AI infrastructure is no longer only a chip story. Credo Technology Group, ticker CRDO, shows how the plumbing around the chips is becoming a serious market in its own right.

The numbers are large enough to make investors pay attention. They are also large enough to make sensible people ask what happens if AI data center spending slows.

Credo moves from niche to noticed

Credo reported revenue of $437 million, up 157%. That is not a normal growth rate for a hardware supplier. It is what happens when a small company sits in a part of the stack that suddenly becomes scarce.

The gross margin was 68.3%. Free cash flow was $177.5 million. For a company tied to semiconductors and data center equipment, that combination matters. High growth is nice. High growth with cash generation is less common.

The market is not paying for yesterday’s revenue. It is paying for the possibility that Credo becomes a core supplier to large cloud operators building AI factories. That phrase sounds like marketing, but the engineering problem is real. The machines need to talk to each other fast, with low delay, and without silly failure rates.

There is a quiet lesson here. Compute gets the headlines. Networking gets the invoices. If the network is weak, expensive graphics processors spend too much time waiting. Nobody buys a racing car and then accepts a dirt road to the office.

Optical revenue becomes the test

Management guided for more than 80% revenue growth in fiscal 2027. Optical products are expected to contribute more than $600 million. That is the line investors will watch, because it moves Credo from impressive growth story to possible infrastructure standard.

Optical links matter because AI clusters keep getting larger. A model training run is not just one chip doing heroic mathematics. It is many chips, racks, cables, switches, and transceivers acting like one machine. Every extra hop creates delay. Every weak part reduces useful throughput.

This is why bandwidth is becoming a balance sheet issue for cloud companies. Hyperscalers can buy more compute, but the system only performs if data moves where it needs to go. Credo’s opportunity sits in that practical mess.

The more interesting point is margin. If optical products scale while gross margins stay near current levels, earnings can rise faster than revenue. If price pressure arrives, the story changes quickly. Hardware markets are friendly until customers remember they have purchasing departments.

Forecasts assume the AI build keeps running

Analysts expect revenue to move from $2.44 billion in fiscal 2027 to $6.78 billion by fiscal 2031. That would be a very large jump from the current base. It implies that AI networking demand keeps expanding for years, not quarters.

There is a reasonable argument behind it. Large language models, recommendation systems, robotics training, and cloud AI services all need more data center capacity. Companies do not just need processors. They need memory, power, cooling, networking, software orchestration, and a tolerable electricity bill.

Credo benefits from the part of this chain where scale creates pain. Bigger clusters are harder to connect. Reliability becomes more valuable. Latency becomes less forgiving. These are not decorative features. They decide how much useful compute a customer gets from equipment that already cost a fortune.

Still, a forecast is not a contract. The path from $2.44 billion to $6.78 billion assumes continued capital spending by the largest buyers. Those buyers can delay orders. They can redesign systems. They can push suppliers on price. The customer concentration risk in AI infrastructure is not a footnote. It is the business model.

Rates still matter for AI winners

The market backdrop is less forgiving than the AI narrative suggests. Persistent inflation and a more cautious Federal Reserve keep pressure on long duration growth stocks. That matters for companies whose valuations depend on earnings several years away.

The Nasdaq 100 and SPY have both been treated as vehicles for the AI trade. When yields rise, investors ask harder questions about growth that arrives later. Semiconductor and networking stocks are among the first places this question appears, because the expectations are already rich.

This does not mean every AI supplier is overvalued. It means the cost of capital has a vote. A company can execute well and still see its share price struggle if investors demand a cheaper entry point. Mathematics is rude like that.

Credo’s cash flow helps here. Free cash flow gives the business more credibility than a pure promise stock. But valuation still depends on future growth. If the market decides that AI capital spending is peaking, cash flow from one strong year will not settle the argument.

What to watch

The first item is optical product revenue. If it moves toward the more than $600 million expectation without heavy margin damage, the market will have stronger evidence that Credo is gaining real share in AI networking.

The second item is hyperscaler spending. AI data centers are expensive systems, and their budgets can change when financing costs, power constraints, or internal return targets change. A delay at a large buyer can travel quickly through the supplier chain.

The third item is valuation discipline. CRDO has the kind of growth that attracts attention, but attention is not the same as safety. The useful question is not whether AI networking matters. It does. The useful question is how much future perfection is already in the price.