OpenAI CEO Sam Altman has publicly confirmed that OpenAI is working hand-in-hand with Cerebras—a company renowned for its wafer-scale AI chips—to delve into the frontiers of speed optimization. Previously, amidst rumors that OpenAI might pivot to NVIDIA chips, Cerebras’ stock price took a nosedive. However, the announcement of this collaboration sparked an immediate rebound in its stock value. Currently, the spotlight in AI computing power consumption is shifting from the training phase to inference applications, posing a pivotal choice between specialized large chips and general-purpose computing paths.

Cerebras has witnessed remarkable revenue growth in recent years, with its revenue nearing $510 million in fiscal year 2025. Notably, 32% of its total revenue in the second quarter stemmed from OpenAI. Cerebras’ groundbreaking Wafer Scale Engine architecture seamlessly integrates an entire silicon wafer into a single, massive chip. This design drastically cuts down on data transmission distances, providing a significant edge in throughput speed for large language model inference—a benefit that has been successfully translated into commercial gains.

Nevertheless, Cerebras is not without its challenges. It grapples with a high forward price-to-earnings ratio, substantial net outflows of free cash flow, and sluggish progress in recognizing revenue from its $25.4 billion order backlog. Only 22% of this backlog is expected to be converted into revenue within the next 24 months. Furthermore, its customer base is highly concentrated, with the top three clients accounting for a staggering 76% of its second-quarter revenue.

To take on general-purpose chip behemoths, specialized inference chips must achieve fundamental architectural breakthroughs and create a performance gap of at least an order of magnitude. The Cerebras team has risen to the occasion, overcoming numerous engineering hurdles through a disruptive full-wafer integration strategy and optimization of underlying algebraic operations. Looking ahead, the growth trajectory of its cloud service revenue and strides in customer diversification will serve as crucial barometers for assessing the viability of the specialized inference chip model.