---
title: "Citigroup says 14,000 engineers on Devin cut vendor spend 30%; Cognition's $48 billion mark waits on December"
description: "Jane Fraser's disclosure is Devin's first bank-scale usage number, funded by other vendors' shrinking budgets. Cognition's $48 billion round now waits on a December run rate of $4 to $5 billion that exists only as a forecast."
publisher: "The Inference"
section: "Markets"
published: 2026-10-20T13:33:58.536Z
modified: 2026-10-20T13:33:58.536Z
canonical: https://theinference.org/article/citigroup-says-14-000-engineers-on-devin-cut-vendor-spend-30-cognition-s-48-billion-mark-waits-on-december
language: en
keywords: "Cognition, Citigroup, Devin, AI Coding Agents, Banking, CoreWeave"
---

# Citigroup says 14,000 engineers on Devin cut vendor spend 30%; Cognition's $48 billion mark waits on December

> Jane Fraser's disclosure is Devin's first bank-scale usage number, funded by other vendors' shrinking budgets. Cognition's $48 billion round now waits on a December run rate of $4 to $5 billion that exists only as a forecast.

## Fourteen thousand engineers, and the vendor bill behind them

On [Citigroup's third-quarter call on October 13, 2026](https://finance.yahoo.com/markets/stocks/articles/citi-c-q3-2026-earnings-235924215.html), chief executive Jane Fraser gave Devin its first bank-scale number: about 14,000 of the bank's engineers use Cognition's coding agent, products are launching 30 to 40 percent faster, and "we've reduced our tech vendor spend by 30% year-over-year as a consequence, and that's still got a long way to run" [1][2][3]. [Citi](https://theinference.org/markets/companies/citigroup) is re-engineering about 100 manual processes with AI and pulling contracted work back in-house [1]. The causal chain is the chief executive's own, and no bank disclosure separates what Citi pays Devin from what it saves. The disclosure still matters: an agent becomes a budget line by shrinking other vendors' lines.

## The $48 billion is priced on December

Cognition, founder Scott Wu's applied AI lab, raised more than $2 billion at $48 billion on September 8, 2026, four months after a $26 billion round, on run-rate revenue it says rose from $492 million in May to almost $900 million [4]. It passed $1 billion on September 25, a company-reported figure [5]. [The Information expects $4 billion to $5 billion annualized by year-end](https://techcrunch.com/2026/09/08/cognition-hits-48b-valuation-signaling-investors-believe-ai-coding-is-far-from-a-winner-take-all-market/) [4]. Cursor, the comparable exit, sold to [SpaceX](https://theinference.org/markets/companies/spacex) at $60 billion on annualized revenue that had surpassed $2 billion: under 30 times, against about 48 times the $1 billion print for Cognition's September mark [4].

*Chart: **Cognition's stated run rate doubled twice since May; December is a forecast, not a print** Four columns showing Cognition's annualized run-rate revenue: $0.49B in May 2026, $0.9B on September 8, $1.0B on September 25, and a forecast $4 to $5B in December.*

*Annualized run-rate revenue in billions of US dollars as stated by Cognition; the September 8 figure is the company's 'almost $900 million'. The December bar is The Information's forecast, drawn as an estimate, not a company print. Sources: TechCrunch, September 8, 2026; Unite.AI, September 25, 2026. [4][5]*

## Rationing by output, while the rent runs

October's other Cognition news runs the other way. [Higher Devin usage limits now depend on estimated engineering hours returned per dollar of session cost](https://www.tokenpost.com/news/technology/34146), validated at a 0.74 log-scale correlation across 233 held-out sessions [6]. The [Nvidia](https://theinference.org/markets/companies/nvidia) cluster Cognition rents from [CoreWeave](https://theinference.org/markets/companies/coreweave), the Nasdaq-listed AI cloud, costs hundreds of millions a year, and 2026 burn could reach $800 million [4]. [OpenAI's $20 ChatGPT login now meters Devin's OpenAI calls](https://theinference.org/article/openai-s-20-chatgpt-login-is-now-the-meter-for-devin-and-15-other-developer-tools), with nothing disclosed between the companies [7]. The landlord is repricing too: CoreWeave bonds issued in June at 9.675 percent [traded near 15 percent yields in the week to October 15](https://www.msn.com/en-us/news/other/coreweave-bond-rout-tests-markets-math-on-ai-buildout/ar-AA2emqm6), and the shares fell 15.1 percent, even as Cognition's own engineers measured 4.8 times the token throughput of a GB200 baseline running [the first production workloads on Nvidia's Vera Rubin racks](https://theinference.org/article/cognition-s-48-billion-devin-runs-on-nvidia-s-newest-racks-rented-from-coreweave) [8][9][10].

## Two readings decide it

The bull case is arithmetic: December at $4 billion to $5 billion compresses the $48 billion to 10 or 12 times revenue, below what [Cursor](https://theinference.org/markets/companies/anysphere) carried, though that forecast is The Information's, not Cognition guidance [4]. Watch the December print, and the first enterprise contract [Chris Degnan, Cognition's new chief revenue officer from Snowflake](https://theinference.org/article/cognition-hires-snowflake-s-chris-degnan-as-revenue-chief-with-devin-s-run-rate-at-1-billion), signs without a consultancy on it; [Cognizant](https://theinference.org/markets/companies/cognizant) has trained more than 3,000 associates on Devin and owns most delivery today [11]. The first receipt came out of a bank's vendor budget. The verdict comes from Cognition's own December number.

## Takeaway

Fraser's 30 percent vendor-spend cut is Devin's first bank-scale number; Cognition's $48 billion round now hinges on a forecast $4 to $5 billion December run rate, not yet a print.

## Sources

1. [Motley Fool via Yahoo Finance, Citi (C) Q3 2026 Earnings Call Transcript, 13 October 2026](https://finance.yahoo.com/markets/stocks/articles/citi-c-q3-2026-earnings-235924215.html)
2. [Banking Dive, What Wells Fargo, Citi and Goldman Sachs' CEOs are saying about AI, 15 October 2026](https://www.bankingdive.com/news/ai-citi-fraser-wells-fargo-scharf-goldman-sachs-solomon-morgan-stanley-pick-bofa-moynihan-jdimon/833087/)
3. [American Banker, Citi hires Google's director of engineering, shuffles AI leaders, 19 October 2026](https://www.americanbanker.com/news/citi-hires-googles-director-of-engineering-shuffles-ai-leaders)
4. [TechCrunch, Cognition hits $48B valuation, signaling investors believe AI coding is far from a winner-take-all market, 8 September 2026](https://techcrunch.com/2026/09/08/cognition-hits-48b-valuation-signaling-investors-believe-ai-coding-is-far-from-a-winner-take-all-market/)
5. [Unite.AI, Cognition Says Annualized Revenue Run Rate Has Passed $1B, 25 September 2026](https://www.unite.ai/cognition-says-annualized-revenue-run-rate-has-passed-1b/)
6. [TokenPost, Cognition Links Devin Usage Limits to Estimated Engineering Output, October 2026](https://www.tokenpost.com/news/technology/34146)
7. [The Inference, OpenAI's $20 ChatGPT login is now the meter for Devin and 15 other developer tools, 10 October 2026](https://theinference.org/article/openai-s-20-chatgpt-login-is-now-the-meter-for-devin-and-15-other-developer-tools)
8. [Business Wire via StreetInsider, CoreWeave Delivers NVIDIA Vera Rubin NVL72 Performance at Production Scale, Starting With Cognition, 30 September 2026](https://www.streetinsider.com/Business+Wire/CoreWeave+Delivers+NVIDIA+Vera+Rubin+NVL72+Performance+at+Production+Scale%2C+Starting+With+Cognition/27126145.html)
9. [Reuters via MSN, CoreWeave bond rout tests market's math on AI buildout, 16 October 2026](https://www.msn.com/en-us/news/other/coreweave-bond-rout-tests-markets-math-on-ai-buildout/ar-AA2emqm6)
10. [The Inference, Cognition's $48 billion Devin runs on Nvidia's newest racks, rented from CoreWeave, 9 October 2026](https://theinference.org/article/cognition-s-48-billion-devin-runs-on-nvidia-s-newest-racks-rented-from-coreweave)
11. [The Inference, Cognition hires Snowflake's Chris Degnan as revenue chief with Devin's run rate at $1 billion, 15 October 2026](https://theinference.org/article/cognition-hires-snowflake-s-chris-degnan-as-revenue-chief-with-devin-s-run-rate-at-1-billion)


---

Citigroup says 14,000 engineers on Devin cut vendor spend 30%; Cognition's $48 billion mark waits on December — The Inference. Canonical: https://theinference.org/article/citigroup-says-14-000-engineers-on-devin-cut-vendor-spend-30-cognition-s-48-billion-mark-waits-on-december