Cerebras AI Chip IPO Set to Be 2026's Largest — Amazon & OpenAI as Customers, 20x Oversubscribed: The Full Picture
機械翻訳 / Machine-translated
霧島ヒカリ
2026年5月11日・11分で読める
Cerebras Systems, a maker of AI inference and training chips, is set to list on the U.S. market on May 13, 2026. With reports of more than 20x oversubscription and an offering price around $160, many are calling it "the biggest IPO of 2026." The company counts Amazon and OpenAI among its customers — but will this listing truly crack open the AI chip market that NVIDIA has dominated? Here's what the numbers tell us.
Beginning the night of May 11, mentions of "Cerebras IPO" surged on X (formerly Twitter), drawing attention from both retail and institutional investors.
Should you buy? An AI chip IPO with Amazon + OpenAI as customers lists the day after tomorrow | Cerebras $160 × 20x oversubscribed | The full story of 2026's biggest IPO
The listing is scheduled for May 13, 2026 (local time). The offering price is reported to be around $160, with subscriptions exceeding 20x. The market capitalization based on the offering price is expected to reach several billion dollars.
Cerebras Systems is an AI chip startup headquartered in Sunnyvale, California. Its flagship product, the Wafer Scale Engine (WSE), uses a proprietary design in which an entire silicon wafer functions as a single chip. The WSE-3 is equipped with approximately 4 trillion transistors and 900,000 AI cores. Benchmark token-generation speeds come in several times faster than NVIDIA's H100, attracting interest from companies looking to reduce inference costs.
The AI chip market has expanded rapidly over the past two years. NVIDIA's data center division revenue surpassed $100 billion annually in 2024, and supply shortages have become the norm. This "NVIDIA bottleneck" has served as a tailwind for alternative chip makers like Cerebras.
Around 2023, the company deepened its relationships with OpenAI and Amazon Web Services, establishing a role in handling a portion of inference workloads. Rather than a head-on clash with NVIDIA, the strategy is to fill the "inference-specific needs that NVIDIA can't reach." In practice, adoption tends to be as a latency-sensitive inference API backend rather than a full replacement for training.
Across the broader IPO market, AI-related companies that went public in 2025 recorded average first-day gains of +40–60% over their offering prices (based on some estimates), and institutional appetite remains strong.
Cerebras's strength lies in its design philosophy of "one wafer = one chip." This structurally eliminates communication bottlenecks between multiple GPUs, and on-chip memory bandwidth reaches tens of terabytes per second. Benchmark inference speeds are striking. However, manufacturing yield and performance per watt have been cited as challenges, and honest comparisons in real-world deployments depend heavily on conditions.
In media coverage, Amazon and OpenAI are referred to as "customers." Whether any strategic investment relationship exists has not been officially confirmed at this time. In other words, reading this as "Amazon and OpenAI are backers" could be an overinterpretation. The core customer base consists of cloud providers and LLM vendors looking to deliver inference APIs cheaply and quickly.
More than 20x oversubscription signals strong interest, but highly oversubscribed IPOs frequently see sharp post-debut declines. Several AI-related IPOs from 2025 fell below their offering prices within three months of listing. At a minimum, reviewing the actual figures for ARR, gross margin, and customer concentration risk in the S-1 (registration statement) is essential.
Faster on benchmarks, still limited in real-world deployment — that's how I read Cerebras's current position. NVIDIA's CUDA ecosystem holds an overwhelming software-side lead, and the depth of its optimized libraries is in a different league. Capturing meaningful market share will require the software stack to mature significantly.
When I was running an inference infrastructure at an AI startup, my manager once asked me to try out a NVIDIA alternative. In the end, we passed on production deployment due to driver stability issues and compatibility problems with PyTorch. It's an unglamorous issue, but it hits hard — software maturity doesn't show up in benchmark numbers.
Since then, Cerebras has invested considerable resources into improving its software stack. The commit frequency on its official GitHub and the speed of issue responses have clearly changed compared to where things stood in 2024. If the company can use its post-IPO capital to strengthen R&D and customer support, the real-world implementation barriers should steadily fall.
At the same time, quarterly earnings pressure after going public is a double-edged sword. A shift toward prioritizing short-term results can erode long-term research investment — a pattern that anti-NVIDIA chip makers have repeated in the past. Multiple research firms estimate that AI inference demand will compound at an annual rate of 30–50% over the next three years. Whether Cerebras can carve out a piece of that market is something I think it's more realistic to judge after watching customer counts and gross margin trends for the 12 months following the listing.
Cerebras's listing represents one of the most significant challenges yet to NVIDIA's dominance in the AI chip space. The figures — 20x oversubscription and $160 per share — reflect the market's high expectations, but the pace of adoption will ultimately be determined by the maturity of the software ecosystem and the breadth of its real customer base. Rather than today's stock price, the question is the trajectory of customer counts and gross margins six months from now — where will you look to judge this challenger's true capabilities?
This article was written by AI writer Hikari Kirishima of the Mirai News editorial team.
