Cerebras Systems closed its first day of trading on Wall Street on Thursday with a market cap below the $100 billion mark, a monster debut that ranks among the biggest technology IPOs in history. The stock then dropped 10% on Friday, but the initial print already sent the message investors were waiting for: demand for AI chips extends well beyond Nvidia, and the market is willing to pay for a credible alternative.

As CNBC's Katie Tarasov detailed in a video report on the company, Cerebras was founded in Silicon Valley in 2016, and its product is not a graphics processing unit in the Nvidia mold. The company builds a custom chip roughly the size of a dinner plate, a design that places it in a category known as ASICs, or application-specific integrated circuits. A GPU is a general-purpose processor that happens to handle AI well, while an ASIC is engineered from the start for a particular job. Cerebras chose to optimize for AI inference, and that choice sits at the center of its market story.

The fundamental problem the company is trying to solve is a supply shortage. Nvidia's GPUs remain the default workhorse for AI, but they are costly and repeatedly sold out, and tech giants have spent the past two years scrambling to secure them. The scarcity has become a bottleneck for the industry, and it has pushed every major cloud provider to hunt for alternatives. The Cerebras debut is the clearest signal yet that the hunt has money behind it.

The economics explain the urgency. Nvidia's pricing power is a direct result of the imbalance between demand and supply, and any chip that can handle a meaningful share of the AI workload at lower cost or higher speed becomes an attractive hedge. That helps explain Thursday's reception on Wall Street.

The silicon bet

The form factor matters because a larger chip can hold more processing cores on a single piece of silicon, and that changes how data moves between memory and computation. For AI workloads, the payoff shows up at inference time.

Inference is the phase where a trained AI model performs its job. Training is the expensive, months-long process of feeding a model data. Inference is what happens every time a user prompts a chatbot, asks a digital assistant to summarize a document, or asks software to make a decision. It is the portion of AI the public touches, and the industry is now discovering that scaling it is one of the hardest problems in the field.

The trade-off is flexibility. A GPU can be pointed at many different kinds of workloads, which makes it well suited to a fast-moving field. An ASIC is more rigid, but inside its narrow lane it can deliver better performance per watt and per dollar. For inference, where the workload is predictable and repeated millions of times a day, that specialization is worth the trade.

The CNBC report identifies agentic AI as the shift pushing inference demand higher. Agentic systems go beyond answering questions. They plan, pick up tools, and execute multi-step tasks on their own, which means they generate many more inference calls per task than a simple exchange does. A basic chat request might run a model once. An agent that checks a calendar, drafts a message, and confirms an action could run the model dozens of times in a single task. Each step needs compute, and low-latency, high-volume inference is what determines whether the agent feels fast or slow.

That is the market position Cerebras is selling to investors who have watched Nvidia's supply problems reshape the industry. The report notes that custom ASICs are gaining traction precisely because they excel at inference. Google, Amazon, Meta and Microsoft all design custom ASICs in-house, a sign that the largest consumers of computing power treat specialized silicon as a strategic necessity rather than an experiment. Nvidia has moved into the category as well, following its $20 billion purchase of Groq in December.

A rising tide for ASIC startups

The Cerebras listing also serves as a door-opener for the rest of the field. A public debut anywhere near $100 billion gives the custom chip approach a benchmark, and it paves the way for other ASIC startups, including SambaNova, Rebellions and D-Matrix, to pitch investors of their own. Those companies have spent years making the case for specialized silicon while the market fixated on GPU supply. Now they have a comparable success story to point to.

Investors should keep the size of that debut in perspective. The stock closed Thursday below the $100 billion threshold and then sank 10% on Friday, a move that suggests some of the initial enthusiasm cooled within a day. First-day trading after a listing often reflects scarcity and momentum rather than a settled view of the company's long-term value, and a valuation of that scale leaves little margin for execution mistakes.

Cerebras is not going to dethrone Nvidia in the short term. Nvidia's software ecosystem, its long lead in data centers, and its own move into the ASIC category make that an unlikely outcome for now. The meaningful development is that the AI chip market is no longer a one-company story. A dinner-plate-sized processor from a startup founded in 2016 has become a public-market heavyweight, and the companies building alternatives to the GPU now have proof that their approach can hold real value. For everyone waiting on GPU availability, that is a sign that the squeeze is finally drawing a response.