Cerebras, a specialized AI chip maker, just completed 2026's largest initial public offering with a $48.8 billion valuation, but a closer look at its financial filings reveals a more complicated picture than the "Nvidia challenger" narrative suggests. The company reported $510 million in revenue and a $237.8 million profit, but nearly all of that profit came from a one-time accounting gain unrelated to actual business operations. When adjusted for real operational performance, Cerebras actually lost $75.7 million in 2025, a significant deterioration from its $21.8 million loss the year before.

What Makes Cerebras Different From Nvidia?

Cerebras is not trying to compete with Nvidia across all computing tasks. Instead, the company has carved out a specialized niche in AI inference, which is the process of running trained AI models to generate responses. The company's WSE-3 chip contains 4 trillion transistors and 900,000 AI cores, allowing it to process requests with remarkable speed. In independent benchmarks, Cerebras' CS-3 system outputs over 2,500 tokens per user per second when running Llama 4 Maverick, a large language model with 400 billion parameters. By comparison, Nvidia's flagship DGX B200 achieves approximately 1,000 tokens per second, while competitors Groq and SambaNova reach 549 and 794 tokens respectively.

However, this advantage applies only to inference workloads. Cerebras explicitly states in its prospectus that it has no intention or capability to challenge Nvidia's dominance in training large AI models or general-purpose computing. Nvidia's CUDA ecosystem, built over nearly 20 years since 2007, remains entrenched with developer tools, libraries, and community support that Cerebras cannot easily replicate.

Where Is Cerebras' Revenue Really Coming From?

The company's customer concentration presents a significant risk. In 2025, two UAE-based entities accounted for 86 percent of Cerebras' total revenue. Mohammed bin Zayed University for Artificial Intelligence (MBZUAI) contributed 62 percent, while G42, a related entity, contributed 24 percent. MBZUAI alone represents 77.9 percent of the company's accounts receivable, meaning money owed to Cerebras is heavily concentrated with a single customer.

This concentration echoes a previous crisis. In 2024, Cerebras attempted to go public but withdrew its application after the Committee on Foreign Investment in the United States (CFIUS) launched an investigation into G42's role as a dominant customer. Rather than diversifying its customer base, the company appears to have shifted G42's revenue weight to MBZUAI, which is also located in the UAE and maintains ties to G42.

How Does OpenAI Fit Into Cerebras' Business Model?

OpenAI's involvement in Cerebras extends far beyond a simple customer relationship. The two companies signed a computing power contract valued at over $20 billion, with OpenAI committing to purchase 750 megawatts of computing capacity. However, the same agreement includes several other arrangements that blur the lines between customer, lender, and investor. OpenAI provided Cerebras with a $1 billion loan, received nearly free warrants for 33 million shares of Cerebras stock, and included exclusivity clauses in the Master Relationship Agreement that restrict Cerebras from selling to certain competitors.

This multi-layered relationship means OpenAI simultaneously serves as Cerebras' customer, lender, upcoming shareholder, and strategic controller. The arrangement raises questions about whether the revenue from OpenAI represents a genuine market transaction or a form of financial engineering designed to support Cerebras' valuation.

Steps to Understanding Cerebras' Financial Reality

• Examine Non-GAAP Adjustments: The company reported $237.8 million in GAAP net profit, but $363.3 million of this came from a one-time, non-cash gain related to the extinguishment of a forward contract liability with G42, not from selling chips.
• Review Customer Concentration: Two UAE-based entities account for 86 percent of revenue, with MBZUAI alone representing 77.9 percent of accounts receivable, creating significant dependency on a narrow customer base.
• Analyze Related-Party Transactions: OpenAI's role includes providing a $1 billion loan, receiving warrants for 33 million shares, and imposing exclusivity restrictions, making it difficult to assess the arm's-length nature of the $20 billion computing contract.
• Compare Valuation Multiples: At $48.8 billion valuation on $510 million revenue, Cerebras trades at a price-to-sales ratio of 95, demanding extraordinary growth assumptions to justify the valuation.

What Does the Broader Market Look Like for Inference Chips?

Cerebras is not alone in pursuing AI inference as a specialized market. Nvidia, the dominant player, is investing heavily to maintain its lead. The company plans to spend over $18 billion on research and development in fiscal year 2025 and acquired Groq, an AI inference startup, for $20 billion in December 2025. Nvidia also invested $4 billion in photonics technology companies in March 2026 to advance its capabilities.

Meanwhile, other major technology companies are developing their own chips. AMD's MI400 has reached 320 billion transistors, Google offers TPU v6, Amazon provides Trainium 3, and Microsoft developed Maia 2. Nvidia's upcoming Vera Rubin architecture, unveiled at GTC 2026, boasts 336 billion transistors and claims a performance leap of 5 times over its current Blackwell generation.

In this competitive landscape, Cerebras' specialized advantage in inference speed may prove valuable, but the company faces relentless competition from well-capitalized rivals with broader product portfolios and established customer relationships. The $48.8 billion valuation assumes Cerebras will capture a significant share of a growing inference market, but execution risk remains substantial given the company's narrow customer base and the aggressive competition ahead.