Explain why CRDO is so expensive

$Credo Tech(CRDO.US) It's a matter of less than a week, and a single stock can swing more than 10%. Its market cap is over $30 billion, not some small, ignored stock. The elasticity in the AI optical communication track is just this high.

CRDO is Credo Technology—the Credo here is not the one for credit cards, it's the Credo known as "Broadcom's little brother." I'll explain it in plain language, focusing on answering three questions: what was its foundational business, why and what benefits did it gain in optical-electrical connectivity, and where is future tech iteration headed.

First question, what was its foundational business?

CRDO started by making SerDes (Serializer-Deserializer, serial/parallel conversion) chips. This sounds very technical, but it's essentially one thing—combining data transmitted slowly over multiple wires onto one wire for fast transmission, then splitting it apart at the other end. This component is found in all data centers, servers, network cards, and hard drives. They've been in this foundational business for over a decade, not a dominant player but stable.

But what really made it expensive is another product: AEC, Active Electrical Cable. Let me use a life analogy—your home ethernet cable is a passive cable, plug and play, fine for short distances. But in AI data centers, when the distance between GPUs gets a bit longer (5 meters, 7 meters), high-speed signals start to attenuate in passive cables. How does active cable solve this? It embeds a CRDO SerDes chip at each end as a signal "refueling station"—high transmission rate, sufficient distance, and crucially, much cheaper than optical modules. This is the niche market CRDO carved out in AI server internal interconnects.

Second question, what benefits did it gain in optical-electrical connectivity?

Here we need to break down "optical-electrical connectivity." Data transmission in AI data centers has three layers:

Within GPU clusters (inside a rack, a few meters) → This is the main battleground for CRDO's AEC active cables. NVIDIA, Amazon, Meta, and Google's AI clusters heavily use CRDO AEC to connect ToR switches and servers. This area forms a division of labor with traditional optical modules (handled by Coherent, Innolight, Xinyisheng) of "use AEC for short distances, optical modules for long distances."

Rack to rack (tens of meters) → Here, optical modules are dominant, and CRDO also makes optical DSPs (the "signal processing brain" inside optical modules). This area competes head-on with MRVL (Marvell)—MRVL is the leader in optical DSPs, CRDO is the challenger.

Between data centers (kilometer scale) → Not much to do with CRDO, that's the coherent optical communication field, handled by Lumentum, Coherent, etc.

CRDO benefits most explosively from the first layer—the explosive demand for AI server internal interconnects. Every NVIDIA HGX server, every ToR switch, needs AEC. If the number of GPUs doubles, AEC demand roughly increases by 2-3 times (interconnect is an n×n relationship). That's why its financial report growth far exceeds that of GPU sellers.

Third question, where is future tech iteration headed?

There are three paths, each meaningful for CRDO:

First, PAM4 → PAM6 → 224G. This is the generational leap in SerDes speed. CRDO is already established in the PAM4 generation, now competing with MRVL for the next generation of 224G/PAM6. Whoever mass-produces first and has a more stable BER (bit error rate) gets the orders.

Second, AEC moving from 800G to 1.6T. AI cluster bandwidth demand doubles every two years. CRDO is currently pushing 800G AEC, with 1.6T already under customer validation. This is its cash cow and growth engine overlapping product line.

Third, CPO (Co-Packaged Optics). This is a longer-term technology, packaging the optical engine directly next to the GPU/switch chip. NVIDIA, Broadcom, TSMC are all pushing it. CRDO's position in the CPO chain is currently not as prominent as COHR (Coherent, another stock I'll write about later today)—this is its hidden concern.

My judgment is that CRDO is not the most expensive stock in the AI chain (its P/S ratio is lower than ARM's and NVIDIA's), but its "expensiveness" is relative to its own history. Its current P/S of around a dozen times is because the market is pricing in the premium of being "NVIDIA's supporting supplier in the AI server interconnect field." This story is still playing out—AI clusters are indeed expanding, AEC orders are indeed increasing. But the sector's violent volatility (10% swings in a week) shows it's already a market darling, and it falls fast during pullbacks too.

In short, CRDO is not a stock you just buy on dips—it moves in sync with the entire AI optical communication sector (COHR, Lumentum, MRVL, Innolight). If the sector pullback starting around 5/13 continues for 1-2 weeks and CRDO falls back to the 170 range, I'd consider buying in batches. If it surges above 220, I'd take half off first—valuation expectations could become overly stretched.

If you still have doubts about whether its "fundamentals are strong enough to hold the current price," watch two things: whether the AEC revenue share continues to rise in the next earnings report, and whether Q3 provides a specific timeline for 1.6T AEC mass production. If either of these disappoints, the story will be repriced.