---
title: "Cognition's $48 billion Devin runs on Nvidia's newest racks, rented from CoreWeave"
description: "Devin just took the first production slot on Nvidia's new Vera Rubin racks while doubling Cognition's run rate to $1 billion in September. The model underneath is built on weights Cognition does not own, priced under a frontier that keeps cutting."
publisher: "The Inference"
section: "Compute"
published: 2026-10-09T15:05:18.665Z
modified: 2026-10-09T15:05:18.665Z
canonical: https://theinference.org/article/cognition-s-48-billion-devin-runs-on-nvidia-s-newest-racks-rented-from-coreweave
language: en
keywords: "AI Agents, Coding, Data Centers, Benchmarks, Valuations"
---

# Cognition's $48 billion Devin runs on Nvidia's newest racks, rented from CoreWeave

> Devin just took the first production slot on Nvidia's new Vera Rubin racks while doubling Cognition's run rate to $1 billion in September. The model underneath is built on weights Cognition does not own, priced under a frontier that keeps cutting.

## First in line on Nvidia's newest racks

The first production workloads anywhere on [Nvidia](https://theinference.org/markets/companies/nvidia)'s Vera Rubin NVL72 [went live on September 30](https://finance.yahoo.com/technology/ai/articles/coreweave-delivers-nvidia-vera-rubin-130000022.html), and the tenant is not a frontier lab [1]. It is Devin, the coding agent from [Cognition](https://theinference.org/markets/companies/cognition), which runs training, reinforcement learning and inference on rented [CoreWeave](https://theinference.org/markets/companies/coreweave) racks and scaled from bridge capacity to thousands of GPUs in under nine months [1].

## A run rate that doubled in four months

Cognition was on track in September to generate [$1 billion in annualized revenue](https://www.bloomberg.com/news/articles/2026-09-25/ai-coding-startup-cognition-hits-1-billion-in-annualized-revenue), roughly double the $492 million of May, on a figure the company declines to confirm; it had said that month its run rate topped $900 million [2]. Weeks earlier it closed $2 billion at a $48 billion valuation, three months after raising $1 billion at $26 billion [2][3][4]. Nvidia is both a backer and a customer, alongside Citigroup and Mercedes-Benz [2][3].

## Free weights, one reinforcement-learning run

SWE-2, the model behind Devin since September 11, is post-trained from Moonshot's open-weight Kimi K3, 2.8 trillion parameters, with a reinforcement-learning run Cognition says adds five to six points [5]. That puts it within a point of Fable 5.1 on FrontierCode 1.1 Main, at a cost Cognition puts 64% lower, about a quarter of GPT-6 Astra [5].

Provenance is the catch: Cognition wrote the benchmark, filled in missing competitor scores itself, and no independent test of SWE-2 exists [5]. There is no printed per-task price and no API; it runs only inside Devin [5].

*Chart: **Within a point of Fable 5.1, at a claimed 64% lower cost** Bar chart of FrontierCode 1.1 Main pass rates: GPT-6 Astra 53.3%, Fable 5.1 50.9%, SWE-2 50.0%, Grok 4.6 48.0%, GPT-5.6 Sol 47.5%, Kimi K3 44.2%, SWE-1.7 42.0%*

*Pass rate on FrontierCode 1.1 Main, a benchmark Cognition's researchers wrote; best score across effort settings as published by Cognition on September 11, 2026. No independent test of SWE-2 exists. [5]*

## The landlord gets paid either way

CoreWeave, listed on Nasdaq, [booked $2.58 billion of revenue in the June quarter](https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001769628), up 112% year on year, while spending $6.4 billion on capex in the same three months [6]. Cognition's own engineers [measured up to 4.8 times the total token throughput for SWE-2 inference on Rubin racks against a GB200 baseline](https://theinference.org/article/cognition-s-4-8x-benchmark-proves-throughput-not-cheaper-agent-work) [1]; CoreWeave separately claims 10 times the tokens per megawatt, the number that binds when power, not floor space, caps a data hall [1][7]. Whoever wins the model layer, the rack owner bills them all.

*Chart: **Capex has exceeded revenue in every one of CoreWeave's ten reported quarters** Column chart of CoreWeave quarterly revenue and capital spending from Q1 2024 to Q2 2026, with capex above revenue in every quarter and both climbing steeply into 2026*

*CoreWeave quarterly revenue and capital expenditure in US dollars, from its SEC filings, Q1 2024 to Q2 2026. [6]*

## The discount is racing the price cuts

Above Cognition, the frontier is getting cheaper. [Anthropic](https://theinference.org/markets/companies/anthropic) [cut Haiku 5.5 prices 90% on October 7](https://venturebeat.com/technology/anthropic-launches-claude-haiku-5-5-with-90-api-price-reduction-matching-gpt-6-luna) to $0.10 per million input tokens, matching [OpenAI](https://theinference.org/markets/companies/openai)'s Luna, and its launch table scores that $0.10 model 46.4% on the same FrontierCode 1.1 main evaluation [8]. OpenAI halved Pro 200's usage allowance and now lets subscribers spend plan usage inside Devin, one of 16 partner tools [9]. Every cut narrows the discount that justifies Devin's price; on the hardest terminal benchmark, SWE-2 already trails GPT-6 Astra 27.3% to 57.9% [5].

## What would settle it

Watch for one independent SWE-2 benchmark, one printed per-task price, and the first sign of frontier cuts in Cognition's renewal rates. The talent arbitrage continues regardless: Cognition's former chief of staff raised $60 million on October 8 for the agent startup Hone, with Cognition itself a customer [10]. Free Chinese weights, one reinforcement-learning run, rented American racks: that is the whole cost structure, and the exposure sits with the secondary buyers now marking Cognition at $48 billion.

## Takeaway

Cognition's $48 billion Devin is first in production on Nvidia's Vera Rubin racks, but its cheap SWE-2 model rests on unowned weights, a self-published benchmark and a price discount that frontier cuts are eroding.

## Sources

1. [Business Wire via Yahoo Finance, CoreWeave Delivers NVIDIA Vera Rubin NVL72 Performance at Production Scale, Starting With Cognition, 30 September 2026](https://finance.yahoo.com/technology/ai/articles/coreweave-delivers-nvidia-vera-rubin-130000022.html)
2. [Bloomberg, AI Coding Startup Cognition Hits $1 Billion in Annualized Revenue, 25 September 2026](https://www.bloomberg.com/news/articles/2026-09-25/ai-coding-startup-cognition-hits-1-billion-in-annualized-revenue)
3. [GoEarlyBirdLab, Devin, the AI Software Engineer: Inside Cognition's $48 Billion Bet on Autonomous Coding Agents, 6 October 2026](https://goearlybirdlab.com/2026/10/06/devin-the-ai-software-engineer-inside-cognitions-48-billion-bet-on-autonomous-coding-agents/)
4. [TechCrunch, Cognition's Scott Wu says AI coding agents shouldn't replace humans, 29 May 2026](https://techcrunch.com/2026/05/29/cognitions-scott-wu-says-ai-coding-agents-shouldnt-replace-humans/)
5. [hamirev, SWE-2 review: Cognition's cheaper coding model for Devin, 4 benchmarks and cost, 9 October 2026](https://hamirev.com/swe-2-review/)
6. [CoreWeave, quarterly figures from its SEC filings, retrieved 9 October 2026](https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001769628)
7. [The Model Press, CoreWeave Vera Rubin NVL72 goes live with Cognition, 30 September 2026](https://themodelpress.com/news/2026/10/coreweave-vera-rubin-nvl72-cognition)
8. [VentureBeat, Anthropic launches Claude Haiku 5.5 with 90% API price reduction, matching GPT-6 Luna, 7 October 2026](https://venturebeat.com/technology/anthropic-launches-claude-haiku-5-5-with-90-api-price-reduction-matching-gpt-6-luna)
9. [The Next Web, OpenAI halves Pro 200 usage and launches a $500 ChatGPT plan at DevDay, 29 September 2026](https://thenextweb.com/news/openai-devday-pro-200-usage-cut-pro-500-plan)
10. [Crypto Briefing, Hone raises $60 million to build AI agents for business operations, 8 October 2026](https://cryptobriefing.com/hone-ai-agents-60-million-seed/)


---

Cognition's $48 billion Devin runs on Nvidia's newest racks, rented from CoreWeave — The Inference. Canonical: https://theinference.org/article/cognition-s-48-billion-devin-runs-on-nvidia-s-newest-racks-rented-from-coreweave