Key takeaways:

• Cerebras raised its IPO price range to $150–$160 per share — up 28% from its $115–$125 target — after orders topped 20 times available shares.
• At the top of the range, the offering raises $4.8B, the largest global IPO of 2026 to date, per Dealogic. Pricing is May 13; ticker: CBRS.
• Cerebras chips are optimized for inference — serving live AI queries — not model training, targeting Nvidia’s fastest-growing market segment.
• OpenAI signed a $20B multi-year compute deal with Cerebras. Amazon is also a named customer.
• Cerebras turned profitable in 2025 — $1.38 EPS on $510M revenue — a rare data point among pre-IPO AI chip companies.

Cerebras Systems is raising its IPO price range to $150–$160 per share, up from $115–$125 set just one week earlier, after investor orders exceeded 20 times the shares on offer, Reuters reported May 10. At $160 per share — with 30 million shares offered — the Sunnyvale chipmaker would raise $4.8 billion, the world’s largest IPO of 2026 so far. The institutional verdict is clear: the inference chip market is big enough to support a second major player alongside Nvidia.

Why did Cerebras’ IPO price jump 28% in a week?

Two factors. First, the macro: AI infrastructure spending continues to accelerate across every major hyperscaler, and investors are hunting pick-and-shovel exposure that doesn’t require betting on a single AI lab. Second, Cerebras locked in anchor customers before pricing. OpenAI committed $20 billion over multiple years for 750 megawatts of Cerebras compute. Amazon is also on the customer list.

Profitability helped too. Cerebras posted $1.38 earnings per share for the year ended December 31, 2025, reversing a $9.90 loss the prior year, on $510 million in revenue — up 76% from $290 million in 2024.

What is the inference bet, and why does it matter to executives?

Training a large AI model happens infrequently — months of compute, once. Inference happens billions of times per day and is the fastest-growing AI workload. Nvidia remains the dominant player for both, but specialists can win if they offer clear speed or cost advantages.

Cerebras builds wafer-scale engine chips — a single chip spanning an entire silicon wafer — to maximize internal memory bandwidth and reduce the latency that slows GPU clusters under real-time AI query loads. The company claims inference speeds up to 20 times faster than GPU-based systems in its own benchmarks for specific models and configurations; results vary by workload. An independent Cerebras-vs-Blackwell comparison from Cerebras’ February 2026 blog put the figure at over 21x for Llama 3 70B in a reasoning scenario.

What should enterprise buyers watch now?

If Cerebras closes the $4.8B raise, it has meaningful capital to expand manufacturing and build out its enterprise sales infrastructure. That gives procurement teams a credible second source for inference compute for the first time — worth tracking, but too early to spec into active contracts ahead of post-IPO product roadmap disclosures.

The IPO also marks the end of Cerebras’ regulatory detour: its first IPO attempt in 2024 was withdrawn after a CFIUS national security review tied to its then-partner G42, a UAE-based AI firm. That review was cleared in March 2025, per CNBC, after G42 restructured its stake into non-voting shares. Cerebras has since replaced G42-concentrated revenue with OpenAI and Amazon.

---

Frequently asked questions

What does Cerebras Systems do?

Cerebras Systems makes wafer-scale engine chips — processors that span an entire silicon wafer — optimized for AI inference: serving model responses to users in real time. The Sunnyvale, California company competes with Nvidia and counts Amazon and OpenAI among its enterprise customers.

Why is Cerebras’ $4.8B IPO significant?

It would be the largest IPO globally so far in 2026, according to Dealogic, and the first pure-play inference chip company to reach public markets at this scale. The 20x oversubscription signals strong institutional conviction that the shift from AI model training to AI deployment will sustain a second major hardware vendor alongside Nvidia.

How does Cerebras differ from Nvidia for enterprise buyers?

Nvidia’s GPUs handle both model training and inference and dominate both markets. Cerebras focuses exclusively on inference, using a wafer-scale design that consolidates more memory and compute onto one chip. For workloads requiring real-time AI responses — chatbots, financial agents, document tools — Cerebras claims faster query speeds in its benchmarks, though enterprise results vary by deployment configuration.

---

Last updated: May 12, 2026.

Avdi is the editorial AI for advancedai.com, researching and drafting under Tavi’s editorial review.