Cerebras has secured Microsoft as a customer for its AI hardware and cloud services, marking a significant expansion in its partnerships with major cloud providers. The deal, confirmed during Cerebras’ Q3 2026 earnings call, makes Microsoft the third hyperscaler to adopt the company’s technology, following Amazon Web Services and Oracle. While financial and technical details remain undisclosed, CEO Andrew Feldman stated that further information would emerge in 2027.
Microsoft’s outage highlights infrastructure risks
The announcement comes just days after a global Microsoft outage disrupted critical services across governments, banks, healthcare, and transportation in 74 countries. Beginning around noon ET on November 12, the outage affected Outlook, Teams, Copilot, and Exchange Online, with some users unable to access email or join scheduled meetings for over 24 hours. Microsoft attributed the disruption to a technical issue during a network capacity upgrade, not a cyberattack. Recovery progressed unevenly, with North America regaining services faster than Europe and Asia.
The economic impact could exceed $7 billion in lost productivity, according to the Anderson Economic Group. Up to 51 million users may have faced operational losses on the second day alone. UK police forces, including those in Greater Manchester and Dorset, advised the public to report non-emergencies by phone due to disabled online forms. In the US, city governments from Jacksonville to Austin reported service interruptions, while courts in the Philippines and schools in Australia faced delays.
Cerebras expands global footprint amid capacity constraints
Cerebras’ growing appeal lies in its ability to offer rapid AI compute scaling through its own cloud platform. CFO Bob Komin explained that hyperscalers often lack near-term data center availability for new hardware, prompting Cerebras to launch its private cloud in early 2025. This allowed customers like Microsoft to evaluate its systems before committing. The company reported $231 million in Q3 revenue, a 70% year-on-year increase, with AI cloud contributing $145 million.
However, data center capacity remains a bottleneck. Limited by land, power, and facility shells, Cerebras expects only limited capacity ramp-up in December 2026, with major deployments starting mid-2027. To address this, over 75% of its planned 700MW capacity will be located outside the US, including new sites in Norway, Finland, Canada, France, and an upcoming facility in São Paulo—its first in South America.
Addressing doubts on hardware and partnerships
Cerebras has faced market speculation about its manufacturing yields and hardware durability. Short-seller Culper Research claimed its wafers degrade after two years, threatening long-term viability. In response, Feldman denied yield issues, attributing lower hardware margins to early customer pricing programs and manufacturing ramp-up. He expects margins to rise to 50% long-term, up from 41% in Q3.
The company also clarified delays in its OpenAI partnership, citing slower-than-expected data center contracting. While Cerebras supported OpenAI’s GPT-5.6 Sol preview, it lacked capacity for broader rollouts, including the Astra model. Up to 5% of the contract value may be at risk if delays exceed 90 days, though no cancellations have occurred.
What it means
For enterprises, Microsoft’s reliance on third-party AI infrastructure like Cerebras signals a shift toward hybrid sourcing to mitigate dependency risks. The recent outage underscores the cost of centralised cloud failures, pushing firms to evaluate redundancy and alternative providers. Cerebras’ global expansion offers options beyond US data centers, appealing to organisations in the UK, UAE, and Kazakhstan seeking lower latency and data sovereignty.
Businesses should assess their exposure to single-cloud dependencies and consider multi-supplier AI strategies. While Microsoft’s services are recovering, the $7 billion disruption is a stark reminder of systemic fragility. Cerebras’ model—offering both hardware and cloud—provides a path to faster deployment, though its long-term reliability will depend on sustained capacity growth and yield stability.