The AI Trade Isn’t Dead: Semiconductors, Memory and Photonics Roar Back

The AI trade came roaring back this week.

After several punishing sessions across semiconductors, memory and optical-networking stocks, buyers rushed back into the market. Micron, Sandisk, Marvell, Coherent and Lumentum all rebounded sharply, while the Philadelphia Semiconductor Index delivered its strongest session in more than a month.

The move was powerful enough to make the previous week’s panic feel distant. It should not.

This remains a volatile momentum market operating inside the historically difficult third quarter of a midterm-election year. The rebound may be tradable, and the fundamental AI infrastructure story remains intact, but the easy part of this cycle is probably behind us.

The opportunity now is not simply to buy every stock that bounced. It is to determine whether the selloff represented the beginning of a broken AI trade—or a violent reset that cleared leverage out of some of the market’s strongest long-term themes.

What triggered the selloff?

The Philadelphia Semiconductor Index lost approximately 10% last week as concerns spread through the AI complex.

The initial pressure came from several directions:

• Questions about whether hyperscalers can sustain current AI capital expenditures
• Fear that more efficient models could reduce future compute and memory requirements
• Concern that new semiconductor capacity will eventually create oversupply
• Extremely elevated valuations after enormous year-to-date gains
• Leveraged speculation and forced selling in South Korea
• Profit-taking following SK Hynix’s U.S. market debut

South Korea became the epicenter because Samsung Electronics and SK Hynix dominate its stock market and sit at the center of the global memory industry.

SK Hynix’s Korean shares dropped more than 15% in a single session following the company’s Nasdaq debut. The Kospi fell 9%, triggering a temporary trading halt, while the damage spread into Micron, Sandisk, Western Digital and semiconductor stocks throughout Asia and the United States. Reuters reported that leveraged products, margin debt and profit-taking substantially amplified the move.

That distinction matters.

A stock declining because its earnings expectations are collapsing is very different from a stock declining because overleveraged investors are being forced to sell it at any price.

The Korean decline contained elements of both fundamental concern and market-structure stress, but the speed and violence of the move strongly suggest that forced liquidation played a major role.

Has Korea finally found a bottom?

Morgan Stanley’s latest commentary has encouraged investors who believe the memory selloff went too far.

The firm described the decline in U.S. memory stocks as creating a “strong entry point,” arguing that the shortages supporting the industry still show no signs of disappearing. Morgan Stanley estimates that memory prices are rising approximately 25% during the current quarter. That is slower than previous increases, but hardly evidence of a collapsing cycle.

The analysts acknowledged several legitimate risks. Customers are signing longer-term supply agreements, redesigning products to use less memory and attempting to contain rapidly increasing costs. Nvidia is also believed to have reduced the memory content of some systems.

However, Morgan Stanley’s larger conclusion is that the underlying shortage remains intact and that the recent decline improved the risk-reward profile of Micron and Sandisk. The firm still favors Nvidia and Broadcom, but believes memory is catching up quickly following the correction. Investopedia summarized the Morgan Stanley note.

The more defensible interpretation is not that Korea is suddenly safe or that every memory stock has permanently bottomed. It is that the forced-selling phase may have reached exhaustion.

A durable bottom will still require several confirmations:

1. The Kospi must stop producing cascading liquidation events.
2. SK Hynix and Samsung must begin forming higher lows.
3. Micron and Sandisk must hold their rebounds instead of immediately surrendering them.
4. Hyperscaler earnings must confirm that AI capital expenditures remain intact.
5. Semiconductor breadth must improve beyond a handful of momentum leaders.

The first rebound is encouraging. The follow-through will tell us whether institutions are accumulating shares or traders are merely covering shorts.

Memory remains the heartbeat of the AI buildout

The market’s long-term memory thesis remains compelling.

Advanced AI accelerators are only as useful as the systems feeding them data. High-bandwidth memory allows enormous quantities of information to move rapidly between processors and memory stacks. As models grow, inference workloads expand and AI agents become more active, the demand for memory bandwidth continues increasing.

Traditional DRAM and NAND are also benefiting. AI data centers need storage, servers, networking infrastructure and enormous pools of working memory. That is why the rebound was not limited to Nvidia or AMD. It spread through Micron, Sandisk, Western Digital and SK Hynix.

The bearish argument is that the industry will eventually build too much capacity. Samsung, SK Hynix and Micron are spending aggressively, and additional supply arriving in 2027 or 2028 could pressure prices.

That risk is real. Memory has historically been one of the most cyclical parts of technology.

But the timing matters. New fabrication capacity cannot be switched on overnight, and advanced HBM production is considerably more complex than simply producing additional commodity DRAM. For now, demand continues to exceed available supply, and customers are signing multiyear agreements to secure access.

The cycle will eventually normalize. The market may simply be trying to price that normalization years before it arrives.

Photonics is becoming the next AI bottleneck

The rebound in Coherent and Lumentum may be even more revealing than the move in memory.

AI clusters increasingly require high-speed optical connections to move data between accelerators, racks and data centers. Copper works over shorter distances, but its power consumption and signal limitations become increasingly problematic as clusters grow larger.

That makes photonics essential infrastructure.

Coherent supplies optical communications components, lasers and related technologies used throughout advanced networks. Lumentum provides optical and photonic products positioned directly in the path of expanding data-center bandwidth.

These stocks have become high-beta expressions of the AI infrastructure cycle. When investors believe capital expenditures are accelerating, photonics names can move faster than the broader semiconductor sector. When investors question AI spending, they can decline just as violently.

This week’s rebound suggests the market is not yet prepared to abandon the optical-interconnect thesis. It also confirms that these stocks should be traded with smaller position sizes and clearly defined risk.

The midterm-year warning

The bullish rebound is occurring at a difficult point in the calendar.

Midterm-election years have historically produced weaker returns and higher volatility than other years in the presidential cycle. BlackRock calculates that U.S. equities have returned an average of 7.5% during midterm years, compared with 12.4% across all years. The two most recent midterm years—2018 and 2022—were both negative. BlackRock’s historical analysis also shows that markets have frequently strengthened after election uncertainty begins to clear.

Nasdaq Dorsey Wright’s data is more cautious, finding that the S&P 500 has averaged a slight decline during midterm years since 1960, followed by stronger performance beginning in the fourth quarter. Its 2026 analysis specifically warns that midterm years tend to produce unusually high volatility.

History is not destiny. Earnings, inflation, interest rates, energy prices and geopolitical developments matter more than the election calendar alone.

But the seasonal pattern supports one important conclusion: an aggressive July rebound does not eliminate the possibility of another correction during August or September.

That favors defined-risk trades, partial entries and disciplined profit-taking.

Potential August trade setups

The following setups use the August 21, 2026 monthly expiration and reference prices available around July 22. They are illustrative structures, not personalized recommendations. Option premiums and strikes must be checked before entry.

1. SMH: The diversified rebound

Structure: August 21 $580/$620 bull call spread

• Buy the $580 call
• Sell the $620 call
• Avoid paying more than approximately 40% of the spread’s width
• Consider exiting if SMH loses the rebound low or semiconductor breadth deteriorates

SMH offers exposure to the larger semiconductor recovery without requiring one company to deliver the entire move. It is the cleaner choice for traders who believe the sector has stabilized but do not want concentrated memory or photonics risk.

The short call reduces the cost and partially offsets elevated implied volatility. The trade sacrifices unlimited upside, but an August position does not need unlimited upside—it needs a realistic target and controlled downside.

2. Micron: The memory leader

Structure: August 21 $950/$1,050 bull call spread

• Buy the $950 call
• Sell the $1,050 call
• Prefer entry on a controlled pullback rather than another double-digit gap
• Keep the maximum debit below approximately 40% of the spread width

Micron is the most direct liquid U.S. expression of the AI memory thesis. It also carries substantial momentum risk after gaining roughly 200% this year and then falling sharply from its June peak.

The ideal setup is not to chase the opening surge. It is to watch whether Micron can digest the rebound, hold above the recent liquidation zone and build a higher low.

If it can, the decline may have reset positioning without breaking the underlying trend.

3. Coherent: The photonics catch-up trade

Structure: August 21 $310/$350 bull call spread

• Buy the $310 call
• Sell the $350 call
• Use a smaller position than the SMH setup
• Avoid entry if the stock immediately loses the $300 area

Coherent provides direct exposure to optical networking and photonics while offering a less extreme share price than Lumentum.

This is the highest-risk bullish setup of the three. It may also provide the greatest tactical upside if investors rotate back into optical-interconnect names and hyperscaler spending remains firm.

Lumentum offers a similar thesis, but its recent volatility makes it better suited to experienced traders willing to accept wider spreads and more violent price movement. I would choose one photonics position—not both.

4. SMH: Protection against a failed rebound

Structure: August 21 $560/$520 bear put spread

• Buy the $560 put
• Sell the $520 put
• Use it as a partial hedge rather than a standalone oversized bearish bet
• Consider closing it if SMH establishes a sustained breakout above the rebound range

This position protects against the possibility that the current move is merely a short-covering rally.

The hedge becomes especially relevant if semiconductor stocks rally into disappointing hyperscaler guidance, renewed Korean liquidation or evidence that AI capital expenditures are moderating.

Investors do not need to be outright bearish to own protection. In a midterm-year third quarter, hedging a concentrated book can be more rational than constantly trying to predict the next reversal.

What would invalidate the bullish thesis?

The rebound becomes much less convincing if:

• The SOX gives back most of its recovery within several sessions
• Micron breaks below its recent liquidation low
• Korea experiences another wave of margin-driven selling
• Hyperscalers reduce AI capital-expenditure guidance
• Memory pricing expectations begin falling rather than merely slowing
• Coherent and Lumentum underperform even while the broader market rises

The strongest confirmation would be a pullback that holds.

Healthy trends do not move vertically every day. They rally, consolidate, attract buyers at higher lows and eventually resume. If the momentum names cannot survive even a modest retracement, the rebound was probably mechanical rather than fundamental.

The bottom line

The AI trade is wounded, not dead.

The rebound across semiconductors, memory and photonics reflects more than indiscriminate dip-buying. It suggests investors still believe the AI infrastructure cycle has room to run and that the recent selloff was intensified by leverage, crowding and forced liquidation.

Morgan Stanley’s memory call reinforces that view. Shortages remain, pricing remains strong and the long-term demand picture has not collapsed.

But this is no longer a market where investors should confuse a compelling story with a safe entry.

Momentum remains elevated. Valuations remain demanding. Korea remains unstable. The third quarter of a midterm-election year has historically punished complacency.

The best August trades may therefore be the ones that participate in the rebound while assuming it could fail: defined-risk spreads, smaller sizing, staggered entries and a willingness to take profits before the market takes them back.