# Cerebras WSE-4 and 2026 IPO

> Source: https://aiwiki.ai/wiki/cerebras_wse_4
> Updated: 2026-07-17
> Categories: AI Hardware
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution to "AI Wiki (aiwiki.ai)".

The **Cerebras WSE-4** (Wafer Scale Engine 4) is the anticipated fourth generation of the dinner-plate-sized [wafer-scale](/wiki/wafer_scale_engine) AI processor built by [Cerebras](/wiki/cerebras) Systems. As of June 2026 the WSE-4 has not been officially announced or specified, and the third-generation [WSE-3](/wiki/cerebras_wse_3) remains the company's shipping chip. The label is most often paired with the larger 2026 milestone for Cerebras: the company's long-delayed initial public offering, which priced on May 13, 2026 and began trading on the [Nasdaq](/wiki/nasdaq) the next day under the ticker CBRS.[1][2] The offering sold 30 million shares at $185 each and raised about $5.55 billion, which CNBC described as the largest US technology IPO since Uber in 2019.[2][6] It valued Cerebras at roughly $56 billion at the offer price, before a first-day surge pushed the market value much higher. The IPO had been stalled for more than a year and a half, largely by a national-security review of a minority investment from the Emirati group [G42](/wiki/g42).[13]

## Overview

Cerebras Systems, founded in 2015 and led by chief executive [Andrew Feldman](/wiki/andrew_feldman), designs [AI chips](/wiki/ai_chip) that span almost an entire [semiconductor](/wiki/semiconductor) wafer rather than being cut into many small dies. Where companies such as [Nvidia](/wiki/nvidia) slice a 300 mm (12-inch) wafer into hundreds of separate GPUs, Cerebras keeps the wafer whole and operates it as a single processor. The result is a chip with vastly more cores and on-chip memory than a conventional GPU, which Cerebras argues makes it especially well suited to running large neural networks at high speed. Each chip is paired with a system, the CS series, that supplies power, cooling, and data movement.

The current product is the WSE-3, announced on March 13, 2024 and fabricated by [TSMC](/wiki/tsmc) on a 5 nm process.[8] The 2026 IPO, the OpenAI inference agreement, and a still-heavy reliance on G42 are the developments that have brought the company the most attention. This article covers the prospective WSE-4 and the 2026 listing; the broader corporate history and the WSE-3 chip itself are treated on the [Cerebras](/wiki/cerebras) and [WSE-3](/wiki/cerebras_wse_3) pages.

## WSE-4 specs vs WSE-3

No official WSE-4 specifications exist as of June 2026. Cerebras has not published transistor counts, core counts, a process node, or a launch date for a fourth-generation part, and the WSE-3 remains the latest announced design. Trade and analyst commentary published around the IPO generally expects a WSE-4 to arrive in roughly 2027, most likely on a TSMC 3 nm class process, continuing the company's two-to-three-year generational cadence (WSE-1 in 2019, WSE-2 in 2021, WSE-3 in 2024). Several outlets noted that a portion of the IPO proceeds is expected to fund next-generation research and development. None of those WSE-4 figures should be treated as confirmed.

What is confirmed is the WSE-3 baseline that a WSE-4 would have to surpass. The WSE-3 carries 4 trillion transistors and 900,000 AI-optimized cores on a die of 46,225 mm squared, about 57 times the area of an Nvidia H100 (roughly 814 mm squared).[8][10] It holds 44 GB of on-chip SRAM with about 21 petabytes per second of memory bandwidth, and Cerebras rates its peak AI performance at 125 petaflops at 16-bit precision (12.5 petaflops dense).[8][11] The company says the WSE-3 doubles the performance of the WSE-2 at the same power and price.[8]

The published Cerebras generations compare as follows:

| Generation | Year | TSMC process | Transistors | AI cores | On-chip SRAM |
|---|---|---|---|---|---|
| WSE-1 | 2019 | 16 nm | 1.2 trillion | 400,000 | 18 GB |
| WSE-2 | 2021 | 7 nm | 2.6 trillion | 850,000 | 40 GB |
| WSE-3 | 2024 | 5 nm | 4 trillion | 900,000 | 44 GB[8] |
| WSE-4 | Not announced (expected ~2027) | reported as likely 3 nm | Not disclosed | Not disclosed | Not disclosed |

## The CS-4 system

There is likewise no announced CS-4 system. A CS-4 would, by the company's naming convention, be the chassis built around a future WSE-4, but Cerebras has not detailed one. The current platform is the CS-3, which houses a single WSE-3. Cerebras pairs the CS-3 with an external memory subsystem it calls MemoryX, offered in capacities reported up to roughly 1.2 petabytes, and a fabric, SwarmX, that lets many systems be linked.[9] The company says a single CS-3 can train models with up to 24 trillion parameters, and that a cluster of up to 2,048 CS-3 systems reaches about 256 exaflops at 16-bit precision.[8] Until a WSE-4 is unveiled, the CS-3 remains the basis of both Cerebras' on-premises sales and its inference cloud.

## The 2026 IPO

Cerebras first moved toward the public market in 2024, filing a confidential registration with the US Securities and Exchange Commission in September 2024 and targeting a listing as early as that October.[13] The plan stalled almost immediately. G42, an Abu Dhabi AI conglomerate, had agreed to buy about $335 million of Cerebras stock, a stake of more than 5 percent, and that purchase triggered a review by the Committee on Foreign Investment in the United States ([CFIUS](/wiki/cfius)).[13] The concern centered on G42's past ties to Chinese firms, including Huawei. The companies restructured the deal so that G42's shares would be non-voting, and CFIUS cleared the investment in early 2025, but the offering did not proceed amid a soft IPO market.[12]

In October 2025 Cerebras formally withdrew its registration and raised private bridge capital instead, including a round reported at about $1.1 billion.[12] It refiled confidentially in early 2026 and made its prospectus public on April 17, 2026.[14] Demand proved strong: Cerebras twice raised its expected price, from an initial $115 to $125 range to $150 to $160, and finally priced at $185 per share on May 13, 2026, above the range.[2] The sale of 30 million shares brought in about $5.55 billion and implied a fully diluted valuation near $56.4 billion.[2][6]

Trading opened the next day, May 14, 2026. The stock began at about $350, roughly 89 percent above the offer price, and closed its first session at $311.07, up about 68 percent.[1] At that close CNBC put the company's market capitalization near $95 billion on a fully diluted basis, while The Register and some other outlets, counting basic shares, cited a figure closer to $66 billion.[1][7] The debut made paper billionaires of Feldman, whose stake was valued near $1.9 billion, and co-founder and chief technology officer Sean Lie, at roughly $1 billion.[3] The shares gave back part of the gain in the days that followed.[4]

The prospectus also clarified the company's finances. Revenue reached $510 million in 2025, up about 76 percent year over year.[14] Cerebras reported GAAP net income of $237.8 million, but that figure was lifted by a one-time, non-cash gain of about $363.3 million tied to the extinguishment of a forward-contract liability related to G42; on an operating basis the company still ran a loss of roughly $146 million for the year.[14] The business is therefore growing quickly but is not yet profitable from operations.

## Inference business and the G42 and OpenAI context

Cerebras has increasingly positioned itself around AI [inference](/wiki/inference), the work of running already-trained models to answer queries, rather than only model training. It launched a hosted Cerebras Inference service in 2024, later listed it on the AWS Marketplace, and markets the speed advantage of keeping an entire [large language model](/wiki/large_language_model) on one chip's SRAM. Independent benchmarks by Artificial Analysis cited in 2026 reported the CS-3 running OpenAI's gpt-oss-120B at more than 2,700 tokens per second versus about 900 on an Nvidia B200,[15] and in May 2026 Cerebras reported 981 tokens per second on the Kimi K2.6 model, which it said was about 6.7 times the next-fastest GPU cloud.[16] Cerebras' own materials claim the CS-3 is roughly 21 times faster than an Nvidia DGX B200 at one-third the cost and power for some inference workloads; these are vendor figures and should be read as such.[15]

Customer concentration remains the company's central risk. In 2024 G42 accounted for the overwhelming majority of revenue, reported at about 85 to 87 percent.[14] That share fell to 24 percent in 2025 as other customers grew, but the prospectus disclosed that about 86 percent of 2025 revenue still came from two UAE-affiliated entities: the Mohamed bin Zayed University of Artificial Intelligence at 62 percent and G42 at 24 percent.[14]

The most consequential new relationship is with [OpenAI](/wiki/openai). On January 14, 2026 the two companies announced a multiyear agreement, reported by CNBC as worth over $10 billion, to supply 750 megawatts of inference capacity through 2028.[5] Cerebras' April 2026 prospectus described a binding Master Relationship Agreement valued at more than $20 billion once expansion options are included, with provisions allowing OpenAI to scale the commitment toward 2 gigawatts by 2030.[14][17] Reporting around the deal and the IPO also described OpenAI taking an equity position of about 11 percent in Cerebras and extending a roughly $1 billion loan, at about 6 percent interest, to help build the data centers that will house the chips.[17] The arrangement diversifies Cerebras beyond its Gulf customers while binding it tightly to a single, very large US buyer.

## Significance

Cerebras' listing was the first large US technology IPO of 2026 and was widely read as a test of investor appetite for an AI hardware company that competes, in a narrow lane, against Nvidia. Nvidia held an estimated 86 percent of the AI data-center market by revenue in its most recent fiscal year, with revenue above $215 billion, and dominates training.[18] Cerebras does not challenge that breadth; instead it argues that its single-wafer design wins on latency for high-speed inference, an area analysts have suggested could grow into a market worth hundreds of billions of dollars by 2030 and where buyers want alternatives to a single supplier.

The open questions after the IPO are the same ones the prospectus raised: whether Cerebras can convert headline contracts such as the OpenAI agreement into recurring, diversified revenue; whether it can fund and ship a competitive next chip, the still-unannounced WSE-4, against Nvidia's annual cadence; and whether its dependence on a small number of customers, several with foreign-government ties, leaves it exposed to export rules and concentration risk. The WSE-4 itself, when and if it is detailed, will be the clearest signal of whether the wafer-scale approach can keep pace at the leading edge.

## References

1. CNBC, "Cerebras pops 68% in Nasdaq debut, pushing the AI chipmaker's market cap to $95 billion," May 14, 2026. https://www.cnbc.com/2026/05/14/cerebras-cbrs-stock-trade-nasdaq-ipo.html
2. CNBC, "Cerebras prices IPO above expected range, as Wall Street braces for AI tsunami," May 13, 2026. https://www.cnbc.com/2026/05/13/cerebras-prices-ipo-above-expected-range-wall-street-expects-ai-flood.html
3. CNBC, "Cerebras IPO mints two billionaires, sets stage for potential AI wave," May 14, 2026. https://www.cnbc.com/2026/05/14/cerebras-ipo-mints-two-billionaires-sets-stage-for-potential-ai-wave.html
4. CNBC, "Cerebras stock falls after blockbuster IPO debut," May 15, 2026. https://www.cnbc.com/2026/05/15/cerebras-stock-ipo-debut-ai.html
5. CNBC, "Cerebras scores OpenAI deal worth over $10 billion ahead of AI chipmaker's IPO," January 14, 2026. https://www.cnbc.com/2026/01/14/cerebras-scores-openai-deal-worth-over-10-billion.html
6. TechCrunch, "Cerebras raises $5.5B, kicking off 2026's IPO season with a bang," May 14, 2026. https://techcrunch.com/2026/05/14/cerebras-raises-5-5b-kicking-off-2026s-ipo-season-with-a-bang/
7. The Register, "Cerebras risked it all on dinner plate-sized AI accelerators a decade ago. Today it's worth $66B," May 15, 2026. https://www.theregister.com/ai-ml/2026/05/15/cerebras-wafer-scale-ai-bet-delivers-blockbuster-ipo/5240821
8. Cerebras Systems, "Cerebras Announces Third Generation Wafer-Scale Engine" (WSE-3 press release), March 2024. https://www.cerebras.ai/press-release/cerebras-announces-third-generation-wafer-scale-engine
9. The Next Platform, "Cerebras Goes Hyperscale With Third Gen Waferscale Supercomputers," March 14, 2024. https://www.nextplatform.com/ai/2024/03/14/cerebras-goes-hyperscale-with-third-gen-waferscale-supercomputers/1642584
10. ServeTheHome, "Cerebras WSE-3 AI Chip Launched 56x Larger than NVIDIA H100," March 2024. https://www.servethehome.com/cerebras-wse-3-ai-chip-launched-56x-larger-than-nvidia-h100-vertiv-supermicro-hpe-qualcomm/
11. IEEE Spectrum, "Cerebras WSE-3: Third Generation Superchip for AI." https://spectrum.ieee.org/cerebras-chip-cs3
12. Tech Startups, "AI chipmaker Cerebras revives IPO plans after $1.1B raise and CFIUS clearance," December 19, 2025. https://techstartups.com/2025/12/19/ai-chipmaker-cerebras-revives-ipo-plans-after-1-1b-raise-and-cfius-clearance/
13. GovCon Wire, "Report: Cerebras Likely to Delay Plans of Going Public Due to National Security Review of G42 Deal." https://www.govconwire.com/articles/report-cerebras-likely-to-delay-plans-of-going-public-due-to-national-security-review-of-g42-deal
14. U.S. Securities and Exchange Commission, Cerebras Systems Inc. Form S-1, April 2026. https://www.sec.gov/Archives/edgar/data/2021728/000162828026025762/cerebras-sx1april2026.htm
15. Cerebras Systems, "Faster inference from Cerebras, beats Blackwell." https://www.cerebras.ai/blog/blackwell-vs-cerebras
16. CryptoBriefing, "Cerebras reports 981 tokens per second on Kimi K2.6 model, 6.7x faster than GPU cloud," 2026. https://cryptobriefing.com/cerebras-981-tokens-kimi-k2-gpu-speed/
17. CryptoBriefing, "OpenAI commits $20B for potential 11% stake in Cerebras," 2026. https://cryptobriefing.com/openai-20b-cerebras-stake/
18. The Motley Fool, "Nvidia vs. Cerebras: Better AI Stock to Buy Now," May 17, 2026. https://www.fool.com/investing/2026/05/17/nvidia-vs-cerebras-better-ai-stock-to-buy-now/