Verified
AMD
Scaling- Semiconductors
- ~31,000 employees
- Aria AI Suite, built by IT
- 30% software productivity gain (Aug 2026)
- No user count
AMD deploys AI inside the company through platforms its IT organization built and runs partly on AMD hardware: the Aria AI Suite, with more than a dozen models, and ACE, an agent in Microsoft Teams. AMD engineers also use OpenAI's Codex, Anthropic's Claude Code, GitHub Copilot and Synopsys design tools. Chief software officer Andrej Zdravkovic said in Aug 2026 that AI had raised AMD's software productivity 30%. AMD has published no count of employees using its AI tools.
amd.comTheir approach to AI
- The platform
- AMD IT built the Aria AI Suite, a web platform for chat across more than a dozen models plus translation, text-to-voice and flowchart tools. AMD IT also built ACE, an agent in Microsoft Teams that handles HR, IT, engineering knowledge and compute requests (Sep 2026).
- The models
- AMD's Aria AI Suite offers cloud-hosted frontier models and open-source models running in AMD's own data centers on Instinct accelerators. AMD engineers also use OpenAI's Codex, Anthropic's Claude Code, GitHub Copilot and Synopsys AI design tools (2026).
- The reach
- AMD has published no count of employees using Aria or ACE. GitHub says more than 20,000 AMD engineers work on GitHub Enterprise, and AMD says AI wrote more than 20% of its shipped source code by early 2026.
- The catch
- AMD's reported AI returns come from 2 internal units, IT's engineering compute grid and software development, and AMD measured and published both itself. AMD reports no firm-level AI figure and no adoption count, and its chip design gains carry no cost figure.
Snapshot
- Employees
- ~31,000
- Dec 27, 2025, up from about 28,000 a year earlier (2025 Form 10-K)
- Technology budget
- Not disclosed
- No IT or AI budget published; R&D expense was $8.1B in 2025, up 25%
- AI/ML headcount
- Not disclosed
- About half of AMD engineers are software engineers (Andrej Zdravkovic, Apr 2025)
- Who owns AI
- Hasmukh Ranjan, CIO
- AMD IT leads internal AI adoption; Ranjan and Andrej Zdravkovic report AI results to Lisa Su
- Platform
- Aria AI Suite and ACE, built in-house
- ACE is an agent in Microsoft Teams; both named by AMD IT in Sep 2026
- Models
- More than a dozen models
- Cloud-hosted frontier models and open-source models on AMD Instinct; Codex, Claude Code and GitHub Copilot for engineers
- Reach
- No user count published
- More than 20,000 engineers on GitHub Enterprise, according to GitHub
- Use cases
- 100+ documented
- Internal catalog, Jul 2025; deployments in IT, engineering, HR, procurement and finance
Key findings
AMD chief software officer Andrej Zdravkovic wrote in IEEE Spectrum on Aug 17, 2026 that AI had delivered a 30% overall productivity gain in AMD software development, beating AMD's Apr 2025 target of 25% over 2 to 3 years. Zdravkovic also wrote that AI-generated code passed 20% of AMD's shipped source code at the start of 2026.
AMD IT said on Mar 12, 2026 that AI-powered management of its engineering compute grid saved an average 20% to 25% of operating and maintenance costs in 2025, through improved utilization and automated processes.
AMD IT built its own employee AI platforms: the Aria AI Suite and ACE, an agent in Microsoft Teams (Sep 2026). In Aria, AMD staff choose among more than a dozen models, from cloud-hosted frontier models to open-source models in AMD's data centers on Instinct accelerators.
AMD's software agents fixed 75% of reported issues in Radeon Software eXperience by Jun 2026, up from 6% when AMD started using them in Oct 2025. AMD improved the agents with a learning loop that refined their objectives instead of retraining models.
AMD has published no count of employees using its AI tools. Lisa Su said in Jan 2026 that AMD is hiring heavily, favoring people who are AI forward. Su said AI raises the number of products AMD can bring up at once rather than replacing staff.
Where AMD sits
- Stage
- Scaling
AMD is at Scaling because it has reported returns it credits to AI in 2 functions within the last 18 months. AMD IT said on Mar 12, 2026 that AI-powered management of its engineering compute grid cut operating and maintenance costs by an average 20% to 25% in 2025. Chief software officer Andrej Zdravkovic wrote on Aug 17, 2026 that AI had delivered a 30% overall productivity gain in AMD software development, past a 25% target set for 2 to 3 years. On rollout, AMD runs company-wide AI platforms built by IT under CIO Hasmukh Ranjan: the Aria AI Suite and the ACE agent in Microsoft Teams. AMD says Ranjan and Zdravkovic report AI results to Lisa Su against targets, but AMD has not said how many employees use the platforms. AMD falls short of Embedded because it reports no firm-level AI effect on cost, headcount, revenue or margin, and no share of employees using AI each week.
How the stages are defined - Ownership
- CIO Hasmukh Ranjan runs internal AI; Andrej Zdravkovic leads AI in software engineering
- Platform
- Hybrid: Aria AI Suite and ACE built by AMD IT; coding agents bought
- Models
- More than a dozen models in Aria; Codex, Claude Code, GitHub Copilot
- Hosting
- Own data centers on AMD Instinct for open-source models; cloud-hosted frontier models (providers not named); Kore.ai HR agent on AWS
- Reach
- No company-wide user count published
How they deploy
Customer zero on its own hardware
AMD IT calls itself AMD's customer zero: CIO Hasmukh Ranjan's organization leads internal AI adoption and runs it on AMD's own products. AMD IT uses EPYC and Ryzen PRO processors, Instinct GPUs, Versal adaptive SoCs, Pensando DPUs and the ROCm software stack. AMD IT also runs unreleased products and reports their performance back to AMD's product teams. AMD's case study says AMD built its AI transformation strategy on AMD compute.
A maturity model with business metrics
AMD ranks internal AI work against a 4-phase maturity framework: Assist, Action, Automate and Autonomous. Functional teams and AMD IT pick the workloads with the highest expected return, rank them across the company and define business metrics for each. Chris Wire, then AMD's vice president of business applications, told CIO.com in Jul 2025 that AMD evaluates the cost, benefits and suitability of each AI project before proceeding.
Many models in a single platform
AMD IT wrote in Sep 2026 that AMD does not treat enterprise AI as a one-model environment. The Aria AI Suite lets employees pick among more than a dozen models, and AMD staff also use commercial tools for collaboration, software development and code generation. AMD IT describes its job at this stage as coordinating those tools and connecting AI to enterprise knowledge and processes.
From assistants to autonomous processes
AMD IT's Sep 2026 plan, which it calls Click to, has AI agents run entire processes from one request. The plan covers marketing claims review, customer return merchandise authorizations, field support and the quarterly financial close. AMD IT says self-healing systems already detect, diagnose and resolve some IT issues without people.
Augment engineers, keep hiring
Lisa Su said in Jan 2026 that AI augments AMD's productivity in the number of products AMD can bring up at any given time and that AMD is hiring people who are AI forward. Andrej Zdravkovic wrote in Aug 2026 that AMD's goal is to equip its workforce with AI, not to reduce headcount. AMD IT made the same point in Sep 2026.
- Deployment model
- AMD runs internal AI as central platforms built by IT plus engineering-owned tooling. AMD IT, under CIO Hasmukh Ranjan, builds and runs the Aria AI Suite, the ACE agent in Microsoft Teams, the OPTIMA data intelligence platform and the AI champion community. AMD IT also works with functions such as HR, finance and procurement on their use cases. AI in software engineering, from code generation to agentic bug fixing, sits with chief software officer Andrej Zdravkovic's organization. AI in chip design, built on Synopsys tools, sits with engineering under Brian Amick and CTO Mark Papermaster. In 2024 a review committee of the CIO, the president, the chief software officer and the chief legal officer approved internal AI projects. AMD says Ranjan and Zdravkovic report AI initiatives to Lisa Su against targets.
- Model providers
- AMD's Aria AI Suite offers more than a dozen large language models, from cloud-hosted frontier models to open-source models in AMD's own data centers (Sep 2026). AMD has not named the frontier model providers. AMD software engineers run multi-agent workflows through OpenAI's Codex and Anthropic's Claude Code while AMD builds its own multi-agent systems (Aug 2026). For code review, AMD fine-tunes models on its own code base and documentation (Apr 2025). AMD made GitHub Copilot available to developers in Sep 2023, and GitHub says it worked with AMD to train Copilot on hardware description languages such as Verilog. In chip design AMD uses Synopsys reinforcement learning, generative and agentic AI. Stella Laurenzo, a senior director in AMD's AI Group, reported in Apr 2026 that Claude Code had regressed on complex engineering and that her team was moving to another provider for the time being.
- Data and infrastructure
- AMD runs open-source models for the Aria AI Suite in its own data centers on AMD Instinct accelerators. AMD IT's infrastructure combines EPYC and Ryzen PRO processors, Instinct GPUs, Versal SoCs and Pensando DPUs on the ROCm software stack. Hasmukh Ranjan's remit includes expanding cloud capacity and elastic computing for engineering. The Kore.ai HR agent runs on AWS, using Amazon EKS, Amazon S3 with encryption at rest, AWS KMS and Amazon Bedrock. AMD IT built OPTIMA, a data intelligence platform that turns enterprise data into knowledge graphs for agents, first for IT and now as a platform shared across AMD. AMD IT published OPTIMA's reference architecture on GitHub in Jul 2026. AIR, an AI researcher on the same platform, answers questions across IT, observability and finance data.
Platforms and internal tools
Aria AI SuiteNamed Sep 2026
Aria AI Suite is AMD's custom-built employee AI platform: AI Chat with a choice of more than a dozen models, plus translation, text-to-voice, flowchart generation and other content tools (Sep 2026). Aria matches the consolidated AI productivity suite AMD IT described in Sep 2025, which offered more than a dozen internal models, document and data sheet analysis, and thumbs-up and thumbs-down feedback.
- Models
- More than a dozen models, cloud-hosted or self-hosted on AMD Instinct
ACE (AMD Concierge for Employees)Named Sep 2026
ACE is AMD's internal agentic system in Microsoft Teams that connects employees to HR, IT, engineering knowledge and compute resources and gathers approvals from different systems in one place (Sep 2026). ACE matches the universal agent AMD IT described in Sep 2025, a supervisor bot in Teams that routes requests to IT, HR and finance agents by intent.
OPTIMA data intelligence platform and AIRJul 2026 (OPTIMA 2.0 described)
OPTIMA is AMD IT's open, modular data intelligence platform that turns enterprise data into contextual knowledge graphs for agent-driven automation across IT, security and the business. AMD published the reference architecture on GitHub. AIR (AI Researcher) is a conversational layer over the platform that suggests role-relevant questions and returns ranked, explainable recommendations (May 2026).
Software development agents2024
AMD began building AI systems for code generation, test automation, bug analysis and code review in 2024. AMD now runs agents at every stage of its software lifecycle, from triaging problem reports to preparing release summaries for engineers' approval. AMD engineers use Codex and Claude Code as agent harnesses alongside AMD's own multi-agent systems.
- Models
- OpenAI Codex, Anthropic Claude Code, in-house multi-agent systems
GitHub Enterprise and GitHub CopilotSep 2023 (Copilot)
AMD made GitHub Copilot available to developers in Sep 2023 and found after 6 months that just under 50% of Copilot users used it daily. GitHub says AMD has since consolidated more than 20,000 engineers on GitHub Enterprise. According to GitHub, AMD uses the Copilot coding agent to triage about 1 million nightly pre-silicon simulations and Copilot Autofix to patch legacy vulnerabilities.
- Users
- More than 20,000 engineers on GitHub Enterprise (GitHub, undated)
- Models
- GitHub Copilot, trained with AMD on hardware description languages
Kore.ai HR agent
AMD's generative AI HR agent, built with Kore.ai, answers role- and region-specific HR questions, completes high-volume HR transactions, lets managers approve requests in Microsoft Teams and escalates critical issues to HR staff. AMD's HR contact center has about 15 employees serving a workforce of more than 30,000.
Vendors and partners
MicrosoftSep 2023 (GitHub Copilot)
Microsoft Teams hosts ACE; Microsoft 365 Copilot training for staff; GitHub Enterprise and Copilot for engineers · AMD's AI champion community runs specialized sessions on Microsoft 365 Copilot (Sep 2025), and GitHub says more than 20,000 AMD engineers work on GitHub Enterprise.
OpenAI
Codex agent harness for AMD software engineers · Andrej Zdravkovic wrote in Aug 2026 that AMD uses multi-agent workflows extensively through Codex and Claude Code.
Anthropic
Claude Code agent harness for AMD software engineers · Stella Laurenzo of AMD's AI Group filed a public report in Apr 2026, based on 6,852 Claude Code sessions, that the tool had regressed on complex engineering work.
Synopsys
AI chip design tools: reinforcement learning, copilot assistance and agentic workflows · The World Economic Forum selected Synopsys and AMD for its MINDS program in Jan 2026 for AI in chip design across design, verification and signoff.
Kore.ai
Generative AI HR agent platform · Kore.ai gave AMD an AI Business Impact award for its global deployment of AI in HR.
Amazon Web Services
Hosts the Kore.ai HR agent, with Amazon Bedrock for conversational AI
Leadership
AI, data and platform leaders
Executives
AI org map
IT-run enterprise AI platforms; engineering and software organizations own AI in chip design and code
AMD's internal AI program is led by CIO Hasmukh Ranjan's IT organization, which builds the employee platforms and works with each function on use cases. Chief software officer Andrej Zdravkovic's organization runs AI across the software lifecycle, and engineering under Brian Amick and CTO Mark Papermaster applies AI to chip design and verification. In 2024 a review committee of the CIO, the president, the chief software officer and the chief legal officer approved internal AI projects. AMD says Ranjan and Zdravkovic report AI initiatives to Lisa Su against targets.
AMD has not published the size of its AI workforce. AMD IT, under CIO Hasmukh Ranjan, builds the employee platforms, the data platform and agents for HR, finance, procurement and support, and has added roles specific to AI compute. AI in software engineering sits under chief software officer Andrej Zdravkovic, and about half of AMD's engineers are software engineers. AI in chip design and verification sits with engineering leaders such as Brian Amick and Corporate Fellow Alex Starr. AMD IT also runs an AI champion community on a train-the-trainer model, with champions in each function.
Reporting lines
Chair and CEO
Information Technology
Hasmukh Ranjan, SVP and CIO
Also runs information security and indirect procurement; Ranjan reports on AI initiatives to Lisa Su, but his formal reporting line is not published
Information Security
Chief Information Security Officer (not named in the 10-K)
The 2025 Form 10-K says the chief information security officer reports to the CIO
Chief Technology Officer
Mark Papermaster, EVP and CTO
Technology and Engineering
Brian Amick, SVP
Centralized engineering, CPU core strategy and central design methodology; the line to the CTO is inferred, not published
GPU Technologies and Engineering Software
Andrej Zdravkovic, SVP and Chief Software Officer
Reporting line not published
AMD's leadership page lists Hasmukh Ranjan, Andrej Zdravkovic and Brian Amick among its senior vice presidents but does not publish reporting lines. The 2024 review committee included AMD's president, but AMD's leadership page, checked Nov 9, 2026, lists no president. AMD has not said who now sits on the committee.
Where the AI workforce sits
Software engineering
Estimate from Andrej Zdravkovic's Apr 2025 statement that about half of AMD engineers are software engineers; AMD gives no count of AI specialists.
~50%Hardware design, verification and other engineering
Estimate: the other half of AMD's engineers, where AI is used in design exploration, verification and simulation triage.
~50%
Named people
Use cases
AMD has documented more than 100 internal AI use cases (Jul 2025) across IT, engineering, HR, procurement, finance, sales and marketing. AMD credits AI with 2 business returns: a 20% to 25% average cut in operating and maintenance costs for IT's engineering compute grid in 2025, and a 30% overall productivity gain in software development (Aug 2026). AMD's chip design, verification, HR, customer service and finance use cases come with time savings, throughput measures or vendor figures, not cost or headcount, and AMD reports no firm-level AI return.
| Use case | Area | Status | ROI |
|---|---|---|---|
| AI-managed engineering compute grid AMD IT uses AI to detect when engineering compute resources need to be reallocated and act in real time, automating grid processes and letting users report issues to a chatbot that resolves many within minutes. | IT infrastructure | In production | Grid operating and maintenance costs down an average 20-25% in 2025 (Mar 2026) |
| AI across the software development lifecycle AMD runs agents for problem triage, debugging and code changes, unit and integration test generation, and release summaries for engineers' approval, using Codex, Claude Code and its own multi-agent systems. AI wrote more than 20% of AMD's shipped source code at the start of 2026, counting only code that passed review and testing. | Software engineering | In production | 30% overall software development productivity gain from AI (Aug 2026) |
| Agentic bug fixing in Radeon Software eXperience AMD began using AI agents to debug and fix reported issues in its Radeon Software eXperience graphics settings app in Oct 2025; the share of issues the agents resolved rose from 6% to 75% by Jun 2026. | Software engineering | In production | |
| AI in graphics software testing AMD trained convolutional neural networks on more than 20,000 defect-free and 2,000 distorted images to spot rendering artifacts and file bug reports. AMD also combines vision models and generative AI to play games automatically for functional testing. AMD said in Apr 2025 that AI testing cut manual test effort 15% and raised the scenarios tested 20%. | Software engineering | In production | |
| Pre-silicon simulation triage with GitHub Copilot GitHub says AMD uses the GitHub Copilot coding agent to analyze logs from about 1 million nightly pre-silicon simulations, match known issues and propose fixes, reducing manual morning triage to zero. | Chip design and verification | In production | |
| Synopsys AI design flows AMD uses Synopsys reinforcement learning, copilot assistance and agentic workflows across design, verification and signoff; the World Economic Forum said the approach doubled the speed of AMD chip design (Jan 2026). | Chip design and verification | In production | |
| AI in CPU architecture and verification AMD CPU engineers use AI to evaluate more design options, speed verification and narrow choices for the Zen architecture (Sep 2026). Mark Papermaster said in Jul 2026 that some design tasks that took many months now take weeks or days. | Chip design and verification | In production | |
| Automated vulnerability fixes GitHub says AMD runs a historical agent with GitHub Copilot Autofix that combs years of legacy code for dormant vulnerabilities, and that about 70% of vulnerability classes are now resolved automatically. | Security | In production | |
| ACE agent in Microsoft Teams ACE routes employee requests to HR, IT, finance, engineering knowledge and compute agents and consolidates approvals from several systems in Microsoft Teams. | Employee services | In production | |
| Aria AI Suite Aria gives AMD employees chat with more than a dozen models plus translation, text-to-voice, flowchart and document analysis tools in one web platform. | Employee services | In production | |
| HR agent AMD's Kore.ai HR agent answers HR questions and completes transactions for more than 30,000 people. Kore.ai reports an 80% cut in HR resolution time and 50% of inquiries moved to self-service; AMD has not published these figures itself. | HR | In production | |
| R&D tax documentation and finance automation AMD uses generative AI to find qualifying R&D activities and costs in thousands of pages of technical notes, cutting tax document preparation from weeks to hours. CIO.com reported in Jul 2025 that AI-driven automation delivers 15% productivity gains in AMD finance. | Finance | In production | |
| Agentic quarterly close AMD's Finance Modernization program is working toward digital agents that execute close activities, reconcile financial datasets, analyze results and run continuous automated audit and review under human oversight (Sep 2026). | Finance | Announced | |
| Click to Claims An AI review agent checks AMD marketing claims against the applicable guidelines before human reviewers see them, giving submitters faster feedback. | Marketing | In production | |
| Click to RMA A digital agent reviews failure logs and historical return rates for customer return merchandise authorizations and recommends decisions. The agent also automates part of the approval, generates return labels and keeps the customer informed. | Customer operations | In production | |
| Click to Support AMD describes agents that summarize a support case, propose a resolution and create the work order (Sep 2026). The agents then book the service appointment, match a technician and capture the outcome for reuse. | Customer operations | Announced | |
| Procurement AI orchestration AMD uses AI to extract data from quotations and order forms and pre-fill procurement intake. | Procurement | In production | |
| Self-healing IT and data intelligence AMD IT runs self-healing systems that detect, diagnose and resolve issues, and uses AIR on the OPTIMA platform to surface cost drivers, failure modes and next actions across IT, observability and finance data. | IT operations | In production | |
| Yield analytics and compute demand forecasting AMD uses machine learning to improve semiconductor manufacturing yield, which CIO.com reported in Jul 2025 can translate into millions of dollars, without giving a figure. AMD also uses AI-driven analytics to forecast engineering compute demand. | Manufacturing and planning | In production |
Adoption
AMD has spread AI from engineers outward. GitHub Copilot reached developers in Sep 2023, and by 2024 AMD had consolidated internal AI work around 2 audiences, engineers and everyone else. AMD IT added a universal agent and a consolidated AI suite by Sep 2025, then agentic processes in finance, marketing and customer operations in 2026. AMD has not published how many employees use Aria, ACE or any AI tool. The adoption figures AMD has given are partial. Just under 50% of Copilot users used it daily 6 months after launch (Apr 2025), and AI wrote more than 20% of shipped source code by early 2026. GitHub says more than 20,000 AMD engineers work on GitHub Enterprise. AMD IT wrote in Sep 2026 that AI is embedded in how work gets done across AMD and that its impact shows in AMD's top-level performance measures, without giving figures.
By the numbers
Rollout, first pilot to today
Sep 2023
AMD made GitHub Copilot available to its developers. After 6 months just under 50% of Copilot users used it daily, and 75% said they would use it more if suggestions fit AMD's code better.
May 2024
Hasmukh Ranjan said AMD had consolidated internal AI efforts about a year earlier around 2 audiences, engineers and everyone else, with a review committee and a test framework every project must pass before production.
Apr 2025
Andrej Zdravkovic set a target of a 25% software productivity gain from AI over 2 to 3 years and said AMD was training an in-house coding assistant on its own code.
Jul 2025
AMD's internal catalog documented more than 100 AI use cases, and one AMD hackathon produced 100 new AI ideas in a day, several of which reached production.
Sep 2025
AMD IT described a universal AI agent in Microsoft Teams, a consolidated AI productivity suite with more than a dozen models, and an AI champion community trained on a train-the-trainer model.
Oct 2025
AMD began using AI agents to fix reported issues in Radeon Software eXperience, resolving 6% at first.
Early 2026
AI-generated code passed 20% of AMD's shipped source code, counting only code that passed review and testing.
Jun 2026
AI agents resolved 75% of reported Radeon Software eXperience issues.
Sep 2026
AMD IT named its platforms ACE and the Aria AI Suite and described agents running marketing claims review, customer returns and, in development, field support and the financial close.
Economics & controls
Spend and ROI
Technology budget
AMD does not publish an IT or AI budget. AMD's research and development expense was $8.1 billion in 2025, up 25% from $6.5 billion in 2024, mainly from headcount added for its AI strategy, according to the 2025 Form 10-K. The R&D figure covers product development, not internal AI tools.
What AI itself costs
AMD has not disclosed what it spends on internal AI tools, models or tokens. AMD runs part of its internal inference on its own Instinct accelerators in its own data centers, and says each AI initiative must present a business case before it proceeds.
Reported ROI
AMD reports 2 returns it credits to AI, both for internal units and with no firm-level total. AMD IT said in Mar 2026 that its investment in AI-powered grid efficiency saved an average 20% to 25% on the engineering compute grid's operating and maintenance costs in 2025, through improved utilization and automated processes. Andrej Zdravkovic wrote in Aug 2026 that AMD achieved a 30% overall productivity gain in software development through AI. Zdravkovic noted that the share of AI-generated code is the one objective metric AMD has tracked from the start. AMD IT wrote in Sep 2026 that AI's impact appears in AMD's top-level performance and productivity measures, without figures.
Productivity markers
AMD's other AI figures measure time, throughput or quality rather than cost. Preparing R&D tax documents went from weeks to hours, and AI-driven automation gives 15% productivity gains in finance (CIO.com, Jul 2025). AI testing cut manual test effort 15% (Apr 2025), and agents resolve 75% of Radeon Software eXperience issues (Jun 2026). GitHub says Copilot Autofix spares AMD 10,000 engineer hours on vulnerability fixes, and Kore.ai reports an 80% cut in HR resolution time; neither figure comes from AMD. AMD says machine learning yield gains translate into millions of dollars, without a figure. Mark Papermaster said in Jul 2026 that AI cuts months from chip design schedules, without a cost figure.
Headcount
AMD had about 31,000 employees at Dec 27, 2025, up from about 28,000 a year earlier. AMD cut about 4% of its workforce under a restructuring plan in the fourth quarter of 2024, and the 10-K does not link the cut to internal AI use. AMD's IT Service Desk kept technician headcount flat as user volume grew, which AMD credits to automation and improved workflows (Feb 2026). Lisa Su said in Jan 2026 that AMD is hiring lots of people, and Andrej Zdravkovic wrote in Aug 2026 that the goal of AMD's AI program is not to reduce headcount.
Governance and risk
Review committee and test framework
AMD set up a review committee for internal AI use cases with the CIO, the president, the chief software officer and the chief legal officer. Hasmukh Ranjan said in May 2024 that every project must pass a test framework before production. AMD says Ranjan and Andrej Zdravkovic report AI initiatives to Lisa Su against targets.
Responsible AI program
AMD's Responsible AI program reviews internal AI applications for safety, ethics, regulatory compliance and trustworthiness (Sep 2025), and AMD's adoption playbooks embed data privacy, ethical review and operational resilience into each deployment.
Intellectual property
Andrej Zdravkovic wrote in Apr 2025 that AMD guards against 2 intellectual property risks from AI coding tools: using suggestions that infringe third-party IP, and disclosing AMD source code to public tools that may train on it. AMD keeps a human in the loop for AI-generated code and counts only code that passes all reviews and testing (Aug 2026).
Security and disclosed risks
AMD's 2025 Form 10-K says AMD uses AI tools to support internal functions and operations and that these systems are increasingly exposed to cyber threats, including malicious code introduced through AI-generated source code. AMD's chief information security officer reports to CIO Hasmukh Ranjan, who also leads information security. The 10-K lists responsible use of AI and the EU AI Act among AMD's risks.
Agents under policy
AMD IT says its finance agents operate within predefined policies and human oversight (Sep 2026), and that AI agents in its autonomous processes bring employees in when judgment is required.
Recent signals
Sep 10, 2026
AMD IT described ACE, its agentic employee concierge in Microsoft Teams, and the Aria AI Suite with more than a dozen cloud-hosted and self-hosted models. AMD IT also set out agentic Click to processes for claims, customer returns, support and the financial close.
Sep 2026
In AMD's Advanced Insights series, CTO Mark Papermaster and chief CPU architect Mike Clark said AMD engineers use AI to evaluate more Zen design options and speed verification; AMD gave no measurement of the time saved.
Aug 17, 2026
Andrej Zdravkovic wrote that AMD had passed its AI productivity target with a 30% overall software productivity gain. Zdravkovic added that AI produced more than 20% of shipped code by early 2026 and that agents fixed 75% of Radeon Software eXperience issues by Jun 2026.
Jul 22, 2026
AMD IT presented its enterprise AI journey, from experimentation to scaled deployment and data platform investment, in a breakout session at AMD Advancing AI 2026.
Jul 21, 2026
AMD IT published the reference architecture of OPTIMA, its data intelligence platform for agent-driven automation, on GitHub.
Jul 2026
Mark Papermaster told the RAISE Summit in Paris that AMD designs its chips with AI help and that productivity gains had jumped from the 10% range to far larger over the previous 6 months.
May 15, 2026
AMD IT described AIR, an AI researcher that turns IT, observability and finance data into ranked, explainable recommendations.
May 11, 2026
AMD IT published a playbook for its AI hackathons, held across several global sites with hundreds of participants.
Apr 2026
Stella Laurenzo, a senior director in AMD's AI Group, filed a public GitHub issue based on 6,852 Claude Code sessions saying the tool had regressed on complex engineering. Laurenzo said her team would use another provider until Anthropic fixed it.
Mar 12, 2026
AMD IT reported that AI-powered grid efficiency saved an average 20% to 25% on operating and maintenance costs of its engineering compute grid in 2025.
Feb 25, 2026
AMD IT said its service desk cut ticket hand-offs from 11% to 4% of tickets and kept technician headcount flat as users grew, crediting automation and improved workflows.
Feb 3, 2026
AMD's 2025 Form 10-K reported about 31,000 employees and said AMD uses AI tools to support internal functions and operations.
Jan 29, 2026
Synopsys said the World Economic Forum selected Synopsys and AMD for its MINDS program for AI in chip design, with AMD's Brian Amick saying Synopsys AI reduces design time for AMD engineers.
Jan 7, 2026
Lisa Su told CNBC at CES 2026 that AMD is hiring lots of people who are AI forward and that AI augments the number of products AMD can bring up at once.
Quick answers
- How does AMD deploy AI?
- AMD deploys AI inside the company through platforms its IT organization built and runs partly on AMD hardware: the Aria AI Suite, with more than a dozen models, and ACE, an agent in Microsoft Teams. AMD engineers also use OpenAI's Codex, Anthropic's Claude Code, GitHub Copilot and Synopsys design tools. Chief software officer Andrej Zdravkovic said in Aug 2026 that AI had raised AMD's software productivity 30%. AMD has published no count of employees using its AI tools.
- What does AMD use AI for?
- AMD has documented more than 100 internal AI use cases (Jul 2025) across IT, engineering, HR, procurement, finance, sales and marketing. AMD credits AI with 2 business returns: a 20% to 25% average cut in operating and maintenance costs for IT's engineering compute grid in 2025, and a 30% overall productivity gain in software development (Aug 2026). AMD's chip design, verification, HR, customer service and finance use cases come with time savings, throughput measures or vendor figures, not cost or headcount, and AMD reports no firm-level AI return.
- Who leads AI at AMD?
- AI at AMD is led by Hasmukh Ranjan (Senior Vice President and Chief Information Officer); Andrej Zdravkovic (Senior Vice President, GPU Technologies and Engineering Software, and Chief Software Officer); Brian Amick (Senior Vice President, Technology and Engineering). Ownership of AI on the org chart: CIO Hasmukh Ranjan runs internal AI; Andrej Zdravkovic leads AI in software engineering.
- What AI platform does AMD use?
- AMD IT built the Aria AI Suite, a web platform for chat across more than a dozen models plus translation, text-to-voice and flowchart tools. AMD IT also built ACE, an agent in Microsoft Teams that handles HR, IT, engineering knowledge and compute requests (Sep 2026).
- Which AI models does AMD use?
- AMD's Aria AI Suite offers cloud-hosted frontier models and open-source models running in AMD's own data centers on Instinct accelerators. AMD engineers also use OpenAI's Codex, Anthropic's Claude Code, GitHub Copilot and Synopsys AI design tools (2026).
- How many AMD employees use AI?
- AMD has published no count of employees using Aria or ACE. GitHub says more than 20,000 AMD engineers work on GitHub Enterprise, and AMD says AI wrote more than 20% of its shipped source code by early 2026.
- What return has AMD reported from AI?
- AMD is at Scaling because it has reported returns it credits to AI in 2 functions within the last 18 months. AMD IT said on Mar 12, 2026 that AI-powered management of its engineering compute grid cut operating and maintenance costs by an average 20% to 25% in 2025. Chief software officer Andrej Zdravkovic wrote on Aug 17, 2026 that AI had delivered a 30% overall productivity gain in AMD software development, past a 25% target set for 2 to 3 years. On rollout, AMD runs company-wide AI platforms built by IT under CIO Hasmukh Ranjan: the Aria AI Suite and the ACE agent in Microsoft Teams. AMD says Ranjan and Zdravkovic report AI results to Lisa Su against targets, but AMD has not said how many employees use the platforms. AMD falls short of Embedded because it reports no firm-level AI effect on cost, headcount, revenue or margin, and no share of employees using AI each week.
Open questions
- How many AMD employees use the Aria AI Suite, ACE and AI coding tools each week, and what share of AMD's roughly 31,000 staff is that?
- Which frontier model providers sit behind the Aria AI Suite, which open-source models run on AMD's own Instinct accelerators, and in which clouds?
- How does AMD measure its 30% software productivity gain, beyond the share of AI-generated code, and what does it mean for cost or headcount?
- What is the grid cost base behind AMD IT's 20% to 25% saving, in dollars, and how much of the saving came from AI rather than the annual hardware refresh?
- What do AMD's chip design uses of AI, including Synopsys flows, Copilot simulation triage and agentic design work, return in cost, headcount or time to revenue?
- What do the HR agent, Click to Claims, Click to RMA and the finance automation return, given that AMD has published only vendor figures or time savings for them?
- How much does AMD spend on internal AI tools and tokens, and how is that budget governed across IT and engineering?
- Who sits on AMD's internal AI review committee now that AMD's leadership page lists no president, and does the committee still approve every project?
- Did AMD's AI Group return to Claude Code after Anthropic addressed the Apr 2026 regression reports, and which provider did it move to?
Sources
AMD
- AMD IT: AI Productivity at Scale, ACE and Aria (Sep 10, 2026)
- AMD IT: Autonomous Mega Processes, Click to (Sep 10, 2026)
- AMD IT: Data Intelligence Platform, OPTIMA (Jul 21, 2026)
- AMD IT: Data Intelligence Platform and AI Researcher (May 15, 2026)
- AMD IT: AI hackathon playbook (May 11, 2026)
- AMD IT: Grid infrastructure efficiency with AI (Mar 12, 2026)
- AMD IT: Enhancing IT support at AMD (Feb 25, 2026)
- AMD IT: Transforming AMD with AI (Sep 2, 2025)
- AI @ AMD: Insider view of AI customer zero (case study)
- AMD IT home and resources
- Advancing AI 2026: Inside AMD IT, our enterprise AI journey (Jul 22, 2026)
- AMD leadership team
- Lisa Su biography
- Mark Papermaster biography
- Hasmukh Ranjan biography
- Andrej Zdravkovic biography
- Brian Amick biography
Filings and executive writing
Partners
News
- The Next Web: Mark Papermaster at the RAISE Summit (Jul 9, 2026)
- Digg: AMD's CPU architects on AI in Zen design (Sep 2026)
- InfoWorld: Developers question Claude Code's reliability, Stella Laurenzo (Apr 7, 2026)
- PC Gamer: AMD's senior director of AI on Claude Code (Apr 7, 2026)
- ITPro: Lisa Su on AI and AMD's hiring at CES 2026 (Jan 7, 2026)
- Business Today: Lisa Su tells Wired AMD is still hiring engineers (Aug 18, 2025)
- CIO.com: 4 things that make an AI strategy work, AMD's Chris Wire (Jul 2, 2025)
- Fortune: Employers toggle the deployment of AI, Hasmukh Ranjan (May 2, 2024)
Data confidence
- Confirmed
- Stated in a primary or reputable public source.
- Estimated
- Derived or approximated from public reporting.
- Unconfirmed
- Reported, but not yet corroborated.
AMD profile · Enterprise AI diffusion report by Imagine · facts checked against their sources on Nov 9, 2026
