The AI chip manufacturer Cerebras, as it pursues a potential initial public offering, appears to have secured a significant cloud computing client: Oracle. During an analyst call following Oracle's quarterly earnings report on Tuesday, one of the software provider's co-CEOs, Clay Magouyrk, stated that the company's infrastructure incorporates chips from Cerebras, in addition to GPUs from market leader Nvidia and competitor AMD.

Magouyrk commented, "The infrastructure we have built is flexible and versatile, capable of supporting workloads ranging from the smallest to the largest. We continuously provide the latest accelerators, from the newest products by Nvidia and AMD to emerging designs from companies like Cerebras and Positron, another AI hardware startup."

Cerebras offers cloud services utilizing its large-scale WSE-3 chip. The company had filed for an IPO in 2024 but withdrew the application last October. Shortly after, it announced securing $1.1 billion in funding, reaching a valuation of $8.1 billion. Its CEO, Andrew Feldman, indicated that Cerebras still plans to go public.

For potential investors, one of the most notable concerns in Cerebras's original prospectus was its heavy reliance on a single Middle Eastern customer. G42, which is backed by Microsoft and headquartered in Abu Dhabi, UAE, accounted for 87% of Cerebras's revenue in the first half of 2024.

Expanding its partnership roster with a name like Oracle could provide a substantial boost for Cerebras. This development follows a major announcement earlier this year. In January, Cerebras stated it had secured $10 billion in commitments from OpenAI and other firms. OpenAI itself relies on cloud services from providers such as Oracle. The following month, OpenAI announced it was collaborating with Cerebras to launch a research preview of Codex-Spark for ChatGPT Pro customers, a fast-response AI model designed specifically for software development.

Oracle's earnings call occurred after the company reported better-than-expected results. It raised its guidance for fiscal year 2027 and noted that its remaining performance obligations had more than tripled from the previous year's level to $553 billion. After mentioning Cerebras and other chipmakers, Magouyrk concluded, "All in all, we are confident that our current investments in data centers, computing power, and customer relationships will only become more valuable over time."

Despite Cerebras attempting to compete as a newcomer against the world's most valuable company, the market it operates in exhibits an almost insatiable demand for computing power, driven by AI model developers scaling up rapidly to meet user needs. Nvidia is leveraging its substantial cash reserves to expand into new product areas. Last December, the company acquired core assets of AI chip startup Groq for approximately $20 billion. Nvidia plans to introduce a new architecture, incorporating technology from Groq, at the upcoming GTC developer conference in California next week.

Magouyrk noted on the call that there would be some "key announcements" at GTC. He also emphasized that the speed of responding to requests depends not only on strategically located data centers but also on innovative technology. "It depends on the type of hardware deployed, which is why we see so much innovation happening around these AI accelerators," he said. "If you look at what Groq, Cerebras, or Positron are doing, all these different types of customers are asking: How do we not only reduce inference costs but also significantly cut inference latency?"