Three AI Picks-and-Shovels Names: Applied Digital, Credo, Amphenol — Where the Alpha Actually Sits

The Real AI Winners Aren't the Chips — They're What Sits Around Them

Nvidia can make all the GPUs they want. Without buildings to put them in, cables to move data between them, and electricity to power them, the cycle doesn't roll. In the 19th-century gold rush, the most consistent money was made selling picks and jeans. The AI cycle has the same layer underneath it.

Today I'm looking at three of them: Applied Digital, which builds the buildings; Credo Technology, which moves the signals between chips; and Amphenol, which lays down the physical infrastructure that every byte travels through.

1. Applied Digital — A Contract Backlog Larger Than the Market Cap

Applied Digital isn't a generic data center company. They don't do cloud, enterprise, and AI. They only build AI-workload-only 'AI factory campuses.'

Start with the numbers.

• April new lease: 15-year deal with an investment-grade hyperscaler, $7.5 billion
• Total contracted backlog: over $23 billion
• Current market cap: roughly $13 billion

Already-signed contracts are worth nearly 2x the entire market cap. Even if the company never closes another deal, executing the existing book should drag the market cap up to it.

Operationally, Polaris Forge 1 in North Dakota is the flagship campus. CoreWeave is the anchor tenant. The April lease added a second investment-grade hyperscaler at a new campus. The design is liquid-cooled, ultra-high-density compute racks, sited next to abundant natural gas where power is cheap and the grid has headroom.

The roadmap takes contracted capacity from 600 MW today to over 1,000 MW in the pipeline. The cloud computing business has been spun off as a separate entity, with Applied Digital retaining about 97% of it. Practically, that lets the AI real estate and AI cloud businesses each scale on their own capital structures.

2. Credo Technology — Fastest Growth, Lowest PEG

Credo doesn't make chips. They make the connections between chips. Active electrical cables, optical digital signal processors, and the high-speed interconnect that lets thousands of GPUs in a data center talk to each other fast enough to actually work.

Two April datapoints stood out.

1. DustPhotonics acquisition: $750 million for a silicon photonics company. Vertically integrates their high-speed signaling expertise with optics.
2. Optical revenue guidance: greater than $500 million in optical-only revenue by fiscal 2027.

Customer concentration also moved meaningfully. Microsoft used to be 86% of revenue. That number is now 42% — not because Microsoft shrank, but because the rest of the hyperscaler base ramped up fast. Reducing single-customer dependency is a structural change to the multiple they deserve.

Their technical edge is power efficiency. Credo's optical signal processors run at 30 to 50% lower power than competitors at the same speed. When a building holds hundreds of thousands of GPUs, a 30% efficiency difference translates directly into massive operating-cost differences.

And the most striking pair of numbers:

• Year-over-year revenue growth: 201%
• PEG ratio: 0.9

Fastest growth on this list combined with the lowest growth-adjusted valuation. That combination is rare.

3. Amphenol — Quietly Buying Up the Entire Data Center Signal Path

Amphenol isn't usually called an 'AI stock.' That's the opportunity. They make the connectors and high-speed cables that physically link everything inside a data center. The same products serve autos, aerospace, and mobile — but AI is what's accelerating.

Their Q1 was a record on every line.

• IT datacom organic growth: essentially all from AI
• New orders: $9.4 billion in the quarter, up 78% YoY
• Book-to-bill: 1.24 — orders coming in faster than shipments going out

In January, they closed a $10 billion acquisition of CommScope's connectivity and cable solutions business. Amphenol now covers the full data center signal path: high-speed copper, power, active copper, passive fiber, and active optics.

This isn't just an acquisition. It's a positioning statement. They went from 'the company that connects chips' to 'the company that connects everything that moves data inside an AI factory.'

One detail I find interesting. A typical 2026 GPU rack uses 2 to 3 times the interconnect length of a 2024 rack. As compute density rises, cabling density rises in proportion. Amphenol is betting that ratio keeps climbing.

Three Companies, One Table

Company	Position	Key Number	Key Risk
Applied Digital	AI-only data center real estate	$23B backlog vs $13B market cap	Tenant concentration, capital intensity
Credo Technology	High-speed interconnect between chips	Revenue +201% YoY, PEG 0.9	Optical transition risk, customer concentration
Amphenol	Full data center signal path	Q1 orders $9.4B (+78%), B/B 1.24	Acquisition integration execution

All three sit on different floors of the same cycle. If I had to pick one based on growth rate, Credo. Based on stability, Amphenol. Based on asymmetric upside, Applied Digital.

What I Actually Think This Adds Up To

The reason I look at all three together is straightforward. The most visible parts of the AI cycle — GPUs, models — already have institutional money fully positioned. The alpha is in the layers above and below. And the infrastructure layer has one decisive advantage: once a contract is signed, forward revenue visibility locks in for 5 to 15 years.

My personal portfolio is rotating toward infrastructure names over chip names. Lower volatility, locked-in forward revenue via backlog, and indifferent to which AI company wins the model race — because every model runs on top of this infrastructure regardless.