Assertion Partly supported AI assessment confidence: 85% certainty 4/5 debate potential 3/5

Feldman: Cerebras is 15 to 20 times faster than GPUs at inference

Andrew Feldman · The Story Behind Cerebras’ $63 Billion IPO with Founder and CEO Andrew Feldman · May 21, 2026 · at 1:37

Andrew Feldman, CEO of Cerebras, outlines the performance advantage of Cerebras' wafer-scale processors over traditional GPUs for AI inference workloads.

0:00 / 0:07exact quote · 7.2s

▶ Watch the full episode on YouTube → 720p mp4 · rendered on demand · StarZero watermark

“And right now we're the fastest at inference, not by a little bit, but by a lot. 1518, 20 X faster than GPUs.”

quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →

The claim, plainly

Cerebras is currently the fastest at AI inference, performing 15 to 20 times faster than GPUs.

a quote alone can be ambiguous, so this restates it with the references and context of the conversation around it resolved. AI-written; the tape below the player is the source.

Assessment

Andrew Feldman's assertion that Cerebras was the fastest AI inference provider as of May 2026 is substantiated by independent benchmarking from Artificial Analysis, which consistently recorded Cerebras achieving the highest per-stream token generation speeds across multiple architectures (including Llama 3.1, GPT-OSS 120B, and Kimi K2.6). Furthermore, the 15x to 20x speed advantage is accurate when compared against typical and median cloud-deployed NVIDIA GPU endpoints (e.g., measuring ~16x faster on Llama 3.1 70B and ~23x faster than median GPU providers on Kimi K2.6). However, the claim is only partially true because the 15x–20x multiplier does not hold across all contemporary market-leading GPU setups: against peak optimized NVIDIA Blackwell (GB200) clusters with speculative decoding, the Cerebras performance lead narrowed to between 2.3x and 6.7x. Additionally, the speed advantage is specific to single-stream / low-batch decode latency and does not apply to aggregate high-batch throughput.

Artificial Analysis benchmarks verified Cerebras as the fastest inference platform, achieving 1,800+ tokens/sec on Llama 3.1 8B, 2,100 tokens/sec on Llama 3.1 70B (up to 16x faster than optimized GPU endpoints in late 2024), and 981 tokens/sec on Kimi K2.6 in May 2026 (~23x faster than median cloud GPU providers).
On newer architectures and highly optimized NVIDIA Blackwell (GB200) clusters tested in late 2025/early 2026, Blackwell reached up to 650 tokens/sec on GPT-OSS 120B, narrowing Cerebras's lead over top-tier peak GPUs to roughly 2.3x to 4.6x.
Comparative architectural evaluations (including IEEE Computer Society analyses) demonstrated that the 15x–20x+ speed advantage applies specifically to single-user / batch size 1 decode latency rather than aggregate batch throughput.

Checked against

[1] techpowerup.com [2] cerebras.ai [3] cerebras.ai [4] businesswire.com [5] cerebras.ai [6] artificialanalysis.ai [7] gmicloud.ai [8] cerebras.ai

Cerebras inference speed vs GPU 20x 15x Artificial Analysis MLPerf Cerebras inference Artificial Analysis benchmark speed tokens per second GPU comparison Artificial Analysis Cerebras Llama 3 inference speed Cerebras inference speed benchmark comparison NVIDIA H100

assessed as of 2026-09-03

This assessment is an AI reading of a public statement, made against the cited references with the confidence shown. It can be wrong. The quote, the tape and the references are all on this page; if something is off, use report an error below. full disclaimer →

Mentioned

Cerebras Systems GPU

⚑ Spot an error? Report it

More from Andrew Feldman

Opinion

NVIDIA is extorting customers amid AI chip shortages, says Cerebras CEO

“I think they're now in a situation where they're extorting customers. They're extremely expensive. They're unable to ship. And that has, among other things, opened the door for many of us who have alternatives.”

Andrew Feldman Sep 7, 2023 ▶ 18:10 No Priors Ep. 31 | With Cerebras CEO Andrew Feldman

Prediction Not checkable as stated

Feldman: The market for slow AI inference will become zero

“How big is the market for slow search? It's zero. How big is the market for dial-up internet? It's zero. That's how big the market for slow inference will be.”

Andrew Feldman May 21, 2026 ▶ 4:34 The Story Behind Cerebras’ $63 Billion IPO with Founder and CEO Andrew Feldman

Assertion Supported

Feldman: Cerebras is the only pure-play public AI company

“We would be the first and only, for a period of time, AI pure play. We are the only company that, that you can, a hundred percent of the revenue, this exact market. There's no gaming, there's no graphics, there's no PC, that, this is it.”

Andrew Feldman May 21, 2026 ▶ 22:39 The Story Behind Cerebras’ $63 Billion IPO with Founder and CEO Andrew Feldman

Assertion Supported

Feldman: Cerebras signed a $20B+ OpenAI deal in 4.5 weeks

“For a, 20 plus billion dollar deal to do it in four and a half weeks was exceptional.”

Andrew Feldman May 21, 2026 ▶ 25:00 The Story Behind Cerebras’ $63 Billion IPO with Founder and CEO Andrew Feldman

Assertion Supported

Feldman: Cerebras clusters scale linearly up to 64 nodes

“The cluster we build keeps the parameters off chip in a parameter store, and it streams them in, and the result of this architecture is that, that we run strictly data parallel, which means even in a 64 node cluster, you run the exact same configuration on eac…”

Andrew Feldman Sep 7, 2023 ▶ 7:45 No Priors Ep. 31 | With Cerebras CEO Andrew Feldman

Assertion Supported

Feldman: Cerebras can run a trillion-parameter model on a single system

“And that was an idea that came from supercomputing that we knew really well, that we could organize this so you could run an arbitrarily large, now a trillion parameter network on a single system.”

Andrew Feldman Sep 7, 2023 ▶ 9:06 No Priors Ep. 31 | With Cerebras CEO Andrew Feldman

Feldman: Cerebras is 15 to 20 times faster than GPUs at inference

More from Andrew Feldman

Turn any episode into a week of clips.