---
title: "Cerebras moves prompt processing to AMD GPUs, claiming 5x tokens per megawatt in production"
description: "With Q4 cloud revenue guided flat because Cerebras has no spare data center capacity, more tokens per megawatt is the only cloud lever it controls until new power lands from mid-2027. Whether its self-measured 5x holds decides how CBRS grows in between."
publisher: "The Inference"
section: "Chips"
published: 2026-11-18T19:12:59.671Z
modified: 2026-11-18T19:12:59.671Z
canonical: https://theinference.org/article/cerebras-moves-prompt-processing-to-amd-gpus-claiming-5x-tokens-per-megawatt-in-production
language: en
keywords: "Cerebras, AMD, AI Inference, Data Center, Short Sellers"
---

# Cerebras moves prompt processing to AMD GPUs, claiming 5x tokens per megawatt in production

> With Q4 cloud revenue guided flat because Cerebras has no spare data center capacity, more tokens per megawatt is the only cloud lever it controls until new power lands from mid-2027. Whether its self-measured 5x holds decides how CBRS grows in between.

Cerebras said on November 17 that disaggregated inference is running in production in its Inference Cloud: [AMD](https://theinference.org/markets/companies/amd) Instinct GPUs process the prompt, its CS-3 wafer-scale systems write the response, and the company measured five times the throughput per megawatt of a wafer-only setup, a production number rather than the early-test claim of early October [1].

The launch matters this month because [Cerebras](https://theinference.org/markets/companies/cerebras-systems) has run out of power to sell. Last week, CFO Robert Komin [guided fourth-quarter cloud revenue roughly flat](https://finance.yahoo.com/markets/stocks/articles/cerebras-cbrs-q3-2026-earnings-150350547.html) against the third quarter's $145 million: the ramp is "limited entirely by our data center capacity" [2]. Until new megawatts arrive, the only thing Cerebras can sell more of is tokens per megawatt.

## Each chip now does the half of inference it is built for

Every request has two stages. Prefill reads the prompt, a parallel compute job that suits GPUs. Decode emits tokens one at a time, rereading the model from memory at each step, and Cerebras holds model weights in on-chip SRAM for exactly that bandwidth job [1]. Handing prefill to AMD frees wafer capacity for token generation [1]. [AMD announced the joint design in July](https://ir.amd.com/news-events/press-releases/detail/1293/amd-and-cerebras-announce-industry-leading-ultra-low-latency-and-high-throughput-ai-inference-solution) and said Cerebras would deploy its Helios racks in its own data centers [3], so the company built on rejecting GPUs now runs its cloud on both.

## The design targets the constraint that is capping the P&L

New capacity starts arriving in December, and the bulk of about 700 gross megawatts under contract lands only from mid-2027 [2][4]. [Most of the $26.1 billion backlog sits with one customer, OpenAI](https://theinference.org/article/cerebras-faces-its-november-unlock-down-72-with-a-25-4-billion-backlog-resting-mostly-on-openai), under a take-or-pay contract for 750 megawatts [2]. Cerebras is behind on the first tranches and owes delay credits, and CEO Andrew Feldman said capacity was too thin for a broad rollout of [OpenAI](https://theinference.org/markets/companies/openai)'s Ultrafast mode [2]. On the call he framed the new design in the same terms, as a way to produce "even more high-speed tokens when high-speed tokens are constrained" [2].

*Chart: **Cerebras now spends more than twice its revenue building capacity** Column chart of Cerebras quarterly revenue against capital spending, in millions of dollars. Capital spending rises from $98 million in Q1 2025 to $497 million in Q3 2026, while revenue rises from $100 million to $211 million over the same reported quarters.*

*Cerebras Systems revenue and capital spending by quarter, in USD millions, from its SEC filings. The filings retrieved report no figures for Q4 2025, so that quarter is left blank. Source: [8]. [8]*

## The gain is self-measured, and the durability question stays open

The 5x figure is Cerebras' own, measured on November 14 on a single shared endpoint running the open GPT-OSS-120B model against its own baseline [1]. The open question is whether it survives partners' accelerators and workloads as the design rolls toward hundreds of megawatts by end-2027, with [AWS](https://theinference.org/markets/companies/amazon), [Fireworks AI](https://theinference.org/markets/companies/fireworks-ai), Gimlet Labs and General Compute in early access [1].

Culper Research's [October 28 short report](https://finance.yahoo.com/markets/stocks/articles/cerebras-sinks-10-short-seller-173139342.html) claimed [the wafers wear out in about two years](https://theinference.org/article/cerebras-s-customer-pays-3-to-5x-for-2-000-tokens-a-second-culper-research-says-the-wafers-fail-in-2-years) and noted that Cerebras lengthened the systems' assumed useful life between draft IPO filings [5]. Offloading prefill asks less compute of each wafer; whether that touches the durability claim, management has not said, and practitioners were still arguing it in a [92-point Hacker News thread](https://news.ycombinator.com/item?id=50540472) on the launch [6].

## The number to watch is cloud revenue per megawatt

Greenoaks, an $18 billion fund, sold its Cerebras stake in the third quarter [7]. At about 93 times forward earnings, [CBRS](https://theinference.org/markets/companies/cerebras-systems) needs growth that cannot wait for concrete [7]. The number to watch through the first half of 2027 is cloud revenue per megawatt on the fleet Cerebras already runs: if even part of the gain is real, the flat fourth quarter was a power problem, not a demand problem, and making scarce power produce more tokens is exactly the work the market is paying this company to do [7].

## Takeaway

Cerebras has put AMD GPUs in front of its wafer-scale systems in production, and its self-measured 5x gain in tokens per megawatt is the only cloud lever it controls until new capacity lands from mid-2027.

## Sources

1. [Unite.AI, 'Cerebras Puts Disaggregated Inference Into Production With GPUs', 18 November 2026](https://www.unite.ai/cerebras-puts-disaggregated-inference-into-production-with-gpus/)
2. [The Motley Fool via Yahoo Finance, 'Cerebras (CBRS) Q3 2026 Earnings Call Transcript', 13 November 2026](https://finance.yahoo.com/markets/stocks/articles/cerebras-cbrs-q3-2026-earnings-150350547.html)
3. [AMD Investor Relations, 'AMD and Cerebras Announce Industry-Leading Ultra-Low-Latency and High Throughput AI Inference Solution', 23 July 2026](https://ir.amd.com/news-events/press-releases/detail/1293/amd-and-cerebras-announce-industry-leading-ultra-low-latency-and-high-throughput-ai-inference-solution)
4. [RCR Tech, 'Cerebras secures 700 MW of AIDC capacity as it expands global footprint', 18 November 2026](https://rcrtech.com/ai-infrastructure-news/cerebras-700-mw-aidc/)
5. [Yahoo Finance, 'Cerebras Sinks 10% as Short Seller Culper Claims Its Wafers Break Down After Just ~2 Years', 29 October 2026](https://finance.yahoo.com/markets/stocks/articles/cerebras-sinks-10-short-seller-173139342.html)
6. [Hacker News, 'Cerebras Disaggregated Inference' discussion, 18 November 2026](https://news.ycombinator.com/item?id=50540472)
7. [Yahoo Finance, 'An $18 Billion Fund Dumps These 2 Stocks', 17 November 2026](https://finance.yahoo.com/markets/stocks/articles/18-billion-fund-dumps-2-145830408.html)
8. [Cerebras Systems, quarterly figures from its SEC filings, retrieved 18 November 2026](https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0002021728)


---

Cerebras moves prompt processing to AMD GPUs, claiming 5x tokens per megawatt in production — The Inference. Canonical: https://theinference.org/article/cerebras-moves-prompt-processing-to-amd-gpus-claiming-5x-tokens-per-megawatt-in-production
