← 블로그← All posts

2026-09-22

Semiconductors Fall Harder Than Almost Anything, and Keep Coming Back

By FinScope Research · Data as of 2026-09-18

Semiconductors have a reputation for volatility that people repeat without a number attached. So here is the number, measured across every semiconductor company in the FinScope universe: the median lifetime maximum drawdown is −85.8%.

For context, the median across the 763 non-leveraged securities in the same dataset is −71.2%. The typical chip company has fallen roughly fifteen percentage points further than the typical stock, and several have fallen more than 96%.

Then there is the second number, which is the one that changes how you read the first: zero of the 21 failed to recover. Every one of them eventually reclaimed the high it fell from. Across the full universe, 24% never did.

That combination — brutal falls, complete recovery record — is unusual, and it is the whole argument about how to own this sector.

The 21 names, sorted by how far they fell

Two things stand out. Most of the deepest troughs are dated 2002 or 2008 — the dot-com unwind and the financial crisis, when semiconductor demand collapsed alongside everything that used chips. And the recovery column is measured in thousands of days: INTC needed 5,628 of them, ON 5,585, TXN 5,158. Recovery happened, but on a fifteen-year clock.

The bottom of the table is the interesting part. AVGO’s worst-ever fall is −48.3%, shallower than the S&P 500’s own −55.2%, and it recovered in 110 days. That is not because Broadcom is a defensive business; it is because its listed history starts in 2009, after both of the catastrophes above. Lifetime drawdown is partly a measure of what a company lived through, and comparing a 2009 listing with a 1980 listing on that basis is a category error. This is the same trap the common-window guide exists to avoid.

What the index actually buys you

Put the two semiconductor ETFs next to broad tech and the broad market on identical dates, starting 2004-01-30 (the latest common start across the set) and running to 2026-09-18 — 5,695 shared trading days.

Diversifying across ~30 chip companies cut the worst case from roughly −86% (the single-name median) to about −67%. That is a real and large effect. It did not, however, make the sector behave like the market: SOXX still fell about 12 percentage points deeper than QQQ or SPY in the same crash.

The return column explains why people accept that. Over this window semiconductors returned roughly three times what the S&P 500 did. What is less obvious is the comparison one row up: broad technology (VGT) captured about 79% of SOXX’s total return with a drawdown 12 points shallower. If you hold tech for exposure to the same computing cycle rather than for a specific bet on chip capex, the concentrated version was not obviously the better trade.

The last column is the one nobody quotes. VGT and QQQ are within about 3% of their all-time highs right now. SOXX is −18.6%. The sector’s current recovery is not finished, even after the last two years of headlines.

“Semis are at record highs” is hiding a spread

The index number smooths over a dispersion that matters more than the average.

Right now, within those same 21 companies:

• Near their highs: AMD −3.6%, NVDA −5.5%, TSM −8.7%
• Still deeply below: SWKS −48.5%, ON −47.7%, KLAC −41.3%, AMAT −38.4%, ARM −37.3%, LRCX −33.5%, NXPI −30.9%

Seven of the 21 are more than 30% below their own peaks while the sector is described as being in a boom. The names at the top are the AI-datacenter winners; several of the names at the bottom are equipment, analog and connectivity businesses tied to end markets — handsets, autos, industrial — that have not shared the move.

So “semiconductors are at record highs” and “my semiconductor stock is 40% underwater” can both be true at the same time, and an index chart will not show you which situation you are in. Check the individual name.

Single names versus the index, on the same dates

For the two most-watched single names, aligning with SOXX from 2009-08-06 (the latest shared start) gives:

Over this window the single names delivered 7× to 17× the index’s return. They also carried more drawdown, though less than you might expect: NVDA’s worst fall within this window was −66.3%, not the −89.7% from its lifetime record, because the 2002 collapse sits before the window opens.

That gap between −66.3% and −89.7% is the entire reason we publish both figures on every ticker page. A lifetime maximum drawdown tells you what the security has survived. A common-window drawdown tells you what it did during a period you could actually have held it. They answer different questions and they are routinely quoted interchangeably.

What this means in practice

• Size the position against −85%, not against last year. The median chip company in this data lost that much at some point. The sector’s recovery record is excellent, but the path went through losses most people do not hold through.
• Recovery is measured in years, not quarters. Among these 21, the recoveries that followed the 2002 and 2008 troughs typically ran 8 to 15 years. That is the real cost, and it is the subject of our depth-versus-recovery analysis.
• Decide whether you want the cycle or the bet. VGT got most of the return with meaningfully less drawdown. SOXX and SMH are the concentrated expression.
• Do not read a young ticker’s shallow drawdown as safety. AVGO at −48.3% and ARM at −54.0% look tame next to MU at −98.2%, but neither existed as a listed security during the two events that produced those numbers.

An honest limitation

This universe contains companies that are listed today. Semiconductor firms that went bankrupt or were acquired at a distressed price during the 2001 and 2008 downturns are not in it, and they cannot be, because they no longer have a price series. That biases the “zero failed to recover” result upward — the correct reading is “every chip company that is still listed today eventually recovered”, which is a weaker and more honest claim than “chip companies always recover.”

The same caution applies to the return figures. Survivors are, by definition, the ones that survived.

Check any of these on its FinScope page for the full underwater history, or put two of them on the same dates with the comparison tool. For how leverage changes this picture, see SOXL vs SMH and the leveraged ETF census.

FAQ

How far do semiconductor stocks fall in a downturn? Across the 21 semiconductor companies in this universe, the median lifetime maximum drawdown is −85.8%, compared with −71.2% for the 763 non-leveraged securities overall. Five of them fell more than 96%.

Do semiconductor stocks recover? Every one of these 21 eventually reclaimed its prior high, against 24% of the whole universe that never did. But recoveries from the 2002 and 2008 troughs typically took 8 to 15 years, and this count only includes companies still listed today.

Is SOXX safer than owning a single chip stock? It has been materially less severe: about −67% at worst on a common window versus a −85.8% single-name median. It is not close to market-level risk though, falling roughly 12 percentage points deeper than SPY or QQQ in the same crash.

Methodology: measured on split- and dividend-adjusted daily closes as of 2026-09-18. Lifetime maximum drawdown uses each security’s full available history; common-window figures recompute peak, trough and total return from the latest shared start date across the compared set, so securities with different listing dates are not mixed. “Recovery” is calendar days from trough to the first close reclaiming the prior peak. Semiconductor membership is a hand-maintained list of chip designers, foundries and semiconductor-equipment makers in the FinScope universe; VGT and QQQ appear as broad technology and broad growth comparators, not as semiconductor funds. See the methodology page and the editorial policy.