The free plan covers 100 requests a day; other clients are in the setup docs.

1. Pull the history, computed as of each filing

The prompt: "Using Equibles, show Nvidia's valuation multiples over time." The agent calls GetValuationMultiplesHistory, which returns P/E, EV/Revenue, EV/EBIT and EV/EBITDA once a quarter.

The detail that matters: each row is computed on the day that quarter's report was filed, from figures available by then and that day's closing price. A chart that divides today's restated earnings into an old price uses numbers nobody had at the time. This one doesn't.

A few of the 27 rows:

Prices are on today's split basis, so the 2020 close of $7.72 is comparable with the 2026 one.

2. Put the latest value in its range

"For each multiple, give the median, minimum and maximum across all rows, and where the latest value ranks."

              Median     Min       Max      Latest   Rows lower than latest
P/E (TTM)     65.61×    26.51×   202.84×   26.51×         0 of 26
EV/EBIT       59.43×    25.46×   212.99×   25.46×         0 of 26
EV/Revenue    24.28×    12.43×    37.00×   16.60×         4 of 26

On earnings, the August 2026 filing is the cheapest Nvidia has been in any of the 27 samples. In 24 of them the P/E was above 40.

3. Split the price move into its two parts

From the May 2023 filing to the August 2026 one, the price went from $38.95 to $209.66, about 5.4 times. The P/E fell from 202.84× to 26.51×, to 13% of where it was. Dividing price by P/E gives the earnings behind it: trailing EPS went from about $0.19 to $7.91, roughly 41 times.

So the stock rose a lot, and earnings rose about eight times faster. That's the sense in which a stock can go up fivefold and get cheaper.

4. Read the revenue multiple against the earnings multiple

EV/Revenue didn't fall as far. Four earlier rows are lower than today's 16.60×, the 2022 trough among them. The reason is margin: dividing EV/Revenue by EV/EBIT gives the operating margin the market is pricing.

                 EV/Revenue   EV/EBIT   Implied operating margin
Feb 2020           16.49×      63.25×          26%
Aug 2026           16.60×      25.46×          65%

Same price per dollar of sales, very different price per dollar of profit. If margins came back down, the revenue multiple would be the better guide.

5. Compare with peers last

GetValuationMultiples gives the current snapshot with a peer median:

The tool says the peer set is three similar-size semiconductor companies, so I treat it as a sanity check, not a verdict. The snapshot P/E is a little higher than the last history row because it uses the current share price.

The prompt I reuse

Using Equibles, check <TICKER> against its own valuation history:
1. Quarterly P/E, EV/Revenue, EV/EBIT, computed as of each filing date.
2. Median, min, max per multiple, and where the latest value ranks.
3. Split the price change since the peak multiple into earnings growth and
   re-rating.
4. Implied operating margin then and now (EV/Revenue ÷ EV/EBIT).
5. Current snapshot with peer median, and the peer sample size.

What valuation history can't tell you

• Whether earnings hold. A low trailing P/E after a fast run-up in profits says nothing about next year's profits.
• The blanks. Some rows have a dash, EV/EBITDA in fourth quarters for example, and 13 older quarters were left out entirely. The tool skips a ratio rather than estimating a missing input.
• Forward multiples. This is trailing data only.

The same multiples are on each company's financials page, like Nvidia's.

Nothing here is investment advice.