Cerebras Systems shares fell as much as 8.87% on Wednesday after SemiAnalysis reported that OpenAI’s GPT-6.1 Sol Ultrafast runs on Nvidia GPUs with low batch processing rather than Cerebras’ Wafer-Scale Engine. The tier reportedly delivers about 300 tokens per second, compared with roughly 750 tokens per second for the earlier GPT-5.6 Sol Ultrafast tier that Cerebras said it would support in August. OpenAI is Cerebras’ largest customer by committed revenue backlog, making the reported infrastructure change particularly important to investors. Neither company has confirmed or denied the claim. The low-batch detail is significant because Cerebras markets its single-wafer architecture as an alternative for fast, low-latency inference, while Nvidia systems typically gain efficiency from processing larger batches. Shares have also faced pressure from IPO lockup releases, a reported GAAP quarterly net loss of $450.5 million and approximately $266 million in insider share sales over the prior three months. The supplied accounts differ on the IPO reference price, with one citing $185 and the new report saying the stock debuted at $350 in May.