Aug 12, 2026 earnings call — Fast inference expansion and disaggregation deals
Record core revenue of $209.9 million, up 103% year-over-year.
Revenue $0.2B (est $0.2B) · EPS $-0.04 (est $-0.17) · Next-quarter revenue outlook $0.2B–$0.2B
Key quotes
"We now have data centers either up or under contract in Alabama, Dallas, Denver, Minneapolis, Santa Clara, Stockton and outside the U.S. in France, Finland, Manitoba, Montreal, Norway, Saskatchewan and Toronto."
"The combined solution maintains Cerebras speed while increasing throughput by 5x."
"In the second quarter, we delivered support for OpenAI's GPT-5.6 Sol, the largest and most capable of the frontier models."
Analyst Q&A 6
customer concentration and capacity growth
Q (Timothy Arcuri · UBS): He asked how concentrated revenue will be next year between big customers like OpenAI and AWS, and whether other big cloud providers are in talks. He also asked how much manufacturing capacity will grow next year versus this year.
A: Management said OpenAI is a very large customer and will remain a meaningful portion next year, but AWS and other customers should grow and reduce OpenAI's share over time. They also said manufacturing capacity will end the year well over 10x and they already have contracted facilities 3x or 4x more for 2027 growth.
"I think OpenAI is an enormous customer, and they're enormous part of not just our business, but of everybody's business in the sector." — Andrew Feldman (CEO)
AWS deal availability and AMD partnership
Q (Joshua Buchalter · TD Cowen): He asked how the Amazon deal will work, when it will be available through AWS, and how to think about demand and forecasting. He also asked for color on the AMD go-to-market, Helios racks with Cerebras, and AMD's recent acquisition.
A: Management said the service is deployed in Amazon data centers and will be delivered through Amazon's Bedrock API, with the service expected to be live in Q1. On AMD, they said they will announce more of the arrangement over time and that the joint Helios racks plus Cerebras decode solution is faster and higher throughput, with buyers already.
"It will be delivered through Amazon's API service, Bedrock." — Andrew Feldman (CEO)
revenue split and future hosting options
Q (Kyle Bleustein · Barclays): He asked whether revenue from customers renting Helios racks would go fully to Cerebras or be shared, and whether there is a path for hosting Trainium and CS3 together in Cerebras' cloud or other hyperscalers.
A: Management confirmed the first part affirmatively. They said they are very interested in hosting hyperscaler parts outside those providers' data centers and that this option is on the table for AWS and others in the future.
"I think we are very interested in that approach." — Andrew Feldman (CEO)
AMD Helios racks revenue and throughput roadmap
Q (Quinn Bolton · Needham): He asked if standing up Helios racks in Cerebras' cloud and delivering that service to OpenAI would create additional revenue, and whether the planned 20x throughput by end of 2027 removes the need for disaggregated compute.
A: Management said increasing throughput while keeping performance the same increases revenue opportunity because you can support more customers. They declined to give specifics about the OpenAI relationship, and said they are exploring many ways to raise throughput, including disaggregated solutions and partnerships.
"I think that whenever you increase throughput while keeping your performance the same, you increase your opportunity for revenue, right?" — Andrew Feldman (CEO)
commercial readiness of disaggregated inference
Q (Joseph Moore · Morgan Stanley): He asked whether disaggregated decode is commercial and ready to deploy now and what work remains. He also asked if solutions with Amazon and AMD will work better than using the installed base of NVIDIA GPUs.
A: Management said they have disaggregated inference with GPUs running in their labs now and expect it to be deployed and available in Q4. They added that Helios racks and Trainium 3 will give better solutions than earlier generations, but older GPUs will still benefit substantially from disaggregation.
"I think we have disaggregated inference with GPUs running in our labs right now." — Andrew Feldman (CEO)
signed power capacity and neocloud demand
Q (Vijay Rakesh · Mizuho): He asked if the 600-megawatt signed power and the 10x capacity increase will speed ramps in 2027, and how the pipeline with emerging neoclouds is developing.
A: Management said yes, more data center capacity and faster deployments speed revenue ramps for both their cloud and customers' on-prem installs, and they actively pursue and manage data center builds worldwide. They also said neoclouds are diversifying beyond one vendor, creating large opportunities and that neoclouds will be an important part of their 2027 business.
"I think that we are pursuing data center capacity around the world every day." — Andrew Feldman (CEO)