EY Cut AI Cost 15x With Nvidia Nemotron on Microsoft Foundry

EY adapted Nvidia’s open Nemotron models in Microsoft Foundry and cut cost by 15x against proprietary frontier models, Nvidia said on 19 November at Microsoft Ignite.

On the same task, classifying corporate purchasing records, accuracy on EY’s evaluation dataset improved by 55%, the post says.

Business Pill · GPUs Explained

An evergreen Business Engineer explainer on GPUs. It teaches the general idea only and does not discuss EY or Nemotron.

The key insight: EY’s tuned Nemotron model cost 15x less than proprietary frontier models on purchasing-record classification, as Nvidia reports it.

What EY Did

EY used Programmed Training, Microsoft’s serverless post-training offering in Foundry, to tune Nemotron for its purchasing data. Foundry manages the infrastructure; developers control the post-training workflow, Nvidia says.

Nvidia calls the result a demonstration of “how enterprises can customize open models for specialized business tasks”. It did not name the frontier models EY compared against.

EY tuned Nvidia's open Nemotron models in Microsoft Foundry for purchasing-record classification: 15x lower co
EY tuned Nvidia’s open Nemotron models in Microsoft Foundry for purchasing-record classification: 15x lower cost and 55% better accuracy, per Nvidia.

What Else Nvidia and Microsoft Announced

Azure virtual machines on Vera Rubin NVL72 racks are expanding, and the systems are “running in Azure data centers”, the post says. The companies are also extending confidential computing to rack-scale AI hardware.

The Nemotron 3.5 lineup comes to Foundry Managed Compute: Super for long-running agents, Lightning for fast responses. Bolt and Transcribe are “coming soon”.

Azure Local adds RTX PRO 6000 Blackwell GPUs and Jetson Thor and Orin edge platforms. Windows 365 Cloud PCs get Direct Stream, with public preview planned for early 2027.

Microsoft as a Customer

Microsoft AI uses Nvidia Dynamo for reinforcement-learning rollout inference during post-training. Microsoft also runs RTX PRO 6000 Blackwell GPUs for Bing search models, the post says.

In MLPerf Training v6.1, Azure scaled to 10,240 GB200 GPUs on DeepSeek-V3, the round’s largest scale. Its GB300 systems led cloud submissions on DeepSeek-R1 interactive throughput by over 10%, per Signal65.

The Structural Read

As we read it, Nvidia is selling open models as the cheap tier of enterprise AI, with Microsoft’s Foundry as the distribution channel.

As we read the Dynamo and Bing lines, Microsoft is a heavy Nvidia customer for its own models as well as a reseller of Nvidia capacity.

NVIDIA blog, 19 Nov 2026

“how enterprises can customize open models for specialized business tasks”

Three Implications

FOR ENTERPRISES EY’s reported result shows a specialised task tuned on open models at a fraction of frontier-model cost.

FOR MICROSOFT Foundry becomes the place where Nvidia’s open models are tuned and deployed.

FOR MICROSOFT AS A BUYER Microsoft AI uses Nvidia Dynamo in post-training and RTX PRO 6000 GPUs for Bing models, per Nvidia.

What Is Not Established

Nvidia gave no prices, no Azure Vera Rubin capacity figures and no dollar size for EY’s workload. The 55% gain is on EY’s own evaluation dataset; both figures are as reported by Nvidia, not independently tested.

This Week In The AI Supercycle

AI Lost Its Central Banker, the Lab Oligopoly, and the Future of Open Source

This week’s issue: the Nvidia leadership change, read as the AI build-out losing its central banker: who now finances the GPU landlords, backstops the data centres and sponsors open models, and why two labs are becoming the bank.

Read the full issue →

Business Engineer Framework

The AI Supercycle

The Business Engineer map of where value pools in the AI stack.

Read the Map of AI →

The Bottom Line

Alongside Vera Rubin on Azure, Nvidia’s Ignite post puts a customer number on open models: EY’s tuned Nemotron at 15x lower cost than proprietary frontier models, as Nvidia reports it.

97,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.

A note on sourcing. Every fact in this piece is from Nvidia’s 19 November blog post for Microsoft Ignite, including EY’s cost and accuracy results, the MLPerf v6.1 figures and the Signal65 comparison. We have not seen EY’s or Microsoft’s own account of the EY work.

Sources: NVIDIA, ‘NVIDIA and Microsoft Expand Accelerated Computing From Azure to Windows PCs’, 19 Nov 2026

Scroll to Top
FourWeekMBA