Building the AI-Ready Future: From Infrastructure to Skills

Contents

Executive Summary

This multi-speaker summit session explores the comprehensive infrastructure, governance, and organizational frameworks needed to democratize AI access globally and across India specifically. Speakers from AMD, the Indian government, and industry leaders argue that AI readiness requires not just compute capability but also sovereign infrastructure, change management, and an open ecosystem approach—moving beyond the "GPU-centric" narrative to include edge computing, enterprise adoption, and physical AI applications.

Key Takeaways

AI Readiness = Compute + Capability + Change Management: Technical infrastructure alone is insufficient; organizations must address organizational change, talent development, and business case clarity.
Open Ecosystems Enable Sovereignty: Countries and enterprises avoiding vendor lock-in (through PyTorch, Triton, open standards) can innovate faster, reduce costs, and maintain strategic autonomy.
The Indian Opportunity is Multi-Layered: Sovereign infrastructure (MEITY, Matei), multilingual model development, SMB transformation consulting, and physical AI all represent distinct business and research opportunities not yet saturated.
Edge & Physical AI are The Next Frontier: After cloud GPU scaling, the innovation edge moves to robotics, autonomous systems, and industrial AI—requiring different architectures and full-stack solutions, especially for India.
Stay Curious & Question Assumptions: The pace of AI change means yesterday's truths (e.g., "AMD is inference-only," "you need CUDA," "enterprises adopt before individuals") become false quickly. Success requires continuous learning and skepticism of old mental models.

Key Topics Covered

Sovereign AI & National AI Infrastructure: US Department of Energy's Genesis Mission; government-led AI initiatives for scientific discovery, energy, and national security
Compute Architecture & Scalability: AMD's hardware portfolio from AI PCs to exascale systems; the Helios rack and pathway to zettascale computing
Open Ecosystem Philosophy: PyTorch, JAX, Triton, and day-zero model support as antidotes to vendor lock-in
AI Readiness for Indian SMBs & Enterprises: Change management, business case development, and the gap between large enterprises and micro/small/medium businesses
Multilingual LLM Development: Supporting Indian languages and minority languages in foundational models (Lumi Finland case study)
Physical AI & Edge Computing: Moving AI inference from cloud to edge; robotics, autonomous systems, and industrial applications
Agentic AI & Governance: Human-in-the-loop systems for research acceleration and risk mitigation
Enterprise AI Adoption: From chatbots to agentic workflows; managing customer knowledge gaps and use-case variation
Public-Private Partnerships: India's infrastructure strategy combining government labs, private sector, and academia
Talent & Skills Development: Building a foundational knowledge base across startups, researchers, and enterprise workers

Key Points & Insights

The "Genesis Mission" Framework: The US Department of Energy's initiative uses AI to accelerate scientific discovery by shortening the hypothesis-to-analysis cycle, reducing research costs while improving output efficiency. This model is being adapted internationally and aims to integrate government labs, academia, and private sector research.
Compute Density Breakthrough: A single Helios rack delivers 2.9 exoflops of AI compute (FP4) at 220 kilowatts—a dramatic efficiency gain compared to 2007 exascale projections. The path to zettascale within a decade suggests compute constraints should not limit problem-solving ambitions.
Open Standards as Competitive Advantage: AMD's commitment to PyTorch, JAX, Triton, and open-source infrastructure (rather than proprietary CUDA lock-in) enables vendor-agnostic AI development. This is critical for countries seeking technological sovereignty and avoiding supply-chain dependencies.
Change Management as a Barrier, Not Technology: For Indian SMBs and legacy enterprises, the primary obstacle is not understanding AI but accepting workflow transformation, redundancy of existing roles, and adoption of new processes—requiring advisory support, not just software.
Multilingual AI Gaps: With 22 official Indian languages, many with <5 million speakers, training data is fragmented across silos in different forms. The Lumi (Finnish) experience shows that small-language LLM development requires coordinated, nation-level infrastructure and governance to succeed.
Day-Zero Model Support: AMD's commitment to supporting new model architectures (Qwen 3.5, Deep Seek, Indian language models) on release day—via contributions to open frameworks like SG Lang—enables rapid adoption without waiting for vendor optimization cycles.
The "Three I" Framework for SMB Readiness: Intent (leadership commitment to AI transformation), Investment (in talent, infrastructure, and process change), and Implementation (ongoing support and change management). Most SMBs fail at intent and investment phases.
Agentic AI Requires Governance, Not Regulation: The governance model proposed uses human-in-the-loop validation before AI agents can update outcomes—analogous to peer review in academia—balancing autonomy with accountability and preventing unintended consequences.
Physical AI & Edge Deployment: The shift from cloud-centric to edge AI (robotics, autonomous vehicles, industrial systems) requires different hardware (smaller accelerators, low power) and full-stack software solutions. India's startup ecosystem is well-positioned to lead physical AI innovation.
D2C Adoption Outpacing Enterprise: Unlike previous software waves (adopted by enterprises first, then individuals), generative AI adoption is reversed—D2C and individual usage are ahead of enterprise and SMB adoption, creating a unique opportunity for startups and consulting services to bridge the gap.

Notable Quotes or Statements

Thomas Courtney (AMD, former Oak Ridge director): "Innovation, you know, if you think about AI, AI didn't happen magically with Nvidia or AMD. It happened because US government took the risk to invest in first kind systems."
Thomas Courtney: "Governance is not regulation. If you want agentic systems driving innovation, you want to make sure there's a person in the loop before you can update and let this thing drive innovation."
Shrim Paneer Salvi (MEITY/Matei): "What you don't measure, you don't manage. If you really want to build this model of AI readiness quotient, you know, and then slowly integrate and unlearn existing processes—there could be a huge opportunity around building this business."
Shrim Paneer Salvi: "Small and medium enterprises or SMBs are extremely—you should never underestimate their business acumen. If you show value to them, they will buy."
Tim Korklewski (AMD): "The whole point with day zero support is that [the notion that you need a specific GPU to run AI] is absolutely false."
Tim Korklewski: "With PyTorch, you're writing Python code. You're not writing vendor-specific code. It's an open ecosystem. You don't want to be tied [to one supplier]. It's going to stifle innovation. It's going to increase costs."
Jil Garcia (AMD, Physical AI): "GPUs are one aspect of it, but you need very different technology... [that] need to sense, perceive, act, react in such a quick manner that there is no time to go back to the cloud."

Speakers & Organizations Mentioned

Name	Organization / Role	Focus Area
Thomas Courtney	AMD (Senior VP); former Director, Oak Ridge National Lab	Sovereign AI, US Department of Energy Genesis Mission, scientific computing
Tim Korklewski	AMD (Software/Ecosystem Lead)	Open-source ecosystems, multilingual models, enterprise AI
Shrim Paneer Salvi	Ministry of Electronics & IT (MEITY), Matei Startup Hub, India	SMB AI readiness, change management, Indian startup ecosystem
Jil Garcia	AMD (Physical AI, Robotics, Communications)	Edge AI, robotics, physical AI infrastructure
Lisa Su	AMD CEO (referenced)	General direction on "AI Anywhere" strategy
Prime Minister Modi	Government of India	Announced AI for all initiative; mentioned nuclear/defense policy
President Macron	France (referenced)	Announced first French "AI Factory" (Alice Raoul) on AMD MI430X

Institutions & Programs:

US Department of Energy (DoE) & Genesis Mission
Oak Ridge National Laboratory (ORNL)
Lawrence Livermore & Sandia National Labs
Lumi Supercomputer (Finland)
Matei Startup Hub (India Ministry of Electronics & IT)
AMD Developer Cloud

Technical Concepts & Resources

Hardware & Infrastructure

Helios Rack (AMD): 72 GPUs per rack, 2.9 exoflops AI compute (FP4), 220 kW power
MI355/MI430/MI450 GPUs (AMD): Training and inference accelerators
Exascale Systems (Frontier @ Oak Ridge): <20 MW power efficiency goal achieved
Zettascale Computing: ~10x scaling from exascale projected within 10 years
Edge Accelerators (AMD): Lower-power alternatives to GPUs for robotics and physical AI

Software & Frameworks

PyTorch: Primary framework for 99% of customers discussed; vendor-agnostic, Python-based
JAX: Emerging framework for advanced AI research
Triton: Python-like language for GPU kernel optimization; compiler-level optimization without hardware lock-in
SG Lang: Open-source inference engine; AMD day-zero support contributor
Docker Containers (AMD): Pre-packaged software stacks to reduce deployment friction
Hugging Face: Model repository referenced for weights and model distribution

Models & Datasets

Qwen 3.5 Codeex (Alibaba): Day-zero AMD support mentioned
Deep Seek: Noted for multi-head latent attention; AMD achieved better TCO/performance via SG Lang
BLOOM 176B (Hugging Face/BigScience): Multilingual open model; AMD participation cited
Indian Language Models (12 languages): Recent announcement by Prime Minister Modi; day-zero AMD support
Lumi for Finnish/Indic Languages: Case study of developing LLMs for minority languages with <5M speakers

Governance & Frameworks

Human-in-the-Loop Agentic AI: Governance model requiring human validation before agents update outcomes (peer review analogy)
AI Readiness Quotient: Proposed metric for measuring organizational maturity in AI adoption
Three I Framework: Intent, Investment, Implementation for SMB AI readiness
Sovereign AI Infrastructure: Public-private partnerships combining government labs, industry compute, and academic research
Confidential Computing & Security-by-Design: Required for national security and sensitive workloads

Industry Benchmarks & Initiatives

TOP 500 Supercomputer List: Historical context for AMD/NVIDIA leadership milestones
Exascale Computing Project (ECP): US government funding program that drove efficiency innovations
CRM Utilization Metrics: ~28% average utilization in Indian enterprises (illustrative of enterprise AI adoption challenges)

Implicit Risks & Challenges Identified

SMB Change Resistance: Legacy, family-owned businesses in India (>80% of private sector) resist workflow transformation required for AI integration.
Data Siloing: Even large Indian enterprises lack integrated data—different departments unable to synthesize datasets, blocking analytics.
Talent Scarcity: Shortage of advisors/consultants with proven AI implementation experience to guide SMB adoption.
Multilingual Model Fragmentation: Training data for minority languages insufficient; no unified governance to coordinate development.
ROI Unclear: Businesses uncertain of AI's value proposition for their specific use cases; lack of demonstrated success models.
Vendor Lock-in Legacy: Historical reliance on proprietary platforms (e.g., CRM at 28% utilization) creates skepticism about new technologies.

Recommendations for Further Exploration

Review the Genesis Mission framework documentation for sovereign AI science initiatives.
Examine the Lumi Supercomputer case study (Finland) for multilingual LLM development patterns applicable to India.
Explore AMD's Developer Cloud and Docker container offerings for POC validation.
Study PyTorch and Triton documentation for vendor-neutral AI development best practices.
Investigate India's Matei Startup Hub programs for SMB AI readiness consulting models.
Monitor Physical AI and robotics announcements (AMD-Gin01 humanoid, autonomous systems) as emerging opportunity area.