Great. You've heard that pitch a thousand times. Your coworker says it. CNBC says it. The guy at the gym says it between sets.

Here's what they're not saying: the companies printing money right now aren't the ones with the flashy logos. They're the companies that make the machines that make the chips that make the GPUs that power the AI your coworker won't shut up about.

The boring guys. The toll collectors. The picks-and-shovels guys.

And most retail traders are completely ignoring them.

Let's fix that.

Everyone is piling into Nvidia.

Which is fine. NVIDIA is great. NVIDIA has been great. NVIDIA might keep being great.

But here's the thing nobody talks about: Nvidia can't conjure chips out of thin air. Neither can AMD. Neither can any hyperscaler with a trillion-dollar market cap and a CEO who talks like he's already met the robots.

Chips require wafers. Wafers require lithography. Lithography requires tools. Tools require subsystems. Subsystems require precision components, specialty gases, and materials that would make a chemistry teacher's eyes water.

Right now? Every link in that chain is constrained.

Translation: the best trade isn't the chip. It's everything that has to happen before the chip exists.

Let's start with memory pricing, because it's the heartbeat of this entire trade.

TrendForce just upgraded its Q1 2026 contract price forecast. Their numbers: conventional DRAM up 90-95% quarter over quarter. NAND flash up 55-60%.

Read that again. Ninety to ninety-five percent. In one quarter.

That's not a blip. That's a signal flare.

Data centers are now absorbing over 70% of high-end memory output, according to some projections. That means regular humans, the ones buying laptops and gaming PCs, are fighting over the scraps. Fabs can't just flip a switch and make more.

Building a new fab takes years. TSMC confirmed it on its January 2026 earnings call. CoWoS packaging facilities, the ones that matter most for AI chip stacking, have been fast-tracked from 3-5 years down to about 1.5-2 years. Monthly CoWoS capacity is expanding from roughly 70,000 wafers/month at the end of 2025 to 115,000 wafers/month by the end of 2026. That's real progress.

It's also still not enough. The demand is now. The capacity comes later.

This gap-demand-now, supply-later is your trade.

TSMC guided capital expenditures of $52-56 billion for 2026. Let that sit for a second.

$52-56 billion. In one year. From one company.

Here's how they're spending it: about 70-80% goes to advanced process nodes. About 10% to specialty tech. And somewhere between 10-20% to advanced packaging, testing, mask-making, and related infrastructure.

That last bucket, the packaging, testing, and mask-making category, is the one retail traders are sleeping on.

Micron announced a broader $200 billion U.S. investment vision for manufacturing and R&D. Its first Idaho DRAM fab is expected to go online in 2027. SK Hynix is building advanced packaging facilities in both Cheongju, South Korea, and Indiana, totaling an investment of $3.87 billion. Samsung is on track to deliver HBM4 in Q1 2026.

The upstream order book is massive. It doesn't clear overnight.

This isn't financial advice. This is pattern recognition applied to public data. Do your own research. Size appropriately. Don't let any newsletter, including this one, replace a risk management brain.

That said. Here are the names worth having on your radar.

ASML Holdings (ASML)

ASML makes EUV lithography machines. There is exactly one company in the world that makes EUV machines. That company is ASML. If leading-edge chip production is going to scale, ASML has to be involved. Full stop.

The numbers: Q4 2025 total net bookings came in at €13.2 billion, nearly double analyst expectations. For context: €7.4 billion of that was EUV alone. The total backlog stands at €38.8 billion. 2026 revenue guidance is €34-39 billion.

Think about what a €38.8 billion backlog means. That's years of visibility. That's customers pre-paying for capacity that doesn't exist yet because they know if they wait, they won't get it. That's either rational planning or the most expensive FOMO in semiconductor history. Probably both. Q4 also included revenue recognition on two High-NA EUV systems, the next generation of lithography tooling. The cycle isn't peaking. It's stacking.

China exposure is a live risk. China was projected to account for roughly 20% of ASML's 2026 revenue on its Q4 call. One geopolitical headline, one surprise export restriction, and that reprices overnight. It's happened before. It'll happen again.

The trade: ASML is a duration play. This isn't about next quarter's earnings beat. It's about multi-year capex cycles. Call spreads at 9-15 months out are a cleaner way to express than trying to time a single print.

Watch for: EUV output cadence updates, export control headlines, and bookings revisions.

Support levels: 50-day MA around €1,070, 200-day MA around €821 on the Amsterdam listing.

Applied Materials (AMAT)

Applied makes the equipment that deposits, etches, and polishes the material layers that make up a chip. They're in everything front-end process and increasingly in advanced packaging.

The backlog speaks: $15.0 billion as of October 2025. Here's the interesting part about 31% of that backlog is NOT expected to be filled within 12 months. That's a deliberate demand signal. The pipeline is so full it's spilling into the next fiscal year.

As advanced packaging becomes a larger capex bucket (see: TSMC's 10-20% allocation), Applied's exposure matters more than the market currently prices in.

The trade: AMAT is the "steady compounder with upside torque" play. Long call spreads give you leverage to the capex cycle without the full volatility risk. Avoid initiating right after earnings gap-ups.

Watch for: backlog conversion rate, memory re-acceleration signals, quarterly guidance tone on packaging tools.

Support levels: 50-day MA around $296, 200-day MA around $218.

Lam Research (LRCX)

Lam is the etch-and-deposition specialist. Their FY2025 revenue mix was approximately 42% Memory and 45% Foundry, almost perfectly positioned for a market where HBM and AI accelerator production are the two dominant capex themes.

When both memory and foundry capex are expanding simultaneously, Lam gets pulled from both ends. They explicitly called out 3D stacking and HBM-related packaging needs in their documentation. They know where this is going.

The trade: LRCX tracks memory capex cycles closely. Watch Micron, SK hynix, and Samsung capex commentary for leading indicators on Lam's order outlook.

Key risk: export controls on China WFE. Lam specifically warns that trade restrictions and tariffs can hit revenue and margins. This isn't theoretical. It's happened before. It's happening now. Stay hedged around policy events.

KLA Corporation (KLAC)

KLA is process control. Metrology. Inspection.

Here's why this matters more as chips get harder: when you're stacking memory dies and building HBM at nanometer-level precision, yield losses are catastrophically expensive. A bad HBM stack doesn't just waste a chip. It wastes all the chips in the stack, all the packaging materials, and all the process time. KLA's tools are how fabs keep yield up when processes push toward the edge of physical limits.

KLA's FY25 shareholder letter explicitly called out AI infrastructure, advanced packaging, and HBM investments as growth drivers. Backlog came in at $7.86 billion, down from $9.83 billion, but that drop was driven by easing supply constraints and faster shipping. Revenue recognition accelerated. That's a good problem to have. Most companies would kill for it.

The trade: KLA tends to benefit late in the ramp cycle as fabs get serious about yield. Think of it as the "things got real, now we need to not waste product" play.

Support levels: 50-day MA around $1,369, 200-day MA around $1,063.

Entegris (ENTG)

Entegris lives in contamination control and advanced materials. Unglamorous? Absolutely. Necessary? Without question.

Here's the math: HBM stacking is expensive and complex. Yield failures at that stage are brutal. When fabs are running expensive advanced packaging processes, contamination control becomes more valuable, not less. The cost of a defect scales with the complexity of what you're protecting.

Translation: the more valuable the chip, the more Entegris gets paid to make sure it doesn't die on the table.

The trade: ENTG historically trades as "steady growth with cyclical drawdowns." When the memory cycle turns upward, ENTG initially lags, then catches up. Long-dated call spreads are for the patient. Entries near moving average support during broader market risk-off events have historically been the better setups.

AXT, Inc. (AXTI)

AXT sits even further upstream: they make compound semiconductor substrate wafers of indium phosphide (InP), gallium arsenide (GaAs), and germanium for applications where vanilla silicon just doesn't cut it, including high-speed optical links tied to AI data centers.

In that world, InP is the leverage point. AXT has been seeing a sharp ramp in InP demand from data center and passive optical network (PON) customers, and management is framing it as part of a multi‑year data center expansion trend rather than a one‑off blip. They’re responding the only way you can in an upstream squeeze: by more than doubling InP capacity into the back half of 2026.

That’s the whole macro thesis in micro form: demand hits hard, capacity takes time, and the guys closest to the bottleneck get outsized pricing and volume leverage while everyone downstream complains about “shortages.”

The catch: this is not ASML. It’s a small, speculative materials name with customer and geographic concentration that lives very close to the blast radius of export policy and optical-capex mood swings. When it works, it trades like a leveraged call on AI connectivity; when it doesn’t, you get a crash course in gap-downs.

The trade: treat AXTI as a satellite, not a core. Size it like a high‑beta call on InP demand. Defined‑risk options 6–12 months out or small equity with hard, pre‑committed puke points. No “it’s just down, I’ll average forever” heroics.

Qnity Electronics (Q)... This one has a different story.

DuPont spun out its electronics business into Qnity Electronics (NYSE: Q) on November 1, 2025. Pure-play electronics materials. $4.3 billion in net sales in 2024. Direct exposure to chip fabrication and advanced packaging.

Fresh spins are interesting because of mechanics, not fundamentals, at first. When a company is spun out, index funds that held the parent don't necessarily hold the spinoff. Institutional investors doing a DuPont analysis aren't necessarily doing a materials deep dive. Forced selling happens. Mispricing happens. The market takes a few quarters to figure out the right multiple for a standalone materials compounder with AI exposure.

The forward P/E is elevated around 38x, but that's what happens when the market is still price-discovering a new standalone entity. Watch the first two or three quarters of clean standalone results.

Support: 50-day MA around $91.70.

Amkor Technology (AMKR)

Amkor is an OSAT Outsourced Semiconductor Assembly and Test. They package and test chips. As advanced packaging becomes mission-critical for AI accelerators, OSATs become increasingly strategic.

Amkor is building an advanced packaging campus in the U.S. Production expected in 2028. That's not a "this quarter" trade. That's a "the market will eventually re-rate this as strategically critical infrastructure" trade.

The risk: 2028 is far away. A lot can change. Size accordingly. This is a long leash on a slow-moving thesis, which means it's also the kind of position that quietly doubles while everyone else is distracted by the next hot ticker.

ASE Technology (ASX)

ASE is one of the largest OSATs in the world. Direct exposure to the advanced packaging cycle. As the AI packaging mix improves across their customer base, watch for updates on capacity utilization in their quarterly reports.

Less headline-grabby than AMKR. More established. Lower volatility, probably.

Spoiler: "lower volatility" is what people say before they forget to check the position. Don't do that.

Ichor Holdings (ICHR)

Ichor designs fluid-delivery subsystems for semiconductor equipment. Gas delivery modules. Precision flow components. The stuff that has to work perfectly before a $100 million EUV machine can do its job.

The customer concentration here is extreme: Applied Materials, Lam Research, and ASML combined accounted for 73% of Ichor's total revenue in 2024. Three customers. Nearly three-quarters of the business.

That's either terrifying or tradable. It's both.

When WFE ramps, Ichor rides it hard. When WFE pauses, Ichor takes heavy damage. This is levered beta inside semicap.

The trade: tactical, defined-risk options structures around earnings. Not a "core position and forget it." More like a "here's my thesis for the next 60 days" play.

Don't marry this position. Ever. You've been warned.

Rambus (RMBS)

Different kind of upstream. Rambus doesn't make atoms it makes IP. Specifically, the memory interface IP and controller chips that let HBM and DDR5 signal at the speeds AI workloads demand.

FY2025 results: record annual product revenue of $347.8 million, up 41% year-over-year. The story here is design wins driving recurring revenue as next-gen memory architectures proliferate.

The trade: RMBS typically moves on design-win announcements and product-cycle catalysts. Options structures around those events when you have a specific thesis can be cleaner than just holding shares through the noise.

Linde (LIN) / Air Products (APD)

Here's the category the article almost skipped. Don't skip it.

Chip fabrication is one of the most gas-intensive industrial processes on the planet. Ultra-high-purity nitrogen, argon, fluorine compounds, specialty etchant gases, none of this is optional. You don't build a leading-edge fab without a specialty gas supply agreement in place first.

Both Linde and Air Products have long-term supply contracts tied directly to fab construction and ramp timelines. When fabs build, gas suppliers bill. When fabs ramp production, gas consumption compounds.

These aren't explosive upside-down stories. They're compounders with inflation-resistant pricing and captive customers. Think of them as the utility layer underneath the utility layer.

The trade: long-dated equity positions or dividend reinvestment. This is the "boring wins" tier.

Knowing the tickers is half the battle. Knowing how to express the thesis without getting destroyed is the other half.

For the core basket (ASML, AMAT, LRCX, KLAC):

Call spreads at 9-15 months out. Buy a strike near the current price or slightly out of the money. Sell a higher strike to reduce cost. This caps your upside but also limits the damage if the thesis takes longer to play out than expected.

Why spreads and not just calls? These names are already priced for optimism. Pure long calls on high-multiple, high-IV names can decay brutally even when you're directionally right. Spreads let you rent the thesis without overpaying.

For materials (ENTG, AXTI, Q):

Entries near the 50-day moving average support. Be patient. These move more slowly except when they don't. In something like AXTI, think "small satellite bet" rather than "core holding."

For the packaging ecosystem (AMKR, ASX):

These are longer-duration stories. Think about them in terms of fundamental re-rating over 2-3 years, not 2-3 months. If you're options trading these, go further out in time and size smaller.

For tactical names (ICHR, RMBS):

Defined-risk structures only. If you're wrong, your loss should be the premium you paid, not an open-ended equity drawdown. Know your thesis, know your catalyst, know your exit.

For gases (LIN, APD):

Equity. Long-dated. Reinvest the dividend. Check it quarterly. Forget about it otherwise.

Let's be honest about the downside. Because there is one, and pretending otherwise is how people blow up accounts.

The capex pause.

One bad earnings call from a hyperscaler, and this whole thesis reprices in a day. Demand is forward-looking. The stock market moves before the fundamentals do. You'll be wrong before you know you're wrong. That's not a bug in markets, that's the feature.

Memory cycle whiplash.

This has happened before. Memory prices overshoot, fabs oversupply, prices crater, capex freezes. The cycle isn't dead. AI demand is absorbing so much output that the swings are less violent. "Less violent" is not "impossible." Ask anyone who held semiconductor names during the 2022 correction.

Export controls.

This is the wild card. Geopolitical tensions around semiconductor technology are real and escalating. Both Lam and Applied explicitly warn about this in their filings. ASML is staring at roughly 20% China revenue exposure. A surprise restriction announcement can gap these names down overnight. Hedging with SOXX or SMH put spreads around known policy events isn't paranoid. It's basic hygiene.

Names like AXTI, small, upstream, and exposed to optical and comms demand live even closer to this blast radius. If export policy coughs, they catch pneumonia.

Backlog illusion.

Backlog is not revenue. Applied literally says this in their 10-K. Backlogs can be amended, delayed, or canceled. Treat backlog as a sentiment indicator and a capacity proxy, not a guaranteed income stream. The number is real. The timing is not.

You don't need to watch 47 indicators. Watch these.

Everyone is buying the car.

You should be thinking about the assembly line. The tooling. The materials. The factory floor where none of the Instagram photos are taken.

That's where the money is building. Not quietly, just not where most retail traders are looking.

The shortage is real. The capacity takes years to build. Pricing power is already evident in the data.

The question isn't whether this trade exists. The question is whether you're going to be in it before the crowd figures it out or after.

You know which one you'd prefer. Act accordingly.

Tickers referenced in this issue:

ASML · AMAT · LRCX · KLAC · ENTG · AXTI · Q · AMKR · ASX · ICHR · RMBS · LIN · APD · SOXX · SMH