CRDO As An AI Play
Expanding connectivity TAM from Silicon to Facility
Welcome to the premium version of Investing in AI, where we analyze companies through the lens of AI trends and summarize how AI might help or hurt their business models. Please keep in mind we are AI people not financial people so do your own research before making an investment decision. Disclosure: We do not have any position in CRDO.
With all the talk surrounding the compute war, especially with Anthropic’s Opus 4.5 release, the market’s obsession has remained firmly fixed on the “brain” of the AI, or the xPU. Investors have spent the last year scouring the horizon for the next great chip architecture, debating TFLOPS and HBM3e densities. But as models scale toward trillion-parameter complexity, a quiet, more physical crisis has emerged. The “compute war” is no longer being won just by how fast a chip can think, but by how reliably that chip can talk to the rest of the cluster.
As such, those well aware about the various technicalities of AI infrastructure understand the sheer opportunity that lies beyond raw processing power alone, and particularly within the elimination of friction.
In massive-scale environments, “performance without reliability” is essentially a sunk cost. When a single link failure (a “flap”) occurs in a 100,000-GPU cluster, the entire synchronized training run can stall, costing operators millions in wasted power and idle time. The real alpha in this cycle, therefore, belongs to the “Fabric Architects”—those who can turn a chaotic mess of copper and lasers into a singular, resilient high-speed nervous system.
This is where Credo Technology Group (NASDAQ: CRDO) enters the scene. Initially operating as a specialized provider of high-speed Serializer/Deserializer (SerDes) IP and Active Electrical Cables (AECs), Credo is undergoing a profound metamorphosis. It is no longer just a “cable company” or a component vendor; it is becoming the primary architect of the AI Connectivity Fabric.
The company has pivoted from providing discrete parts to delivering a comprehensive, vertically integrated system that solves the “physical layer” crisis of the AI era. While the market was distracted by GPU thermal limits, Credo was busy mapping the entire connectivity stack—moving inward to the silicon and outward across the data center floor.
In this week’s newsletter, we dissect the strategic blueprint that has propelled Credo from a niche component vendor to the structural architect of the AI era. While the broader market remains fixated on the “compute” layer, we highlight why Credo’s shift toward a vertically integrated connectivity fabric represents one of the most asymmetric opportunities in the semiconductor space today. We will explore how Credo is effectively “taxing” every millimeter of the data center—from the internal memory beachfront of the xPU to the facility-wide optical spans—by solving the industry’s most expensive problem: The Reliability Gap.
