Cerebras Systems, the chip company that has spent years positioning itself as the challenger to Nvidia, reported quarterly results on Wednesday evening that showed its business moving in two directions at once. Hardware sales fell 23% from a year earlier, to $54.1 million, while revenue from its AI cloud business nearly tripled, to $126 million.
The contrast underscores how the company’s strategy has shifted since its initial public offering in May. Cerebras built its reputation on wafer-scale chips, processors the size of dinner plates that it says can run AI models faster and more cheaply than Nvidia’s flagship products. But its fastest-growing business is now renting out computing power rather than selling the machines themselves, a model that puts it in competition with the cloud giants whose data centers it hoped to supply.
Total revenue for the June quarter was $180 million, short of the $194 million analysts had expected. The stock fell about 14% in after-hours trading before recovering somewhat, extending a volatile ride for a company that had already seen its shares swing widely since listing. Despite the drop, the stock remains up roughly 42% since the IPO.
The earnings call offered a picture of a company in transition. Executives stressed the growth in the cloud segment, which they described as the clearest evidence that customers are using Cerebras technology for real workloads. The hardware decline was framed as a timing issue: customers who previously bought systems outright are increasingly signing up for cloud capacity instead, a shift that lowers near-term revenue but builds a recurring base.
The cloud business, branded Cerebras Inference, has found a niche in fast inference, the process of running trained AI models to answer questions. The company says its chips can serve responses faster than GPU-based systems, a claim that has won it customers running large language models, coding assistants, and image generators. Revenue in that segment more than doubled sequentially in the quarter, according to the company.
Analysts were split on what the results mean. The bull case is that the cloud growth proves the technology works and that recurring revenue will smooth out the lumpiness of hardware sales. The bear case is that a company that went public as a chip maker is effectively becoming a small cloud provider, a far more crowded and capital-hungry business, and that $180 million in quarterly revenue against the scale of Nvidia’s operations leaves it a niche player.
The financials are complicated by a gap between the company’s GAAP results and the measures it presents as core results. GAAP hardware revenue of $54.1 million was held down by items the company said reflected the way it accounts for certain arrangements, including a system rental tied to customer demand that it will fulfill once dedicated data-center capacity is available. On the company’s core basis, hardware revenue was higher, and management argued that the core view better reflects the underlying business.
The company’s customers range from AI labs running large language models to enterprises using the cloud for tasks such as document processing and code generation. Cerebras says its chips are particularly well suited to models that must answer questions quickly, and it has published benchmark results showing response times faster than systems built on Nvidia’s GPUs. Those claims have helped it win cloud business even as the hardware division sells fewer complete systems.
Whatever the accounting, the market reaction was clear: investors wanted faster growth in the hardware line, the business that justifies the company’s valuation, and instead saw the segment that competes with hyperscalers grow while the flagship product line shrank.
The IPO, completed in May, was one of the year’s most anticipated tech listings, drawing demand from investors who wanted an alternative to Nvidia’s dominance in AI chips. The stock jumped on its first day of trading before giving back much of the gain, and it has remained volatile since. Cerebras has used the proceeds to expand the data centers that power its cloud business, betting that owning its own capacity will let it control costs and performance in a way that renting from others would not. Cerebras said it expects the hardware business to recover as new capacity comes online and as customers who started with cloud trials move to larger commitments.
The company’s position remains unusual. It has a differentiated chip design, a public profile built on beating Nvidia in benchmark comparisons, and a customer list that includes some of the biggest names in AI. It also faces the structural problem that has defeated previous Nvidia challengers: software ecosystems, developer tools, and customer habits all favor the incumbent, and winning share requires more than a faster chip.
For investors, the takeaway from the quarter is that Cerebras is becoming what its revenue mix says it is. The question, analysts said, is whether the market will reward it as a growth story in AI computing services or penalize it for failing to sell the hardware that made it famous. The next few quarters, as new capacity comes online and hardware sales either recover or keep sliding, will supply the answer.
The strategic question is whether a small cloud can compete with the giants. Amazon, Microsoft, and Google offer AI computing at enormous scale with matching software ecosystems, and Cerebras’s cloud business remains a small fraction of their size. Management argues that specialization is the advantage: a company that controls both the chip and the service can offer performance and economics that the hyperscalers cannot easily match. Investors will find out in the coming quarters whether that argument holds.
