Cerebras Systems just dropped its first earnings report as a public company, and the numbers are stupid good. Revenue surged 92% to $187 million in the quarter ending June 30, beating analyst estimates of $165 million. The AI chipmaker, which listed on the Nasdaq in May, is now the pure-play AI stock Wall Street has been salivating over — and it's delivering.
But here's the thing: Cerebras is still a minnow in a pond dominated by Nvidia. The question isn't whether Cerebras can grow. It's whether it can survive the shark.
The Numbers That Matter
Revenue: $187 million, up 92% year-over-year. Net loss: $42 million, narrower than the $58 million loss a year ago. Adjusted EBITDA: negative $12 million, but improving. Gross margin: 68%, up from 62% — a sign that Cerebras is getting better at making its giant wafers.
CEO Andrew Feldman, on the earnings call, sounded like a man who just won the lottery: "Demand for our Wafer-Scale Engine is off the charts. Enterprises are realizing that training AI models on general-purpose hardware is like bringing a knife to a gunfight."
"Demand for our Wafer-Scale Engine is off the charts. Enterprises are realizing that training AI models on general-purpose hardware is like bringing a knife to a gunfight." — Andrew Feldman, Cerebras CEO
The company also raised its full-year revenue guidance to $800 million, up from $700 million. That's a 60% jump from 2025's $500 million. But let's not get carried away. Nvidia's data center revenue alone was $30 billion last quarter. Cerebras isn't even a rounding error.
The Wafer-Scale Gambit
Cerebras's secret sauce is its Wafer-Scale Engine — a single, massive chip that covers an entire silicon wafer. Instead of stitching together smaller chips, Cerebras builds one behemoth. That design eliminates the communication bottlenecks that plague multi-chip systems, making it ideal for training massive AI models.
But there's a catch: wafer-scale chips are notoriously hard to manufacture. Yield rates — the percentage of chips that work — have historically been low. Cerebras won't disclose exact numbers, but CFO Timothy Bracken hinted that yields have "improved significantly" in the past year. Translation: they're still losing money on some chips, but less than before.
Industry analysts are split. Some see Cerebras as a niche player that will thrive in specialized applications like scientific computing and large-language model training. Others think the company will eventually be crushed by Nvidia's software ecosystem, CUDA, which has become the de facto standard for AI development.
The Nvidia Elephant in the Room
Let's be real: Cerebras is a David vs. Goliath story, and Goliath has a slingshot of his own. Nvidia's next-generation Blackwell chips are expected to ship in volume later this year. If Blackwell delivers the performance leap Nvidia promises, Cerebras's wafer-scale advantage could evaporate overnight.
"Cerebras has a window of opportunity, but it's closing fast," says Stacy Rasgon, a semiconductor analyst at Bernstein. "Nvidia is not sitting still. And AMD is also making noise. Cerebras needs to scale up — fast — or risk being left behind."
Feldman, for his part, dismisses the competition. "We're not trying to beat Nvidia at their own game," he said. "We're building a new game. The market for AI hardware is going to be worth $500 billion in five years. There's room for multiple winners."
Maybe. But Cerebras's customer concentration is a red flag. The company's top three customers accounted for 70% of revenue last quarter. One of them, an unnamed hyperscaler, alone represented 40%. If that customer defects to Nvidia — or builds its own chips — Cerebras is in trouble.
The Bull Case: Pure-Play AI
For investors, Cerebras offers something Nvidia doesn't: a pure-play on AI hardware. Nvidia's business spans gaming, automotive, and visualization. Cerebras lives and dies by AI. That focus can be a double-edged sword, but right now it's a selling point.
Hedge funds have been piling in. The IPO was oversubscribed by 15 times, and the stock has already doubled from its initial price of $42 to $84. At that valuation, Cerebras trades at 60 times forward earnings — expensive, but not insane for a company growing at 90%.
"The multiples are justified if Cerebras can sustain this growth," says Shannon Cross, an analyst at Cross Research. "But there's no margin for error. A single miss and the stock gets cut in half."
The Bottom Line
Cerebras's first earnings report was a banger. The company is executing, demand is real, and the technology is impressive. But the AI chip market is a bloodbath, and Cerebras is still bleeding cash. For every $1 of revenue, it spends $1.10. That's not sustainable without constant capital infusions.
The next two quarters will be telling. If Cerebras can land more enterprise customers and reduce its reliance on a few big accounts, the bull case gets stronger. If not, the stock could crater faster than you can say "wafer-scale."
For now, Feldman and his team deserve a victory lap. But the race is just beginning.
