Swing Trading Notes: Finding My Entry Points on the AI Infrastructure Thesis

1. Who I Am: Positioning Myself as a Swing Trader

After enough years doing this, I've settled on where I fit ， I'm neither a long-term value investor nor a short-term speculator. I'm a swing trader.

The distinction, as I see it, is pretty clear:

• A value investor cares about whether a company's intrinsic value is undervalued. Holding periods run in years, and short-term price swings barely register.
• A short-term speculator cares about the day's money flow and order-book dynamics. Holding periods run in days or even hours really a bet on who takes the next handoff.
• A swing trader (me) cares about something different: within a sufficiently high-conviction industry narrative, a stock's price tends to move in a repeating rhythm of "rally — consolidation — rally again." My job is to keep buying during the consolidation phases and keep taking profit during the rally phases, for as long as the underlying narrative hasn't played out.

There's an old saying that captures this well: a stock's price action, at its core, is just a series of peaks and troughs of different sizes. No matter how strong the logic or how good the company, a stock never goes up in a straight line — there will always be pullbacks from profit-taking, cooling sentiment, or news-driven noise. The low points those pullbacks carve out are the troughs; the subsequent highs that fresh buying pushes the price to are the peaks.

My trading system really comes down to two things:

1. Pick the right narrative — find the industry thesis the market currently agrees has the highest conviction, and identify the names most representative of it.
2. Pick the right entry — don't chase the peaks. Wait patiently for the price to work its way down into a trough zone before building a position, then scale out near the next peak.

2. The Strongest Narrative Right Now: AI Infrastructure

If you asked me what the highest-conviction industry narrative in the market is right now, my answer would be: AI infrastructure.

To quickly map out what sits upstream and downstream in this chain, I like to reference industry-chain-mapping sites (industrychain.net's artificial intelligence page is one example). It breaks the AI industry down into upstream hardware infrastructure (compute hardware, memory & storage, optical networking, data center facilities), midstream models and platforms, and downstream applications — a genuinely clear way to see the whole picture. When I hunt for the narrative myself, I mostly follow that same order: upstream hardware moves first, and the mid-to-downstream applications monetize later.

2.1 Where the Signal Comes From: Capex at the Hyperscalers

To judge whether a narrative is "real" rather than hype, the first thing I look at is the signal furthest upstream — capital expenditure at the hyperscalers.

In this 2026 cycle, capex guidance from Google, Amazon, and Meta has stepped up noticeably. Pulling together various public disclosures, Amazon's 2026 capex guidance sits around $200 billion, Alphabet (Google's parent) is in the $175–190 billion range, and Meta's guidance range has been revised up to roughly $115–145 billion. Combined, the major hyperscalers' capex is approaching or exceeding $700 billion — a substantial year-over-year jump. Most of that money is flowing into AI data centers: compute, storage, networking, and power.

To me, the continued upward revisions in hyperscaler capex are the strongest signal that "the narrative isn't done yet." As long as these companies keep raising their investment, it means real-world demand for AI infrastructure — servers, chips, optical modules, PCBs, and so on — still has room to grow, and the thesis hasn't been invalidated.

2.2 How the Signal Propagates: The New Generation of AI Compute Racks

Once you've confirmed the capex signal at the source, the next question is which parts of the supply chain that money actually flows into.

The transmission path this cycle is fairly clear: hyperscalers raise capex → data center buildouts accelerate → Nvidia's next-generation compute racks (GB300, the Rubin platform) ramp faster. And the bill of materials (BOM) inside these racks is itself shifting structurally — teardown analyses of Nvidia's newer platforms show that the value of non-GPU components like PCBs and MLCCs is growing even faster than the GPU itself, meaning the value is spreading from the single chip out to the entire physical carrier.

Following that transmission chain, I've grouped the areas most worth watching into three main branches, plus a handful of smaller ones:

① Optical Modules (Optical Networking) High-speed interconnects within and between data centers rely on optical transceivers. As speeds move from 800G toward 1.6T and beyond, the "volume and price both rising" logic is fairly clean: new data center construction drives volume growth, while generational speed upgrades drive per-unit price increases.

② PCBs (Printed Circuit Boards) Newer racks demand much higher signal integrity, which lifts the value of high-layer-count, high-speed PCBs and the upstream materials that go into them — copper-clad laminate, copper foil, glass fabric, and so on. This is the clearest expression of "non-GPU component value outgrowing GPU value" this cycle.

③ Memory & Storage Chips AI servers use several times more DRAM, HBM, and NAND than traditional servers, and as manufacturers redirect capacity toward HBM, general-purpose memory supply gets squeezed. Memory chips have been in a fairly pronounced pricing up-cycle over the past couple of years as a result.

Beyond these three core branches, liquid cooling, power supplies and UPS systems, MLCC ceramic capacitors, and fiber optic cable are smaller sub-branches that benefit in parallel — logically, they're all different facets of the same underlying story: rising rack power density and rising data throughput.

2.3 Finding Representative Names

Once the narrative and its branches are confirmed, I look within each branch for the most representative "bellwether" names — the companies the market broadly recognizes as being first to benefit, with relatively clear earnings leverage. In optical modules, for example, that means companies leading in global shipment volume and first to ramp higher-speed products; in PCBs, companies that have plugged into leading compute platform supply chains early and are actively expanding capacity; in memory chips, companies with strong pricing power through the up-cycle and a product mix skewed toward enterprise-grade or AI-server applications.

One important note here: I'm deliberately not naming specific tickers in this piece. Partly because this kind of information changes fast and would go stale quickly; partly because stock selection itself reflects each person's own research and risk tolerance. What I want to share here is the how I think about it, not a what to buy list. This post records a methodology and is not investment advice — any specific names still require each reader's own independent research.

3. After Picking the Narrative: Finding the Trough Entry Point

Picking the right narrative and the right names is only step one. The real difficulty in swing trading is timing — judging whether the price is currently sitting in a peak or a trough.

My own approach is fairly simple:

1. First confirm the thesis hasn't been invalidated. As long as fundamental signals — capex, order books, capacity utilization — keep validating the narrative (hyperscalers continuing to raise capex guidance, the industry still in a pricing up-cycle, bellwether companies' order books still extending), the narrative is still "unfinished," and a pullback is more likely a buying opportunity than a signal to exit.
2. Watch the size and duration of the drawdown relative to the prior high. Mid-uptrend, individual stocks often pull back 20–30% or more — profit-taking, a broad market correction, or a temporary piece of bad news. If that drawdown isn't accompanied by any actual damage to the underlying fundamentals, it's often a trough zone.
3. Scale into a position rather than going all-in at once. Because it's genuinely hard to pin down the exact bottom on any given day, I prefer to buy in tranches through the drawdown zone — averaging down my cost basis while keeping some capital in reserve for a more extreme pullback.
4. Scale out in tranches near the prior high, or when there's clear volume expansion with stalling price action or the narrative's marginal momentum fading — rather than waiting indefinitely for some theoretical absolute top.

At the end of the day, swing trading isn't about making money by nailing the exact bottom and the exact top — it's about repeatedly executing "buy low, hold, take profit" on a narrative you actually understand and can keep validating, compounding returns through probability and discipline rather than through precisely predicting every price turn.

4. Closing Note

This post records my personal trading system and current read on the AI infrastructure narrative. It's meant as a record of my thinking and methodology, and it is not investment advice. Markets move fast — capex guidance, supply chain pricing, and company fundamentals will all keep changing. Any decisions should be based on your own independent research and your own risk tolerance.

---

References (public sources for further reading):

• Industry Chain Atlas — Artificial Intelligence Industry Chain Map (industrychain.net)
• Futurum Group, "AI Capex 2026: The $690B Infrastructure Sprint"
• Various financial media coverage of 2026 hyperscaler capex guidance, Nvidia's Rubin-platform BOM teardowns, and the optical module / memory chip pricing up-cycle