Oversold Stocks of August 2026: Semiconductors Down 23% While the S&P Sits at a Record, and Four Questions for Every Drawdown

Two screens, one morning

The morning of 20 August 2026, Seoul. Samsung Electronics announces a shareholder-return programme worth more than 100 trillion won. An hour later SK Hynix adds a 40-trillion buyback. The KOSPI index closes the day up 5.9%.

That same morning, on the other side of the Pacific, the American semiconductor index sits 23% below its peak — the sector’s deepest fall since 2022. The S&P 500, meanwhile, is within one percent of its all-time high, at roughly 7,800.

There is no contradiction here. The index is weighted by company size, so a few of the largest names hold it at the top while the rest of the market gets cheaper. Between those two screens sits a layer of companies that have fallen further than their businesses have deteriorated. Beside them, at identical discounts, sit companies that have fallen exactly as much as they deserved.

Automated stock screens, which sort companies by a column headed «percent below the 52-week high», cannot tell the two apart. To such a program, −53% always looks more attractive than −27%. It does not read lawsuits, and it does not distinguish cash from debt. This article is about what that column does not show.

The depth of a fall is not yet an argument.

What is actually happening

First, the state of play.

From above the picture looks good: the S&P 500 about 1% off its record, the Russell 2000 small-cap index at an all-time high, the VIX volatility index at 14–16 — meaning almost no fear is priced in. The earnings season is formally the best in 15 years: 86% of S&P 500 companies beat analysts’ forecasts.

Now what sits underneath. The 30-year Treasury yield is at a 20-year high. Inflation is stuck above target: consumer prices at 3.4%, core personal consumption expenditures at 3.3%. Futures markets put a roughly 35–38% probability that the Federal Reserve raises rates in September; a cut is not on the table at all. The Fed has a new chairman, Kevin Warsh, and at the July meeting three regional bank presidents dissented in favour of a hike — the first triple dissent since 2016. The average tariff rate on imports is the highest since 1969.

The record earnings season carries a footnote too: the aggregate 29% beat was largely produced by paper revaluations of assets at Alphabet and Amazon. Excluding those two, the figure is 10.9%.

Market breadth is thinning. In mid-August, decliners on the New York Stock Exchange outnumbered advancers by 1.55 to one; the share of stocks above their 50-day average fell from 60% to around 55%. A few of the largest companies hold the index up while the rest has been sliding for weeks.

The calendar ahead is dense:

Date	Event	Why it matters
26.08	NVIDIA results (guidance ~$91bn) and July inflation data	The market’s two key numbers in one evening
27–29.08	Jackson Hole symposium	Warsh’s first appearance as chairman
01–02.09	Credo and Broadcom results	A test of hardware demand beyond NVIDIA
11.09	August inflation data	The last reading before the Fed meeting
16.09	Fed meeting	Hold ~62–65% against a hike at ~35–38%
22.09	Micron results	The decisive point for the whole memory story

Buying everything that has fallen in these conditions is a bad idea. Buying nothing means missing the broadest clearance sale of decent companies in two years. What is needed is a way to tell them apart.

Who fell, and why

Semiconductors, down 23% on the index. The cause was not poor results but the reaction to spending plans. Alphabet beat expectations and fell 7% because it raised its capital-expenditure forecast. Meta and Microsoft went through the same: results above expectations, spending plans above comfort, share price down. Equipment makers fell for weeks even as the industry forecast for equipment purchases was raised — to $145–150bn for 2026 and $185bn for 2027. The market is not saying demand for computing has vanished. It is saying the price of that demand frightens it. These are different statements.

Memory is a case of its own. On 6 August, Western Digital fell 16% in a day. On 13–14 August, SanDisk added 15%. By 18 August the sector was red again. All this against contract DRAM prices up 150–200% year on year and high-bandwidth server memory sold out into 2027. Phones will cost more next year not because of a production shortfall, but because data centres now take about 70% of output. A headline saying «the smartphone market will shrink 13%» reads to an automated screen as demand disappearing. In fact demand has moved to another segment.

Subscription software companies. Salesforce endured the longest losing streak in its history — fourteen consecutive sessions in June. ServiceNow traded 46% below its level of a year earlier at one point. The cause was the fear that AI agents will replace subscriptions sold per seat. In August the assessment began to shift: JPMorgan restored its «overweight» rating on Salesforce, saying the fears were overdone; BofA and Wells Fargo raised their price targets on ServiceNow.

Consumer goods. Nike fell to a twelve-year low — not on its own results but on a competitor’s: On Holding missed its revenue forecast and the market marked down the entire shelf. This is the cleanest case of overreaction: a company punished for someone else’s numbers.

The depth of a fall is not yet an argument.

Four questions for every drawdown

Now the method itself. Four questions an automated screen cannot answer. None requires an expensive terminal — the company’s latest report and ten minutes of news will do.

1. Which way does the company’s own guidance point? What matters is not the depth of the fall but the direction of management’s own forecast. A company down 28% that has raised its full-year guidance and a company down 28% that has cut guidance for the second time this year are two different cases wearing the same price tag.

2. Who pays for the drawdown? Deckers holds $1.7bn of net cash: its fall costs shareholders their nerves. ZoomInfo carries debt of 3.8 times cash flow: its fall costs the company its future.

3. What is hanging in court or with a regulator? Lawsuits, investigations, statements by former employees. None of it appears in the «percent below the high» column.

4. Who is selling at the lows, and who is buying? Insider selling at the bottom is a bad sign. Buybacks and director purchases mean the company is voting with its own money against the current price.

Two August examples show why the check matters. Both looked like bargains on any screen.

First: Alignment Healthcare, down 48% from its high. The picture was ideal: results above forecast on every line, full-year guidance raised — and the shares kept falling. The reason lay outside the report. Since 8 July a lawsuit by a former senior executive has been public, alleging that the company’s first positive operating profit was produced by reclassifying $8–10m of operating costs as capital expenditure. The market was not pricing the report; it was pricing the probability that the report cannot be trusted.

Second: Zoetis, down 53% from its high. Veterinary pharmaceuticals, a strong business, at half its peak price. But the August report cut guidance to a decline in revenue — down one to three percent for the year — with three product lines under pressure at once: generics taking share from older drugs, competitors pressing in parasiticides, and sales of the flagship Librela slowing after regulatory warnings about side-effects. This is not a temporary discount. It is a repricing of a business that has genuinely weakened. Cheapness is a condition of the price. Breakage is a condition of the business. The most expensive mistake of August 2026 is confusing the two.

An automated screen would have shown both companies side by side at almost identical discounts. The check rejects both, but on different grounds: one on the third question, the other on the first.

The depth of a fall is not yet an argument.

Ten candidates, with numbers

Below are ten companies that passed the check. On method: the list was assembled on 20 August in several stages. Candidates were sought along six independent lines (post-results collapses, deep drawdowns at large companies, small and mid-caps, foreign markets, non-technology sectors, special situations); each finalist was then examined specifically for ways to refute the case, with disputed calls reviewed separately. The two that looked like the best candidates failed the check — they are the Alignment Healthcare and Zoetis cases above. Prices are closes or intraday levels on 20 August; they will be stale sooner than this text is indexed.

Ticker	Company	Price	From 52-week high	Why the discount looks excessive	Main risk
MU	Micron	~$956	−24%	DRAM prices up 150–200% year on year, server memory sold out into 2027	Results on 22.09 decide everything
AVGO	Broadcom	~$366	−26%	Fell on someone else’s deal (Marvell–Google), not its own numbers	Results on 02.09
NOW	ServiceNow	~$129	−34%	Price targets raised inside the drawdown; the substitution fear is not in the numbers	The fear may return
INTU	Intuit	~$361	−50%	Revenue up 15%, free cash flow up 26%, valuation below its own history	Results on 25 August
ALNY	Alnylam	~$234	−53%	Down 28% in a day for «post-launch normalisation» while revenue grew 67%	Already bouncing; not cheap
IDXX	IDEXX	~$560	−27%	Price fell while the company raised guidance; return on capital above 40%	A future competitor in Mars
DECK	Deckers	~$89	−29%	$12 of net cash per share, zero debt, guidance raised — and the stock at its low	The Hoka brand slowing
CSGP	CoStar	~$32	−65%	The margin turn is already in the numbers; the chief executive is buying shares	Analysts are split
MTZ	MasTec	~$272	−38%	Down 19% on a one-cent miss with guidance raised and a record order book	An expensive valuation
EWY	iShares Korea	~$179	−19%	Samsung and SK Hynix buybacks plus the state value-up programme	Swings of ±20% within weeks

Below are six months of each. The dashed line marks the 52-week high: the gap between price and dashes is the discount an automated screen displays and the check calls into question.

What the charts show. Micron is not a «cheap stock»: it rose from $320 to $1,213 and gave back a quarter; buying the drawdown means buying the tail of the sector’s biggest run. Intuit and CoStar are the opposite — long slides with no preceding euphoria. Deckers, on the day of publication, sits exactly at its low, with no attempt at a bounce yet. And the Korea fund cut its drawdown from −23% to −19% in a single day of local gains: the speed at which that window closes is an argument in itself.

Who pays, who collects

The distribution of consequences, without judgment.

Paying is the investor who holds the broad market through equal-weighted funds and watches the index at a record while their portfolio is not. Paying is next year’s phone buyer: memory has moved to data centres, and the shortage will show up in the bills of people who have never used a neural network. Paying is the late buyer of AI infrastructure, arriving after the largest cloud companies have already promised $725bn of capital spending a year — 77% more than last year.

Collecting is the Korean minority shareholder, to whom Samsung and SK Hynix are returning money in trillions of won for the first time in a generation. Collecting is the company buying back its own shares: Netflix repurchased a record $4.7bn in a quarter, precisely during the drawdown. And collecting, potentially, is whoever is willing to buy when the screens are all red and the four-question check finds nothing broken.

We wrote about the mechanism itself in «The market as a sensor»: prices aggregate scattered information faster than any committee. The August version of the same rule: the market has already repriced whole sectors — quickly, sharply and, in places, unfairly. That unfairness is what the person doing the checking gets paid for.

The case against the whole idea

Now the objections. Without them this would be an advertisement.

First: the rate regime. Buying drawdowns makes sense in a rate-cutting cycle. Right now the 30-year yield is at a 20-year high and a third of the market expects a hike. «Buy what has fallen» works well in one regime and stops working in another.

Second: valuation. The Shiller CAPE ratio stands at about 41 — the second-highest reading on record after the dotcom bubble. Economists at the European Central Bank published a warning on 17 August about a probable correction in American technology. Bill Dudley, the former head of the New York Fed, writes plainly in a Bloomberg column of 20 August that the AI boom does not remove the risk of a sharp correction, though bubbles often grow larger and last longer than expected.

Third: dependence on a single factor. Half of any list of fallen stocks in 2026 is, directly or indirectly, a bet on the AI investment cycle continuing. If that cycle stalls, a «diversified» basket of candidates falls together. The four-question check separates a damaged business from a temporarily cheap one; it does not cancel the days when everything is sold.

The answer to these objections is not optimism but procedure: enter in stages rather than in one lump; keep part of the money uncommitted through the end of August, because events run weekly from 26 August to 16 September; size positions to survive being wrong; and make no exceptions to your own rules, however attractive the discount. Cheapness is a necessary condition for buying. A sufficient one is not on sale in this market.

The reckoning

The market of August 2026 is right and wrong at the same time. Right that expensive promises in capital-spending plans deserve a discount. Wrong in the details: it is selling sound companies alongside troubled ones, a veterinary near-monopoly alongside pharmaceuticals that have genuinely lost ground, and profitable consumer brands alongside a general panic in the sector.

Sorting through the wreckage is a matter of procedure rather than courage: four questions, ten minutes per company. In those ten minutes an automated screen will show you a hundred discounts. It will always be faster. The bill for that speed simply does not arrive at its address.