Oracle Just Put Cerebras on the AI Chip Map

Oracle's CEO named Cerebras alongside Nvidia and AMD on its earnings call. For a startup that nearly botched its IPO over a single-customer problem, this could change everything.

Ninety days ago, Cerebras was a startup with an IPO problem. Today, it's the chip that Oracle's CEO name-dropped on an earnings call—right next to Nvidia and AMD.

What Actually Happened

On Tuesday, following Oracle's better-than-expected quarterly results, co-CEO Clay Magouyrk confirmed that Oracle's cloud infrastructure now includes Cerebras hardware. "We continually offer the latest in accelerators, from the most recent Nvidia and AMD options to emerging designs from companies like Cerebras and Positron," he said.

That's not a throwaway line. On earnings calls, every word is deliberate. Naming Cerebras alongside the two dominant GPU makers—in front of analysts and investors—is as close to a public endorsement as it gets without a formal press release.

Oracle itself had a strong quarter to report: remaining performance obligations (RPO), a forward-looking indicator of contracted revenue, surged to $553 billion, more than quadrupling year-over-year. The AI infrastructure build-out is not slowing down, and Oracle is positioning itself at the center of it.

Why Cerebras Needed This—Badly

To understand why this matters, you need to know what almost killed Cerebras' IPO.

In 2024, the company filed to go public. What investors found in the prospectus was uncomfortable: in the first half of 2024, a single customer—G42, a Microsoft-backed firm headquartered in Abu Dhabi—accounted for 87% of revenue. That's not customer concentration; that's customer dependency. The IPO was pulled in October 2024.

Within days, Cerebras announced a $1.1 billion funding round at an $8.1 billion valuation. CEO Andrew Feldman insisted the company still intended to go public. But the fundamental problem remained: the customer roster needed to diversify, fast.

Advertise with Us

[email protected]

Since then, the momentum has shifted. In January, OpenAI committed $10 billion in cloud services, with Cerebras as part of the mix. In February, OpenAI announced that Codex-Spark—its AI model for software development, available to ChatGPT Pro users—runs on Cerebras chips. And now Oracle.

The Actual Tech Case for Cerebras

This isn't just a business development story. There's a real technical argument for why companies like Oracle and OpenAI are looking beyond Nvidia.

Cerebras' WSE-3 is, quite literally, the largest chip ever made—it spans an entire silicon wafer. That unusual architecture is purpose-built for one thing: inference speed. When an AI model has already been trained and is responding to live user requests, the bottleneck is often latency—how fast the chip can process and return a result.

Magouyrk made this explicit on the call: "Not only how do we reduce the cost of inferencing, but also, how can we significantly reduce the latency of it?" For a coding assistant like Codex-Spark, where a developer is waiting for a real-time suggestion, milliseconds matter. Nvidia's GPUs are unmatched for training large models, but the inference market is a different competitive landscape.

	Nvidia H100/B200	Cerebras WSE-3
Primary strength	Training + general inference	Ultra-low latency inference
Chip scale	Standard die	Full wafer (world's largest)
Market position	Dominant, ~80% AI chip share	Emerging challenger
Key customers	Virtually all major AI labs	OpenAI, Oracle, G42
IPO status	Public (NVDA)	Targeting IPO (date TBD)

What This Means for Investors

For anyone tracking Cerebras' eventual IPO, the customer diversification story is the single most important variable. The 87% single-customer concentration was the red flag that tanked the original filing. Three significant new customer relationships in under six months—OpenAI, Oracle, and the ongoing G42 relationship—reframe that narrative meaningfully.

But caution is warranted. Oracle has not updated its public price list to include a Cerebras option, and neither company confirmed the relationship on the record. Magouyrk's comment on the call is suggestive, not definitive. The actual revenue contribution from Oracle is unknown.

For those already holding Nvidia stock, the more relevant question isn't whether Cerebras threatens Nvidia's dominance—it doesn't, not yet. It's whether the inference market develops into a structurally distinct segment where multiple chip architectures coexist. Nvidia clearly sees the threat: in December, it acquired key assets from AI chip startup Groq for approximately $20 billion, and plans to unveil a new architecture based on that acquisition at its GTC conference next week.

The AI chip market is no longer a one-horse race. It's becoming a question of which horse runs which course.