Everyone already owns Nvidia. That’s exactly why it’s not where the edge is anymore. The real opportunity in AI is one layer down — the under-the-radar companies quietly selling the parts every data center has to buy. Here’s how to find them, and three names doing it right now.
Educational only — not financial advice.
The market backdrop: patience over chasing
It’s been a whippy stretch. The S&P 500 set a record above 7,600 in early June, then took a sharp chip-led hit — the Nasdaq fell more than 4% on June 4, its worst day since April 2025 — before rebounding on hopes of a US–Iran deal (the Dow jumped roughly 900 points on June 10 as oil eased). The index now sits in the mid-7,000s, with the 10-year Treasury yield near 4.5% and WTI crude in the mid-$80s.
The macro setup is hawkish. May CPI came in hot at +4.2% year-over-year (energy-driven; core +2.9%), the fed funds rate is holding at 3.50–3.75% after four straight meetings without a cut, and the easing bias is gone. The big catalyst ahead: the June 16–17 FOMC meeting, Kevin Warsh’s first as Fed Chair after Jerome Powell’s term ended in May.
Translation: this tape has punished anyone chasing highs. And with the recent volatility centered on chips, a useful way to study AI is one layer down — through the supply chain.
The “picks and shovels” framework
In a gold rush, the people who reliably got rich weren’t the miners — they were the ones selling picks and shovels. Same logic in AI. The front-page names get all the attention and all the valuation. But every one of them depends on a supply chain: chips, power, cooling, optics, and the cables that wire it together.
A supplier only earns a spot if it passes all four tests:
- It sells into a hot name. Customer concentration here is a feature, not a bug.
- It’s under-covered. Off the front page, low retail awareness.
- It has pricing power. Real margins, not a commodity it can’t defend.
- There’s a verifiable demand link. A named customer, a contract, a capex line that flows to them.
Run that filter on today’s AI buildout and three names stand out — and all three have quietly crushed the market over the past year:
1. Fabrinet (FN) — the optics layer
Recent price ~$611 · 52-wk range $248–$749 · about +149% over 1 year
What they sell: Contract optical manufacturing — the high-speed optical transceivers (800G today, 1.6T next) that move data between GPUs across an AI cluster.
Who buys it: Fabrinet is Nvidia’s primary optics supplier (also Cisco, Lumentum). Optical-communications revenue hit roughly $689M in Q4 FY25, up about 15% year-over-year.
Why it’s overlooked: A “boring” Thailand-based contract manufacturer — not a logo retail chases, despite sitting directly inside Nvidia’s networking stack.
The risk: Coherent was just named a second source on Nvidia systems Fabrinet had sole-sourced — watch for share shift. Plus customer concentration and rate sensitivity.
2. Credo Technology (CRDO) — the connectivity layer
Recent price ~$251 · 52-wk range $76–$270 · about +241% over 1 year
What they sell: Active Electrical Cables (AECs) — the “$300–$500 purple cables” that wire AI server racks more reliably than fiber over short runs.
Who buys it: Hyperscalers — AWS (shown in Trainium racks by AWS’s own CEO) and Microsoft. Sales up ~120% to roughly $425M; Credo pioneered the AEC category (est. ~88% share), a market seen reaching ~$4B by 2028.
Why it’s overlooked: A niche connectivity component, not a chip — it flies under the mega-cap radar despite being mission-critical to uptime (“link flaps” waste expensive GPU time).
The risk: SemiAnalysis has flagged that hyperscaler demand may be overstated and margins thinner than bulls assume. Heavy customer concentration.
3. Modine Manufacturing (MOD) — the cooling layer
Recent price ~$275 · 52-wk range $86–$323 · about +193% over 1 year
What they sell: Thermal management and data-center cooling — the gear that keeps dense AI racks from melting as the industry shifts to liquid cooling.
Who buys it: Hyperscaler data-center buildouts. Data-center sales rose ~31% sequentially in Q3 FY26, the order book now stretches roughly five years out, and management guides 50–70% annual data-center growth. A spin-off of its legacy unit (~Q4 2026) leaves a pure-play climate-solutions company.
Why it’s overlooked: An old-line industrial radiator maker re-rating into an AI infrastructure play — the market is still catching up to the pure-play story.
The risk: Cyclical industrial business; spin-off execution risk; the valuation has already moved.
On the bench (same theme, also real demand links): Comfort Systems USA (FIX), the contractor physically building the data centers; nVent Electric (NVT), electrical connection and liquid-cooling enclosures; and Vicor (VICR), power-delivery modules that sit right next to the GPU.
How to study this
This isn’t a “go all-in on small caps” pitch. It’s a way to read the AI buildout. The obvious AI names carry the headlines; the verifiable silent suppliers show you where the money actually flows. Optics and connectivity names (FN, CRDO) are high-beta and swing hard with sentiment. Cooling (MOD) is a longer-horizon industrial re-rate. Knowing which is which is the skill.
The discipline that separates investors from gamblers: write down your thesis before you act, and track whether it actually plays out. That last part — keeping score — is what most people skip, and it’s exactly why they never improve.
The bottom line
A hawkish Fed and energy-driven inflation cap the upside for now, while an Iran deal is the wildcard that could flip sentiment risk-on quickly. Chasing extended mega-cap tech at the highs is the crowded move. The differentiated opportunity is in the AI supply chain — study the verifiable silent suppliers alongside the obvious names, and check every thesis against what actually happens.
FAQ
What are AI supply chain stocks? Companies that sell critical components — optics, cables, power and cooling — into the AI data-center buildout, rather than the headline chipmakers themselves. Examples include Fabrinet (optics), Credo (connectivity) and Modine (cooling).
What is a “picks and shovels” AI investment? Focusing on the suppliers that every AI builder must pay, instead of only the most crowded front-page names — the modern version of selling tools during a gold rush.
Is Fabrinet a supplier to Nvidia? Yes. Fabrinet is Nvidia’s primary optical-components manufacturer, building the high-speed transceivers used to connect GPUs in AI clusters.
Trade the next session with us live. Start your 7-day free trial →
Want the full weekly report?
The MTC Investing Analyst publishes a complete long-term report every Friday — market environment, the silent-supplier spotlight, and the risks to watch. Built to make you an independent investor, not a signal-follower.
Disclaimer: This article is for educational purposes only and is not financial advice. It is not a recommendation to buy or sell any security. Company figures and prices are drawn from public reporting and market data and will change; always confirm current data and do your own research before making investment decisions, considering your own risk tolerance. Meta Trading Club.
Education only. Not financial advice.
