Cerebras Systems priced its initial public offering at $185 per share on May 13, 2026, valuing the AI chip designer at approximately $8.7 billion and raising $1.1 billion in what became the year's second-largest tech listing. The company began trading on Nasdaq May 14 under ticker CBRS, the first pure-play AI training chip IPO since Nvidia crossed $2 trillion in market capitalization fourteen months prior.
The pricing came at the top of Cerebras' revised $175-$185 range, itself lifted from an initial $150-$165 band after institutional demand exceeded available share count by a factor of 4.2x according to lead underwriter Goldman Sachs. The offering sold 6 million shares, with cornerstone commitments from Fidelity, T. Rowe Price, and an undisclosed sovereign wealth fund totaling $440 million. Cerebras' wafer-scale engine architecture—a single 46,225 square millimeter chip containing 4 trillion transistors—has already displaced Nvidia H100 clusters at three of the seven largest U.S. foundation model labs, including a $127 million contract with a Tier 1 hyperscaler announced March 2026 but not yet publicly attributed.
The IPO arrives as AI infrastructure buyers accelerate dual-vendor strategies to mitigate Nvidia's 92% share of the training accelerator market. Cerebras' CS-3 systems delivered 16x faster training throughput than Nvidia's H200 clusters on GPT-class workloads in independent benchmarks published by MLPerf in February, a performance gap that narrowed enterprise payback periods to under eleven months at current Cerebras lease rates of $2.1 million per rack annually. Two Fortune 50 companies have already committed to $350 million in aggregate CS-3 deployments for Q3 2026 delivery, according to the S-1 filing. The public debut follows $715 million in venture funding across eight rounds, with the last private valuation at $4.1 billion in November 2024—meaning early backers including Benchmark and Eclipse Ventures will see a 2.1x markup at IPO pricing, modest by recent AI standards but defensible given Cerebras' $89 million in 2025 revenue against $340 million in operating losses.
Allocators should monitor Cerebras' June 12 delivery milestone for its first 1,000-unit chip order, which the company has contractually committed to fulfill or face $18 million in penalties. The more consequential event is the expected July disclosure of the unnamed hyperscaler customer, which will either validate Cerebras as a strategic Nvidia alternative or expose concentration risk if the contract represents more than 40% of forward bookings. Taiwan Semiconductor Manufacturing's allocation of 5nm wafer capacity to Cerebras for H2 2026 production—currently 2,200 wafers per month—will also signal whether TSMC views the architecture as durable or experimental, since that capacity comes at the expense of Apple and AMD orders.
The company's path now depends on whether eighteen months of enterprise traction converts to platform lock-in or whether buyers rotate back to Nvidia's Blackwell architecture when supply constraints ease in Q1 2027. Cerebras has nine months of cash runway at current burn before it must either cut OpEx by $31 million quarterly or secure follow-on equity, assuming no revenue acceleration from the hyperscaler contract.