AI Competitiveness: Turning Insight into Action

High-level themes

• AI competitiveness depends on translating compute into capability across the full stack — compute, models, data, talent, energy, and supply chains — rather than on raw announcements.
• The landscape is shifting toward a multipolar, “sovereign” AI world: many countries (including India) are building national AI projects across data, models, and infrastructure instead of relying solely on U.S./China stacks.
• Governance and adoption matter as much as hardware. For agentic AI in particular, human-in-the-loop controls, peer-review-style oversight, and public–private–academic collaboration are central to safe scaling.

Technology and infrastructure

• Compute and energy are tightly linked: data center investments and gigawatt-scale infrastructure are accelerating, and energy planning is integral to AI capacity.
• Edge and “physical AI”: high-density compute (for example, a single rack delivering ~2.9 exaflops) is increasingly being pushed to the edge, enabling rural and sector-specific applications.
• Hardware ecosystem: GPUs, multi-GPU workstations, and CPUs (e.g., AMD Threadripper) remain key. Supply-chain limitations and access to high-performance workstations are adoption bottlenecks in some regions.
• Federated and collaborative infrastructure: federated compute and curated shared datasets enable national-scale problem solving without centralizing all data.

Models, product features, and developer tools

• Pretrained models plus fine-tuning: many practical problems can be solved by downloading pretrained weights and fine-tuning rather than building trillion-parameter models from scratch.
• AI PCs and local compute: consumer and enterprise PCs with capable GPUs can enable hands-on experimentation and prototyping.
• Anthropic examples:
  • Claude variants and verticalized products (e.g., Claude for Finance, Claude for Healthcare).
  • Claude Co‑work for agentic teamwork and multitasking agents.
  • Cloud Code for engineers, lowering barriers for non-engineers and enabling agentic workflows via natural input (including audio).
• Agentic AI design pattern: pair an inner-loop of autonomous agent activity with an outer governance loop (human review) to avoid unauthorized autonomous changes.

Governance, safety, and standards

• “Governance” is preferred as a broader, iterative framework over immediate heavy-handed regulation; industry self-regulation has a role where appropriate.
• Industry alliances and self-regulatory commitments (e.g., multilingual training, trusted tech stacks, testing regimes) can accelerate safe scaling.
• Alignment across hyperscalers, model labs, sovereign clouds, and public investment is needed — roles overlap and motivations differ.
• Human-in-the-loop and peer-review analogies are recommended for agentic outputs to control updates and safety-critical decisions.

Workforce, training, and adoption guidance

• Rapid, modular upskilling (certificate programs, short courses, hands-on hackathons) accelerates capability building faster than multi-year degrees.
• Top-down leadership is critical: executive sponsorship and changes to job expectations (CEO messaging, internal incentives) drive organizational adoption.
• Democratizing access: tools that remove coding barriers (natural language, audio, agentic assistants) widen adoption beyond engineers to analysts, HR, and small businesses.
• Local awareness: in India, the largest deployment bottleneck is low awareness of AI risks and benefits outside major cities; skilling and awareness campaigns are required for equitable uptake.

Practical moves recommended

1. Build open, composable, interoperable stacks to attract developers and scale solutions.
2. Ensure broad access to compute, pretrained models, and datasets so talent can practice and prototype.
3. Create cross-sector “lighthouse” problems (public–private–academic) to boost productivity and produce measurable outcomes in health, education, and agriculture.
4. Deploy right-sized governance: industry-led trusted stacks and human-in-loop controls for agentic AI while governments develop regulatory guardrails.
5. Invest in modular skilling programs, hackathons, and incentive structures to diffuse AI into everyday office work and small businesses.
6. Prioritize multilingual training and context-aware models for international deployment.

Measuring impact

• Panelists noted there is no settled set of welfare or impact metrics beyond GDP or enterprise value; governments and institutions are still designing ways to quantify AI adoption outcomes for citizen welfare.

Practical guides and tutorials referenced

• Fine-tuning pretrained models and using AI PCs for hands-on projects.
• Certificate programs, short training courses, and hackathons as rapid talent and prototype pipelines.
• Product-level guidance: Anthropic’s Cloud Code and Claude Co‑work as examples of tools that make coding and agentic workflows accessible.
• Industry “trusted stack” frameworks as governance guides.

Main speakers and contributors

• Dr. Thomas Zakaria — Senior Vice President, Strategic Technology Partnerships & Public Policy, AMD; commissioner on the Geotech Commission (spoke on compute, open stacks, public–private collaboration, and infrastructure).
• Ria (Anthropic) — International Policy / Special Projects Lead at Anthropic (spoke on constitutional AI, Claude products, agentic workflows, international scale and values).
• Pablo Chavez — Adjunct Fellow (Technology & National Security Programs) (spoke on sovereign AI projects, multipolar AI landscape, and allied/sovereign alignment).
• Additional contributors: audience members (e.g., Abashek, Shri Agarwal) who raised adoption/awareness and workforce questions; a moderator facilitated the panel.