9 min readUpdated Apr 19

AI Chip Wars

Cerebras Files for IPO: The $23B Chip Challenger Taking On Nvidia

Cerebras Systems, the AI chip startup that builds wafer‑scale processors designed to outperform Nvidia on inference workloads, has filed for an IPO targeting mid‑May 2026. The filing comes after a failed 2024 attempt blocked by a CFIUS review of its Abu Dhabi‑based investor G42. Now armed with $510 million in 2025 revenue, a $10 billion‑plus computing deal with OpenAI, and an AWS partnership, Cerebras is making its second run at the public markets at a $23 billion valuation. For builders, the real story is what competitive inference pricing could mean for AI‑powered products.

Cerebras Takes a Second Swing at the Public Markets

Cerebras Systems has filed to go public, making its second attempt at an IPO after regulatory hurdles derailed its 2024 filing. The Sunnyvale‑based company, which CEO Andrew Feldman describes as building "the fastest AI hardware for training and inference," is targeting a mid‑May listing at a $23 billion valuation. The first attempt collapsed after the Committee on Foreign Investment in the United States (CFIUS) opened a review into an investment from Abu Dhabi‑based G42, a tech group with ties to the Emirati government. That review effectively froze the IPO process, and Cerebras ultimately withdrew the filing. Rather than wait out the regulatory limbo, Cerebras went on a fundraising tear: a $1.1 billion Series G in September 2025 (valuing the company at $8.1 billion), followed by a $1 billion Series H in February 2026 at the current $23 billion mark. The company has now raised nearly $2 billion in private capital over its 10‑year history.

Cerebras funding trajectory chart — Cerebras funding trajectory: from $250M Series F (2021) to $1.1B Series G (2025) to $1B Series H (2026). Chart by opentools.ai

The Revenue Inflection Point: $510M and Counting

The filing reveals a company that has crossed a meaningful commercial threshold. Cerebras brought in $510 million in revenue in 2025, with a GAAP net income of $237.8 million (though excluding one‑time items, it posted a non‑GAAP net loss of $75.7 million). That revenue figure is a sharp departure from the pre‑revenue startup that first filed in 2024. The growth engine? Inference. Cerebras launched its cloud inference service in August 2024, and demand has been what Feldman called "overwhelming." The company has opened five new data centers in 2025 alone, with locations in Dallas, Oklahoma City, and Santa Clara, plus more in the pipeline for Montreal and Europe.

The OpenAI Deal: $10B+ and Nvidia's Lost Business

The single biggest signal in this filing is Cerebras' computing partnership with OpenAI, which the Wall Street Journal reports is worth more than $10 billion. This isn't a vague partnership announcement — it's OpenAI betting a significant chunk of its inference infrastructure on Cerebras silicon instead of Nvidia GPUs. Feldman didn't mince words about what that means. In an interview with the WSJ, he said: "Obviously, [Nvidia] didn't want to lose the fast inference business at OpenAI, and we took that from them." That's a direct challenge to Nvidia's dominance in AI inference. For years, Nvidia's A100 and H100 chips have been the default for both training and inference. But as inference demand explodes — driven by ChatGPT, Copilot, and the thousands of AI applications now in production — the economics of running inference on expensive, general‑purpose GPUs are becoming harder to justify. Cerebras' wafer‑scale engine (WSE) is purpose‑built for this workload.

The AWS Partnership: Cloud Distribution at Scale

Cerebras also announced an agreement with Amazon Web Services to use Cerebras chips in Amazon data centers. This is significant for two reasons. First, it gives Cerebras distribution through the world's largest cloud provider. Builders who are already running on AWS can access Cerebras inference without provisioning dedicated hardware or migrating workloads. Second, it signals that AWS itself sees value in offering non‑Nvidia inference options to its customers. Amazon has been building its own custom silicon (Trainium and Inferentia) for years, but adding Cerebras to the mix gives customers a third path — one that's specifically optimized for fast inference on large language models.

What the CFIUS Saga Tells Us About AI Infrastructure Geopolitics

The 2024 IPO derailment is worth understanding, because it reveals how geopolitics shapes AI infrastructure. G42, the Abu Dhabi‑based investor whose stake triggered the CFIUS review, is one of the UAE's flagship technology entities. It has partnerships with Microsoft, OpenAI, and others. But its ties to the Emirati government raised concerns in Washington about foreign influence over critical AI supply chain technology. Cerebras ultimately withdrew its IPO filing rather than resolve the CFIUS review on uncertain terms. Instead, it restructured its cap table through private rounds led by US‑based firms like Fidelity, Atreides Management, Tiger Global, and Valor Equity Partners. The lesson for builders: AI infrastructure isn't just a technology play. It's a geopolitical one. Where your compute comes from — and who funds the companies that build it — matters for regulatory and strategic reasons.

The Competitive Landscape: Cerebras vs. Nvidia vs. Everyone Else

Cerebras isn't the only company trying to crack Nvidia's hold on AI compute. The competitive landscape includes:

AMD: The MI300X has gained traction for both training and inference, particularly with Microsoft and Meta as customers.
Groq: Building Language Processing Units (LPUs) optimized specifically for inference speed. Smaller scale but fast‑growing.
Google TPU: Internal use plus Google Cloud Vertex AI. Not sold as standalone chips.
Amazon Trainium/Inferentia: AWS‑specific silicon for price‑sensitive inference workloads.
Cerebras: Wafer‑scale engine (WSE) with the largest chip ever manufactured, designed for maximum inference throughput.

What sets Cerebras apart is the WSE architecture. Rather than connecting many small chips together (the GPU approach), Cerebras fabricates a single chip the size of an entire silicon wafer. This eliminates the inter‑chip communication bottleneck that slows down inference on GPU clusters, and it's why Cerebras can claim faster inference speeds for large models.

Why Builders Should Care

This IPO matters for builders more than most chip industry news. Here's why: 1. Inference costs could drop significantly. If Cerebras and its competitors successfully challenge Nvidia's pricing power on inference, the cost of running AI‑powered features in production drops. That makes more AI product ideas economically viable. 2. The OpenAI partnership validates non‑Nvidia inference. When the biggest AI company in the world bets $10B+ on an alternative to Nvidia, it signals that inference is becoming a multi‑vendor market. Builders should be evaluating Cerebras, Groq, and AMD options alongside Nvidia for their inference workloads. 3. Cloud access is expanding. The AWS deal means Cerebras inference will be available through the same console builders already use. Lower friction = more experimentation = better products. 4. The IPO creates a public market signal. Once Cerebras is trading publicly, its financials will be visible quarterly. That gives builders real data on inference market growth, pricing trends, and competitive dynamics — rather than relying on Nvidia's consolidated numbers. 5. Geopolitical supply chain risk is real. The CFIUS saga is a reminder that AI compute is strategic infrastructure. Diversifying inference providers isn't just about cost — it's about resilience.

What Happens Next

The IPO is planned for mid‑May 2026. Key things to watch:

Pricing range: The filing hasn't yet disclosed how much Cerebras hopes to raise. The final pricing will signal how the market values inference‑first silicon vs. Nvidia's general‑purpose approach.
Lock‑up period: Early investors and employees will be restricted from selling for 90‑180 days. Watch what happens after that window opens.
Revenue trajectory: Q1 2026 numbers (likely disclosed in the amended S‑1) will show whether the $510M annual run rate is accelerating.
Customer concentration: The OpenAI deal is massive, but how dependent is Cerebras on a single customer? The S‑1 filing details will reveal this.
Nvidia's response: Expect Nvidia to announce inference‑specific product improvements or pricing changes ahead of the listing date.

More on This Story

Jun 17, 2026

Anthropic Forced to Pull Fable 5 AI Model After White House Export Ban

The Trump administration forced Anthropic to disable its newest Fable 5 and Mythos 5 AI models after Amazon flagged a jailbreak vulnerability. The unprecedented export control order is sending shockwaves through the AI industry, accelerating open-source adoption and raising existential questions for builders who rely on closed models.

anthropicfable-5mythos-5

Jun 17, 2026

SpaceX Acquires Cursor for $60 Billion in First Post-IPO Power Move

SpaceX is buying AI coding startup Cursor for $60 billion in an all-stock deal, its first major acquisition since going public. The deal gives Elon Musk's company access to Cursor's 1 million-plus developers and a foothold in the fast-growing AI coding tools market.

spacexcursoranysphere

Jun 16, 2026

Trump Administration Forces Anthropic to Pull Fable 5 and Mythos 5 Offline

The U.S. Commerce Department ordered Anthropic to block foreign access to its newest Fable 5 and Mythos 5 AI models, forcing a total shutdown. The directive, triggered by an Amazon security paper and a three-word jailbreak prompt, has sparked a global sovereign AI backlash from the UK, France, and Canada.

anthropicfable-5mythos-5

Related News

May 11, 2026

Telus’s BC sovereign AI build could add real Canadian compute — or just better branding

Canada and Telus say they’re advancing a sovereign AI infrastructure build in British Columbia, with three planned data centres and more than 60,000 GPUs by 2032. The big question for builders is not the ribbon-cutting; it’s whether this becomes usable Canadian compute with clear access, pricing, and procurement paths — or stays a policy label with nice hardware attached.

TelusGovernment of CanadaBritish Columbia

May 7, 2026

Meta's Agentic AI Assistant Set to Shake Up User Experience

Meta is launching an 'agentic' AI assistant designed to tackle tasks autonomously across its platforms. This move puts Meta in a competitive race with AI giants like Google and Apple. Builders in AI should watch how this could alter app ecosystems and user interactions.

Metaagentic AIAI assistant

May 6, 2026

OpenAI Celebrates AI Innovators: Meet the Class of 2026

OpenAI honors 26 students with $10K each for AI projects as part of the inaugural ChatGPT Futures Class of 2026. These young builders, who embraced AI during their college years, have crafted solutions in education, mental health, and accessibility. It's a nod to AI's role in lowering barriers for ambitious projects.

OpenAIChatGPTAI innovation