Cerebras Systems went public on May 13, 2026, priced at $185 per share, and for roughly the first 24 hours the debut looked like a triumph. Shares jumped 68% on the opening day and closed near $311. The momentum did not last.

By August 2026, the stock had fallen more than 50% from its highs, at times dipping below the IPO price. The company is now leaning heavily into its next-generation wafer-scale chip as the answer to its weakened share price, arguing that its specialized silicon will outpace traditional GPUs as AI agents become the dominant computing workload. The slide has turned Cerebras into an early public test of whether purpose-built AI silicon can hold investor confidence once quarterly results, rather than technical claims, become the measure.

A Chip Measuring 46,225 Square Millimeters

Cerebras' Wafer-Scale Engine 3, known as WSE-3, measures 46,225 square millimeters. The size is the point of the "wafer-scale" name: rather than dicing a silicon wafer into many smaller chips, the design builds a single processor across most of the wafer's surface. It packs 4 trillion transistors and delivers 125 petaflops of AI computing power. The company claims the chip offers thousands of times greater memory bandwidth than leading GPUs, translating into 10x to 15x faster token generation for AI inference tasks.

That bandwidth figure carries the strategic weight. Generating each token requires repeatedly moving model data through memory, so bandwidth, not raw compute alone, sets the pace of AI responses. Cerebras is betting that as AI agents become the dominant computing workload, that advantage compounds, because agents must complete many inference steps to finish a single task.

Founded in 2015 by Andrew Feldman, Cerebras employs roughly 708 employees and focuses exclusively on AI training and inference hardware. It has secured partnerships with OpenAI and AWS to deploy its inference capabilities.

The Stock Tells a Different Story

Cerebras raised approximately $5.55 billion through its IPO, earning an initial valuation near $56 billion. The company raised its 2026 revenue guidance to between $880 million and $890 million. Set against the near-$56 billion initial valuation, that guidance works out to more than 60 times annual revenue, a multiple that leaves little room for execution missteps.

On August 12, 2026, shares fell 16% in extended trading despite the raised revenue targets. Cerebras' hardware revenue has shifted unpredictably, and operational challenges following the public listing have drawn scrutiny from analysts watching profit margins. A stock that falls on raised guidance is a market pricing questions about the quality of revenue, not its size.

An Architectural Argument Against GPUs

Cerebras' case is architectural: traditional GPUs were designed for graphics rendering and later adapted for AI. The memory bandwidth advantage is real and measurable, and for specific inference workloads the performance gap could justify the switching costs for large cloud providers.

The partnerships with OpenAI and AWS suggest at least some major buyers agree. But partnerships and purchase orders are different things, and the market will ultimately judge Cerebras on whether those collaborations translate into recurring, growing revenue rather than one-time deployments. For a hardware company whose revenue has already shifted unpredictably, the gap between a one-time deployment and repeat inference business is the gap between erratic quarters and a dependable base.

What Comes Next

For investors who bought in at the IPO price, the stock would need to more than double from its August lows just to revisit its first-day highs. Revenue guidance of $880 million to $890 million is meaningful for a company of Cerebras' size, but it would need to grow substantially to justify a valuation that was approaching $56 billion. The checkpoints come straight from the company's own disclosures: whether the raised guidance is met, whether margins steady as post-listing operational challenges are absorbed, and whether the OpenAI and AWS collaborations turn into repeat orders.