Optimistic about continued growth in AI inference demand, Freedom upgrades Cerebras (CBRS.US) rating to "Buy," wafer-scale chip advantages draw attention.

Freedom Capital Markets upgraded AI chipmaker Cerebras Systems (CBRS.US) from "Hold" to "Buy" and set a price target of $209. The firm believes that although Cerebras recently suffered a market selloff after news that key partner OpenAI switched to NVIDIA Corporation (NVDA.US) GPUs, the recent pullback offers an opportunity to reassess its investment value, given that AI computing demand is expected to continue exceeding supply in the coming years and Cerebras's technological advantages in high-speed AI inference. Cerebras is a company focused on AI computing chips, with its core product being the Wafer Scale Engine (WSE). Unlike traditional GPU architectures, the company integrates a large amount of computing resources on an entire wafer to provide high-performance computing capabilities for large AI models, with particular emphasis on improving inference speed. OpenAI's Adoption of NVIDIA Corporation GPUs Sparks Concerns, Cerebras Stock Falls Freedom analyst Paul Meeks noted in a research report released on Friday that Cerebras's recent stock pressure is mainly related to market concerns about the technology choices of its key partner OpenAI. It is reported that OpenAI adopted NVIDIA Corporation GPUs rather than Cerebras's wafer-scale chips for the "Ultrafast" version of its GPT-1 Sol model. This news sparked investor concerns about whether Cerebras can continue to secure OpenAI orders in the future and about its product's competitiveness in the high-speed AI inference market. Meeks said the related news may be an important reason for Cerebras's recent cumulative stock decline of about 17%. Cerebras completed its initial public offering (IPO) in mid-May this year at an issue price of $185 per share. After listing, the stock quickly rose to above $300, but recently retreated significantly as the market reassessed its competitive outlook. However, Meeks believes investors may have overamplified the negative impact of OpenAI's adoption of NVIDIA Corporation GPUs. In the firm's view, the AI computing market is still in a phase of rapid expansion, and a single customer adopting a competitor's hardware for a certain model does not mean Cerebras's future business opportunities will be fundamentally weakened. Cerebras Still Expected to Benefit from AI Inference Growth Meeks believes AI computing demand may continue to exceed industry supply capacity in the coming years, which will provide growth space for multiple chip suppliers, including Cerebras. He said that even if OpenAI chooses NVIDIA Corporation GPUs for some applications, Cerebras is still expected to continue conducting substantial business cooperation with OpenAI. At the same time, Cerebras also has other important partners, including AMD (AMD.US) and Amazon.com, Inc.'s (AMZN.US) cloud computing platform AWS. He noted that these partnerships are expected to help Cerebras expand market opportunities and reduce investor concerns about a single customer's technology choices. More importantly, as the AI industry's computing demand gradually expands from model training to inference tasks in practical applications, inference speed and operational efficiency are becoming important competitive factors in AI infrastructure. Unlike the training phase, which mainly focuses on model parameter optimization, the inference phase needs to continuously generate output based on user requests. With the widespread adoption of AI applications, how to improve response speed while controlling costs is becoming an important issue that chipmakers need to address. Meeks noted that Cerebras has a competitive solution in high-speed inference and is expected to benefit from this trend. AI Computing Architecture May Move Toward Division of Labor and Cooperation, Cerebras and NVIDIA Corporation Have Room for Complementarity In addition to directly competing for AI chip orders, Meeks also sees the possibility of Cerebras and NVIDIA Corporation forming a complementary relationship in future AI computing architectures. Meeks noted that as AI inference tasks become more complex, the industry may gradually adopt a so-called "computing resource disaggregation" architecture, using different types of chips for different computing stages to improve overall efficiency. Large language model inference typically involves two main stages, Prefill and Decode. The Prefill stage is mainly responsible for processing user input prompts and contextual information, usually requiring strong parallel computing capability; the Decode stage is responsible for gradually generating output content, with different requirements for computing latency and data processing efficiency. Meeks believes that in the future, a computing architecture may emerge in which NVIDIA Corporation GPUs handle Prefill tasks while Cerebras's WSE handles Decode tasks. This division-of-labor model means that although Cerebras and NVIDIA Corporation are in direct competition in the AI chip market, their technologies may also complement each other in specific application scenarios. Meeks even believes that it is not impossible for Cerebras and its main competitor NVIDIA Corporation to leverage their respective advantages in such architectures in the future. However, this remains an analyst's judgment on the future direction of technological development and does not mean the two companies have announced related cooperation. Overall, Meeks believes that Cerebras's recent stock selloff due to OpenAI's adoption of NVIDIA Corporation GPUs has not changed the long-term demand growth trend in the AI inference market. As the scale of AI applications continues to expand, Cerebras's wafer-scale chips are still expected to gain more business opportunities in the high-speed inference field. Based on the above judgment, Meeks upgraded Cerebras from "Hold" to "Buy" with a price target of $209.