War And Rotation

Both Iran and semiconductor stocks got pumled this week.

For three weeks, markets had been pricing the U.S.–Iran ceasefire as a peace. This week made the distinction painfully clear.

The United States resumed large-scale strikes on Iran, Iran threatened additional regional energy exports, and oil rose for a fourth consecutive session. That geopolitical shock hit a market already rotating hard away from its former AI leaders.

Another DeepSeek Moment

The latest AI scare came from Zhipu AI, known internationally as Z.ai, and its new open-source GLM-5.2 model. It offers a one-million-token context window and an architecture designed to reduce the computing required per token.

That immediately revived the DeepSeek argument: if better software lets us run AI with a fraction of the hardware, perhaps the current hardware buildout is being overbuilt.

If that turns out to be true—and if Jevons paradox does not apply—it would be bearish for the AI hardware stack. We should say that plainly.

GLM-5.2 does not demonstrate that, though. A 753-billion-parameter model with a one-million-token context window is hardly evidence that computing power no longer matters. It is evidence that powerful AI is getting cheaper to use.

Lower costs can mean more tokens, more agents, more workloads, and more applications that were previously uneconomic. DeepSeek’s original breakthrough did not stop the current capital-spending cycle, and there is little evidence yet that this one will either.

Leverage Travels

Part of the violence has come from Korea, where repeated semiconductor-led selloffs have contributed to seven KOSPI circuit breakers already this year, most recently on Monday, after an extraordinary run.

Highly levered markets do not unwind neatly. Forced selling travels across borders and across tickers. Margin calls do not stop to ask whether next quarter’s high-bandwidth-memory demand remains strong.

That helps explain why so many adjacent AI names have been hit at once. Price declines can create their own selling pressure even when the underlying demand picture has not changed.

The Tape Is Damaged. Demand Isn’t.

On Wednesday, we laid out the fundamental case in “The AI Buildout Is Accelerating.” The evidence has only gotten stronger.

Micron Technology (MU) reported another enormous sequential increase in revenue and guided higher again. Penguin Solutions (PENG) reported record revenue and raised its guidance.

ASML (ASML) raised its full-year forecast and expanded its capacity plans. Taiwan Semiconductor Manufacturing (TSM) reported another record quarter and committed another $100 billion to U.S. chip capacity.

None of that guarantees these stocks will start going up again soon. Crowded trades can correct brutally while their underlying businesses keep improving.

Those reports are evidence of demand, not its collapse. The tape is damaged. The AI buildout is not.

The Cyclical P/E Trap

The bears are right about one thing: a low trailing P/E does not automatically make a memory stock cheap.

In a cyclical business, the multiple often looks lowest near peak earnings because the denominator is temporarily inflated. Peter Lynch famously suggested that cyclical stocks could be most dangerous when their P/Es looked lowest.

That observation only helps if earnings actually are near their peak.

The real question is where we are in this cycle, and whether AI demand, HBM content, capacity constraints, and longer customer commitments have extended it. Recent results and guidance make a peak-earnings assumption look premature. This may be an unusually long cycle, rather than the end of cyclicality.

Waiting for the P/E to rise is not much of a strategy, either. It can rise because the stock appreciates, but it can also rise because earnings estimates collapse.

If the current cycle has further to run, waiting for a more conventionally reassuring multiple could mean missing most of the remaining move. Meanwhile, adjacent AI names with higher multiples—or no current earnings at all—have been hit much harder. That looks more like a valuation and positioning unwind than evidence that the buildout has stopped.

Better Models, More Demand

Here is one anecdotal data point from running the Portfolio Armor Substack.

The biggest pain point has been tying brokerage fills back to their original trade alerts and documenting every OpEx-week exit in our posts and detailed spreadsheet. That process used to take me more than a dozen hours.

The latest models from Anthropic and OpenAI can now largely automate it.

Within a few hours of trying Anthropic’s latest model, I upgraded from its $20 monthly plan to its $100 plan. That is the other side of the efficiency argument: as AI gets smarter, it becomes much more useful, and users become willing to pay more for it.

Efficiency does not merely save tokens. It creates customers and use cases.

Respecting The Damaged Tape

Finance X is full of victory laps from people who were not along for the climb. The vehemence of some of those victory laps says something about positioning and prior nonparticipation, but nothing about end demand.

Bullish commentary can be colored by ownership too. Price action deserves respect, but it is not, by itself, proof that the underlying thesis has failed.

So we are not fighting the tape. We structured our recent chip and memory options trades to wait out a correction for several months, if necessary. We can wait for their technical setups to repair before adding more exposure.

The AI buildout can remain intact while AI stocks correct. Both can be true.

How We’re Trading It

Instead of adding more new exposure to the most damaged AI-hardware names today, we’re turning to three different themes.

Our first trade is on a growing browser-software company that is incorporating AI features into its products. The second is on a payments and embedded-finance company with additional strategic optionality.

The third is on a late-stage psychedelic-medicine developer. That theme received a major vote of confidence Thursday when Eli Lilly agreed to acquire AtaiBeckley for $2.8 billion upfront, plus a contingent-value right potentially worth another $1 billion.

We already have an uncapped ATAI call from an earlier trade alert. Today’s alert looks to add exposure to another company advancing a psychedelic treatment through late-stage development.

In all three trades, we’re using options volatility to lower our entry costs while keeping the downside measured.

We’re enabling free subscribers to unlock today’s full trade alert.

🚨 Browsers, Payments, And Psychedelics 🚨

Three asymmetric trades for a market that is rewarding selectivity.https://t.co/zP8U9hPtuQ

— Portfolio Armor (@PortfolioArmor) July 17, 2026

If you’d like to see the exact names, option structures, and exit plans, you can become a free Portfolio Armor subscriber by subscribing below.

You can become a free subscriber via the widget below, and you'll also get an email with today's trades before the market opens. 

And if you're worried about market risk here (understandably so), you can hedged with the Portfolio Armor website or iPhone app.