Cerebras wind: strong accounts, halved quotation

Cerebras posted 193.4 million dollars of quarterly revenue. Core revenue grew 92 percent in a year to 191.3 million dollars. A multi-year deal with OpenAI for 750 megawatts, worth more than 20 billion dollars, was announced. Fast inference is on its way to AWS with Amazon. The IPO raised 6.4 billion dollars. It was called the largest chip IPO in history. That same week the stock crashed nearly 20 percent. The lowest price since the IPO was seen. The reason is known too. News arrived that a critical OpenAI job would go to Nvidia hardware. Strong accounts, weak quotation. Both lived in the same week. ^{[1] [2]}

I never read a chip as a chip alone. I read it with energy, money, and supply. Here the picture is clean. Demand exists, money exists, production exists. But the price tells another story. I think the gap sits in expectations, not measurement. The market priced Nvidia fear, not the accounts. Whoever misses this distinction misreads the numbers. My job is to put each number in its place. The income statement says one thing. The stock says another. Both are real. Both belong to the same company.

Down to details. First-quarter revenue rose 13 percent on the previous quarter. Yearly growth reached 94 percent. Hardware revenue grew 59 percent to 110.6 million dollars. Cloud and services revenue jumped 178 percent to 82.8 million dollars. Gross margin printed 45 percent. Hardware showed 41 percent, cloud 49 percent. Operating loss was 15 million dollars. Net loss stayed at 14 million dollars. On the other side the stock closed Friday at 166.43 dollars. It fell by more than half since the debut excitement. Lockup expiry pressure piled on top. The numbers are plain. Do not leave judgment to mood; read it with measurement conditions. ^{[3] [4]}

The technology side is concrete too. Codex-Spark promises output above 1,000 tokens per second. It plays to interactive jobs where latency matters. A disaggregated inference strategy is being built with AWS. AWS Trainium 3 does the prefill. Cerebras CS-3 runs the rest. Market value was 95 billion dollars on day one. It winked at the Meta and Alibaba league. Now it stands a little above 39 billion dollars. The product is a dinner plate-sized custom chip. A circuit built for inference jobs. It rents from its own data center like a cloud service. In January a 750 megawatt job worth over 10 billion dollars through 2028 was signed with OpenAI. Hardware is not sold; capacity rents. That is the model. ^{[5] [6]}

The money side runs in layers. The IPO brought 6.4 billion dollars. On top, a 1 billion dollar Series H closed in February. In January a 1 billion dollar working capital loan arrived from OpenAI. In April an 850 million dollar revolving credit was tied. The data center race runs on money. According to McKinsey, inference workloads will pass training before the decade ends. The bulk of the data center pie shifts there. The supply side moves too. Per the prospectus, 19.4 million shares came out of lockup. That means 8 percent of the total. Selling pressure as locks open is normal. The pressure may pass. Debt and share balance must be watched. ^{[7] [8]}

Let me set down the counter-view too. The bears may be right. Customer concentration is a risk. If OpenAI work slips to Nvidia, the story weakens. Insider sales shake trust. Lockup openings last for months. The bulls may be right too. The inference age is just beginning. If the speed gap is real, customers return. AWS distribution opens doors. I weigh both sides with numbers. Not with feelings. Fear and euphoria both pass. Capacity and contracts stay. Look at which weighs more.

My watch list is set as I close. Under which load are 1,000 tokens per second measured. What does independent measurement say. Is OpenAI volume truly slipping. Does Altman's partnership promise turn into figures. When do lockup openings end. Do insider sales stop. Does the cloud margin hold at 49 percent. Does the hardware margin recover. No verdict without these. I take no analysis seriously that skips measurement conditions. Neither should you. Let numbers speak, with their conditions.