The following risks represent material, specific concerns drawn from the S-1 filing. They are not exhaustive — the S-1 contains dozens of additional risk disclosures — but these are the ones with the highest near-term probability and potential magnitude.

Extreme customer concentration (existential). In FY2025, 86% of total revenue came from two related UAE-based entities: MBZUAI (62%) and G42 (24%). MBZUAI represented 77.9% of accounts receivable at year-end. The S-1 states: "A reduction in demand from, or a material adverse development in our relationship with any of our significant customers...would harm our business, financial condition, results of operations, and prospects."¹ The OpenAI MRA is expected to improve diversification in future periods, but substitutes one dominant customer for another, and the MRA contains termination rights for OpenAI if Cerebras fails to meet service level or delivery obligations.
OpenAI MRA performance and termination risk (existential). Under the MRA, Cerebras must deliver 250 MW capacity tranches across specified data centers by time-based milestones. Failure to meet delivery or service levels gives OpenAI the right to terminate a portion or all of the agreement. Termination also triggers potential immediate repayment of the $1.0 billion Working Capital Loan. The S-1 notes that OpenAI would gain control of the loan funds account upon certain trigger events.¹
Gross margin compression during cloud buildout (headwind). Management has explicitly guided that gross margins will decline below recent levels in 2026 as data center pass-through costs run through the income statement under the OpenAI MRA structure, and as customer warrant contra-revenue begins to reduce reported revenue. Non-GAAP operating losses are already widening despite revenue growth, from $(42.9M) in 2024 to $(96.1M) in 2025.¹
Material weaknesses in internal controls (headwind). The S-1 discloses the identification of material weaknesses in internal control over financial reporting. Cerebras elected to use the Emerging Growth Company extended transition period for accounting standards compliance. An inability to remediate these weaknesses could impair the accuracy or timeliness of public financial reporting and adversely affect investor confidence.¹
Fabless manufacturing concentration (headwind). Cerebras operates a fabless model, relying on third-party foundries for wafer production and module assembly. It does not hold long-term capacity commitments with suppliers and depends on sole-source suppliers for certain components. Given the uniqueness of wafer-scale die fabrication, supply chain disruption could halt production with limited alternative sourcing options.¹
CFIUS and geopolitical exposure (headwind). G42 and MBZUAI are Abu Dhabi-based entities. A prior CFIUS review of G42's Cerebras investment delayed the company's first IPO attempt in 2024.⁸ While restructuring addressed the initial concern, the S-1 discloses ongoing export control obligations and warns that geopolitical or regulatory changes affecting the UAE could materially harm the business.
Competitive response from NVIDIA and hyperscaler-developed silicon (headwind). NVIDIA continues to invest in NVLink interconnect bandwidth and software ecosystem depth. Google, Amazon, and Microsoft develop custom AI accelerators optimized for their inference workloads. The WSE's speed advantage is architectural but not unassailable; sufficiently improved interconnect bandwidth in future GPU architectures could narrow the latency gap, particularly for workloads that can be efficiently parallelized across chips.¹

"We identified material weaknesses in our internal control over financial reporting. If we are unable to remediate these material weaknesses, or if we identify additional material weaknesses in the future or otherwise fail to maintain an effective system of internal controls, we may not be able to accurately or timely report our financial condition or results of operations."
— Risk Factors, Cerebras S-1 Registration Statement¹

Cerebras Systems IPO: The Wafer-Scale Bet on AI Inference Speed

The Power Delivery Gap