Cerebras falls behind on first OpenAI capacity tranches as data center deals lag
AI chipmaker Cerebras Systems is behind schedule on its first capacity deliveries to OpenAI, citing delays in securing data center facilities rather than production problems with its chips.
The AI chipmaker has pinned the delay on data center contracts that are taking longer than planned. Cerebras has described the pace of securing facilities as moving at 'the speed of real estate.' The two companies signed a Master Relationship Agreement on December 24, 2025, under which OpenAI agreed to buy 750 megawatts of inference capacity split into three equal tranches.
Contract Details
The first 250 MW is due by the end of 2026, with another 250 MW following by end of 2027, and the final 250 MW by end of 2028. The total agreement value exceeds $20 billion and includes performance bonuses. OpenAI also holds an option for an additional 1.25 gigawatts of capacity by 2030.
OpenAI extended Cerebras a $1 billion working-capital loan in January 2026 to support the buildout. Cerebras went public in May 2026 and has reported strong cloud revenue growth despite infrastructure constraints, with its backlog now exceeding $25 billion.
Infrastructure Challenges
Cerebras has identified data center buildouts as the critical constraint in meeting OpenAI's rising demand. The company is pursuing new data center sites across North America and Europe, including a 165 MW facility in Finland that it is working to secure.
infrastructure and data center buildouts remain the critical constraint in meeting OpenAI's rising demand
Implications for Stakeholders
For Cerebras shareholders, the delay represents execution risk. The company has significant demand and a marquee customer but needs physical capacity to come online on schedule. Delays could push when backlog turns into recognized revenue.
The performance-linked bonuses mean delivery timing and quality could affect how much value Cerebras ultimately captures from the agreement. The company is already behind on its first deadline of 250 MW by end of 2026, suggesting that power, sites and contracts are the binding constraints in AI compute rather than demand.
Related
