A price is not a property of a company. It records how many individuals, at the margin, are willing to pay today for earnings they expect later. The evidence on Nvidia in late October 2026 fits that view. The shares fell sharply from a record, but the multiple investors pay for each dollar of earnings has fallen further than the business has weakened. The one clear operating deterioration is China. That is a real problem, but a narrow one. Almost everything below was first reported in mid-to-late October, so it describes one moment and not a settled trend.
The drawdown: sharp, sector-wide, and only tentatively arrested
Nvidia reached a record close on October 6, when one post cited a price of $239.82 60. It then fell for most of three weeks. In the week ending October 16 it lost 6.61% 14,15, while its closest peers averaged a 6.1% decline over the same five sessions 22. On October 16 it stood 10.5% below the October 6 record close 93, and a later reading put it 11.8% below 100. It fell a further 4.4% on October 22 29,41,42,52,75, recorded a third straight down week 37,103, and was down 7.6% for October 37,103. The decline came during a wider technology sell-off 44. One post described it as an "AI-jitters sell-off" costing roughly $200 billion in market value 23.
The latest data show a rebound, not a reversal. On October 27 Nvidia gained $1.94, or 0.90%, at the close 32,82. That recovered only part of the damage from three losing weeks 59. Volume was 84.8 million shares against an average of 116.7 million 84,86,91, so few people took part in the bounce.
The stock is still up for the year, though the sources disagree on how much. Estimates range from about 13% 3,76 to 18.43%, the figure with the most sources behind it (four) 31,43,80,97. By October 23 the gain was only slightly ahead of the S&P 500's 76. Nvidia was also not the weakest AI name in the sell-off: ORCL and CRWV lost more than its roughly 3% decline 94.
The multiple has fallen faster than the price
The best-corroborated finding is the collapse in the trailing multiple. Nvidia's P/E is cited as the lowest since June 2019 96. The valuation history has a median trailing P/E of 65.61x, a figure carried by four sources 98. The latest reading is 26.51x, and none of the 26 history rows shows a lower P/E 98. Other sources agree on the level but not the decimal: 26.69x on one page 27 and 27.1x in an article excerpt 20. The same filing-based series shows the P/E at 202.84x in May 2023 98. One commenter says the trailing P/E fell from 39x at January's close while profit grew 61. Another item says the forward multiple has declined since August 2024 despite strong quarterly results 47.
Forward-looking claims point the same way but rest on thinner evidence. One post cites a 17x forward P/E for NVDA against 19x for the S&P 71. Another source's bullish-case earnings imply 19x 69. These are single-source claims and deserve less weight than the trailing history. The direction is better supported than any specific multiple, because the figures come from different dates and methods.
The arithmetic matters because it shows how a stock can fall while earnings rise. One article's illustration: at $8 of EPS, a 30 P/E gives $240. If EPS rises 25% to $10 while the multiple falls to 20, the value drops about 16.7%, to $200 87. The sources do not say why Nvidia's multiple fell, so that is my reading. But the pattern fits: the market is paying less for each unit of earnings, and does not appear to dispute that the earnings are growing.
Nvidia is not alone: the compression is sector-wide
If the cause were something specific to Nvidia, we would expect its peers to look different. They do not. As of October 19, Alphabet, Amazon and Meta traded at lower trailing multiples than Nvidia 102. A commenter says Nvidia's 28.8x trailing multiple was its lowest in five years 61. Technology P/E is said to have derated about 30% over the past year 79. The Mag 7 median forward P/E is 23.9 79, and one source says it has fallen from about 37x to 24x 79. The aggregate US AI basket moved from a forward P/E of 20.4, 6% above its average, on 29 July 2,79, to 17.9x, 7% below the same average, by October 79.
The individual cases show the same thing from different angles:
Broadcom reported record quarterly revenue of $29.6 billion 4,5,6,7,8,74, yet its shares were 20.9% below the April 30 close 34,36,49. Its forward multiple, cited at about 39x in older material 1,84, is now put at roughly 13x by Bank of America's Vivek Arya 10 and 19x elsewhere 73.
TSM fell 9.0% over five trading days to October 16 while its closest peers averaged a 5.7% decline 13. Over a longer window it is still up 8.5% over three months 12,54, and revenue growth of 36.4% is widely repeated 36,38,39,40,50,51,53,54,55.
AMD fell 13.05% in a week 14,15 from a trailing P/E that most sources put above 100 (130.6 is cited by eight sources) 11,35,36,38,40,50,55. It shows that compression has not been uniform: AMD remains priced far above the peer median of 23.6 51.
Compression is concentrated in the large platforms. Some AI sub-baskets still trade at premiums, such as Infrastructure at 20.3 against 16.9 79 and Cooling at 23.6 against 17.0 79. Against this, the broad US market is described as stretched on other measures. The S&P 500 price-to-sales ratio is described as at an all-time high 78. So Nvidia has not been marked down because everything is cheap. It has been marked down while the wider market is still expensive on those measures.
Where the fundamentals do bite: China
The decline in the multiple is not the whole story, and China is the exception to a benign operating picture. IDC data, as relayed in several posts, show Nvidia's China accelerator shipments down 44% year over year in the first half of 2026 24,25,67,72. One account drops the China qualifier and applies the 44% to Nvidia's AI accelerators generally 62. The scope difference is unresolved, and the China framing is more common. IDC is also reported to estimate that Nvidia's share of China's AI-chip market fell to 34%, a claim with five sources 25,28,46,77,83. A harsher claim, that Nvidia has effectively lost its China AI accelerator market, has a single source and goes beyond the 34% figure 26.
Other evidence is suggestive but thin. Reuters reported that Nvidia paused some revenue-sharing deals amid concerns that included the extent of control sought 45. One item notes days sales outstanding rose from 45 to 60 in a year 16. That is a single-source observation, and the evidence does not say what drove it. Even so, China is the best-corroborated adverse operating data in the set, and it is a loss of volume and share, not a change in multiple. The stock's reaction to it cannot be separated from the broader sell-off.
Capital return and insider sales: two signals, opposite readings
Management has acted as though the shares are cheap. The repurchase authorization was expanded substantially 81,89, with $235 billion remaining at the September 28 announcement 87. The dividend was raised 89,92. Neither is large against the size of the company. The net share count fell by 157 million shares, about 0.65% of the beginning balance, over the six months to July 26 87. The dividend yield is cited at 0.24% 30,95, though other yield figures differ 27,48. The market reaction to the buyback news was mixed: one post says the stock ticked higher 60 and another says it closed red 60.
Insiders have acted differently. Insiders sold 1,766,118 shares in the last quarter, a claim with 15 sources 9,46,56,58,63,65,66,77,83,85,88,90,99,101. Sentiment is described as negative across 92 insiders 57, and selling rose relative to earlier in the year 57. Reported sales include CFO Colette Kress at 4.36% of her position 46,58,66 and Director Mark Stevens at 21.03% 58,77,85,88,90. We should not over-read this. The evidence gives no reason for any of the sales, and a person sells for many private purposes, so these sales are not a verdict on the business. Institutional activity is similarly mixed, with Van Eck adding 34.7% in Q2 85 and Silverleafe cutting 55.6% 63. Analysts remain constructive, with targets of $255–$287 implying 18–33% upside 18, though one analyst rates the stock Strong Sell 33,64.
What it implies, and what it does not
The weight of the evidence supports reading this as a de-rating of a still-growing company during an AI-wide risk-off phase. The supports for that reading are a trailing multiple at the bottom of its recorded range, a pullback matched by peers, and a rebound too light to call a turn. China is the exception: it is the one place where the evidence points to real erosion, and it is the main reason not to call the whole decline sentiment.
There are limits on what a low multiple allows us to conclude. The sources themselves warn that forward P/E estimates depend on the earnings definition used and lean on aggressive future earnings expectations 70. They also say a low forward P/E can signal undervaluation or an expectation that unusually strong earnings will normalise 68. A cheaper multiple is therefore not the same as a bargain. It is the market's current price for an uncertain future, and we cannot say from this material whether buyers or sellers will turn out to have judged better. Nor does the buyback by itself look sufficient to halt a decline, given the mixed reaction to it 60.
History gives some perspective, though only as a base rate. Nvidia has had three drops of 20% or more in the last three years, all fully recovered 19,21, and its deepest drop is cited at −36.9% 11. An article notes the stock moved lower initially after four of the last five earnings reports, then had risen by the time of the next report about three months later 17. These come from single articles and do not predict the next move. The sound conclusion is a pattern prediction, not a price one: where earnings keep growing while the multiple stays compressed, the gap between the two is the thing to watch. That gap will close by the multiple recovering, by earnings stalling, or by China's erosion spreading, and the material does not tell us which.