This is a first-quarter earnings call analysis for Cerebras Systems (CBRS), with the host first reading the 8-K/press release before the live call and then reacting to management’s remarks. The core message from management was that AI inference demand is strong, Cerebras claims a major speed advantage over GPUs, and Q1 revenue grew sharply, but the stock sold off on concerns about hardware growth, margin durability, and execution against large contracts like OpenAI and AWS.
Watch on YouTube ›Get the market thesis, key claims, assets, contradictions, and follow-up questions from any financial video — then unlock a version personalized to your portfolio, watchlist, and favorite speakers.
This transcript is centered on Cerebras’ first earnings call as a public company and the market reaction around it. The host opens by scrambling to find the release, then reads the early numbers before the call starts: Q1 core revenue of $191.3 million, core hardware revenue of about $111 million, cloud and services revenue around $80 million, and the stock trading lower after hours. The host’s initial read is mixed: revenue growth is impressive, but the stock’s selloff suggests the market is focused on hardware growth not being as strong as expected and on whether Cerebras can justify its valuation against much larger AI infrastructure names. Andrew Feldman’s prepared remarks frame Cerebras as a company built around one central thesis: fast AI inference is more valuable than slow AI inference, and Cerebras’ wafer-scale architecture makes it “the fastest AI in the world.” He argues …
Tactically, the stock looks vulnerable until the market sees whether Q2 growth, capacity ramp, and customer conversion offset the initial post-IPO selloff. The immediate risk is that investors focus on hardware softness and valuation rather than the OpenAI/AWS backlog.
Over the next few quarters, the setup depends on whether Cerebras can turn contracted demand into visible cloud revenue and keep margins from slipping too much while it scales capacity. If those ramps land, the stock can re-rate; if they slip, the story stays aspirational.
Structurally, the bull case is that inference speed becomes a premium category and Cerebras wins a durable niche or even a broad share of AI inference. The long-run risk is that the market likes the story but still prefers cheaper, more established compute stacks if differentiation does not keep widening.
Cerebras delivers the fastest AI inference in the world, by an order of magnitude, across all model sizes and KV cache sizes.
Feldman demonstrates with a live benchmark showing Cerebras completing a trillion-parameter model inference in 21 seconds versus 4 minutes 37 seconds for a leading GPU (13x faster), and asserts this holds across model sizes and KV cache configurations.
Cerebras signed a definitive agreement with OpenAI on December 24, 2025 for the purchase of more than $20 billion of Cerebras compute over several years.
Speaker states timeline from signature to production was 35 days, implying strong execution.
Fast AI inference tokens are the most valuable because they get more work done in less time, and slow inference is unproductive.
Feldman analogizes slow AI to slow search, slow internet, and dialup — arguing that just as speed drove adoption and value in those markets, it will in AI inference.
Now that you have the definitive agreement with AWS, can you help us think about the timing and your ability to supply that customer?
Andrew says TSMC has been good to them and they have supply for their plan and beyond in 2026, but AWS's impact should be expected in 2027.
You talked about multiple partners for disaggregated solutions — does that imply there's another customer beyond AWS?
Andrew explains the opportunity to provide decode for GPU users is real, noting GPUs struggle with the sequential nature of decode while Cerebras excels at it, making partnerships on that vector sensible.
What do you think your TAM is when you look at the broader AI market, especially regarding your ability to handle larger models?
Andrew argues that slow technologies don't own meaningful market share over time, citing search and dialup as examples. He says Cerebras sees the entire inference market as available for fast inference, acknowledging this is at odds with GPU makers' self-interested views.
Unlock the full claims, asset map, scores, related transcripts, follow-up questions, and AI chat — shaped around your portfolio, watchlist, favorite speakers, and risks.
