Cathie Wood’s Cerebras Systems Investment Signals a Quiet War Against Nvidia

If you’ve been following Ark’s daily disclosures lately, it’s difficult to miss the pattern. Cathie Wood’s funds receive a new tranche of Cerebras shares almost every other trading day. 105,616 shares on May 14. 149,716 more on May 15. 82,779 more on May 20. 42,126 more on May 22. At some point, the running total began to resemble a conviction rather than a position.

On May 14, Cerebras Systems made the kind of debut that most CFOs can only imagine. The stock opened at $350 and closed its first day at $311.07, with a price of $185 following an initial range of $115 to $125. According to reports, demand exceeded supply by more than twenty times, making it the biggest U.S. tech IPO of 2026 to date. Wood didn’t wait for the dust to settle. On the first day, she bought in and continued to do so.

This has a very Cathie quality to it. She has previously worked with Palantir, Coinbase, and Tesla. The trade of convictions. Purchase loudly, buy heavily, and allow the thesis to be demonstrated in public. Additionally, she is combining it with a consistent, nearly methodical reduction in her AMD position. There is no subtlety to the rotation. It seems to be a declaration about the future direction she believes AI hardware will take.

To put it mildly, Cerebras is not your typical chip manufacturer. Cerebras took a completely different approach, whereas Nvidia and AMD have spent years optimizing GPUs measured in square centimeters. Its Wafer-Scale Engine 3 packs 900,000 AI cores and 4 trillion transistors onto a single silicon wafer, which is about the size of a dinner plate. To prevent the device from literally melting through motherboards, engineers had to redesign cooling systems, packaging materials, and even motherboards. Someone at the company decided that smaller wasn’t always better, which is why it’s the biggest commercial chip ever produced.

The current state of the AI industry makes the pitch compelling. Nvidia’s empire was built through training. The next ten years of competition will take place in inference, the unglamorous task of actually running these models millions of times every day. According to Cerebras, its chips can perform inference up to fifteen to twenty-one times more quickly than Blackwell systems from Nvidia. Since the company provided those figures, a healthy dose of skepticism is justified. However, it is more difficult to ignore the customer list. In January, OpenAI committed to deploying 750 megawatts of Cerebras capacity through 2028 as part of a multi-year, $20 billion compute agreement. AWS became the first hyperscaler to install Cerebras systems in its data centers when Amazon did the same in March.

More quickly than most anticipated, the financials have caught up. In 2025, revenue increased from $290 million to approximately $510 million, a 76% increase. The company’s net profit increased from $482 million to $238 million, but a significant portion of that came from non-operating items rather than its main business. The trajectory appears real, though. And for Wood, whose ARK Innovation ETF has been looking for the next Tesla-like tale for years, the combination of improving margins, hyperscaler validation, and architectural differentiation seems to check most of the boxes.

However, it’s important to remember that there are risks associated with purchasing this aggressively in a recently listed stock that aren’t mentioned in pitch decks. The market capitalization is already close to $65 billion. Revenue concentration is high; according to reports, one customer accounted for about 24% of revenue in 2025. Additionally, when a prominent manager loudly supports this, prices may fluctuate in ways unrelated to what is truly being developed within the organization. It’s difficult to avoid experiencing a tiny bit of déjà vu while watching this play out. When it comes to large wagers, Wood has previously been correct. She has also made incredibly incorrect statements, sometimes about the same businesses in different years.

She is obviously not hedging. Somewhere in Sunnyvale, engineers are producing dinner plate-sized chips for consumers who seem to be unable to get enough of them, while Cerebras’ position continues to grow and AMD’s continues to decline. The upcoming quarters should begin to reveal whether that proves to be another cautionary tale about IPO enthusiasm or the next big AI infrastructure story.