Cerebras Files for IPO to Scale Wafer-Scale AI Chips Amid Deals with OpenAI and AWS

Companies don’t usually announce themselves by changing the physics of their category. Cerebras did, and then kept building like that wasn’t enough. Back in 2016, Andrew Feldman (CEO) and a crew forged in the SeaMicro days decided that scaling AI wasn’t about stacking more GPUs like pancakes. It was about going full wafer, one massive piece of compute that doesn’t blink when the model size starts acting disrespectful. Cerebras Systems came out of Sunnyvale with that thesis and didn’t ask for permission.

Fast forward today and the plot tightens. Cerebras files for its IPO, ticker CBRS, after a detour that would’ve rattled a lesser outfit. Regulatory reviews, a pulled filing, and then right back to the table like nothing happened. That’s not luck. That’s conviction with a memory.

The numbers tell their own story if you listen closely. A $1.1B raise in 2025 followed by another $1.0B in early 2026. Multi billion momentum before even ringing the bell. Partnerships that read like a power list. Amazon Web Services bringing Cerebras into Bedrock, threading their wafer scale muscle into the cloud. OpenAI locking in a massive multiyear relationship that signals one thing clearly. When the models get bigger, the old playbook starts sweating.

And the machine itself. The Wafer Scale Engine is not a tweak. It’s a statement. One chip the size of a dinner plate, built to keep data close and latency on a leash. Training and inference stop being a tradeoff and start acting like teammates. That matters when every millisecond has a price tag and every token has expectations.

Leadership here is not theoretical. Andrew Feldman (CEO) still driving, with Jean Philippe Fricker (Chief System Architect), Michael James (Chief Architect, Software), and Sean Lie (Chief Hardware Architect) engineering the guts of the system like it’s a live performance, not a lab experiment. Different disciplines, same rhythm.

The market didn’t reward Cerebras for being early. It rewarded them for being right long enough to survive being early. They built for a future that wasn’t convenient yet, then waited while the world caught up and started begging for compute like it was oxygen.

Now the window is open. Public markets, hyperscaler distribution, and demand that doesn’t look like it’s cooling off anytime soon. Cerebras isn’t just entering the chat. It’s bringing its own infrastructure to the conversation, and suddenly everyone else has to speak a little louder to be heard.