Cerebras Slips Below Its IPO Price: OpenAI Concentration, the Lockup Expiry, and What It Means for Buyers of AI Compute

Cerebras Systems, the AI chipmaker that went public in May at $185 a share, ended the first week of October below its IPO price. The pressure came from the lockup expiry, executive share sales, and a report questioning whether its anchor customer, OpenAI, is moving its fastest model to Nvidia GPUs.

What Happened to the Stock

Cerebras priced its IPO at $185 on May 13, 2026 and began trading on Nasdaq as CBRS; it closed its first day at $311.07. According to 247WallSt, in early October the stock was trading near $177.65, roughly 43% below that first-day close. Other outlets report that the week ended Friday at $166.43, a drop of about 20% over five days and its lowest level since the IPO, before it bounced 9.08% to $181.55 on Monday.

Three Pressures at Once

1. Lockup expiry. Up to 19.4 million shares held by directors, executives, and employees became eligible for sale on September 30. This is a standard mechanism: when a lockup expires, the potential supply of shares goes up.

2. Executive sales. According to 247WallSt, COO Dhiraj Mallick filed a notice of intent to sell 396,000 shares (about $77.9 million) and CFO Robert Patrick Komin Jr. filed for 32,500 (about $6.4 million). A notice of intent to sell is not an executed sale, and such sales often have personal or tax motives, but the timing feeds caution.

3. The OpenAI report. SemiAnalysis claimed that the "Ultrafast" mode of GPT-6.1 Sol would run on Nvidia GPUs rather than on Cerebras systems. This is an analyst report, not a confirmation from either company. Sam Altman responded that "Cerebras is a close partner" and that they have "a deep commitment to pushing the frontiers of speed," which helped the rebound.

Why Concentration Risk Matters

OpenAI has a multi-year agreement with Cerebras worth more than $20 billion and 750 MW of inference compute, according to 247WallSt. When one customer weighs that much, any sign that it is spreading its workload across suppliers moves the valuation: the same outlet notes the stock trades at about 217 times projected earnings, a multiple that only holds up if that contract keeps growing.

Cerebras vs. Nvidia GPUs: The Technical Read

Cerebras sells wafer-scale processors built for inference speed; Nvidia dominates with general-purpose GPUs and a mature software ecosystem. A model provider can use both: specialized chips for low latency and GPUs for general capacity and flexibility. That choice is dynamic and changes by model and by serving mode, which explains why a single report can move a stock so much.