Discuss why enterprise adoption of AI is shifting from 'best model on the leaderboard' to 'best model for the workload.' Examine the implications for India's AI governance strategy.

Benchmark rankings measure raw capability on standardised tests; enterprises deploy AI against workloads with fixed budgets, legal duties and latency needs. As deployment matures, model choice is turning into a fit decision rather than a ranking decision — a shift India's own AI framework has already absorbed.

Why the shift is happening

• Cost discipline: frontier closed models charge per token via managed APIs. India's answer — the IndiaAI Compute Portal's 38,000+ GPUs and 1,050+ TPUs at subsidised rates below ₹100/hour — shows infrastructure economics, not leaderboard rank, drives access [1].
• Data residency and IP: open-weight models (trained weights self-hostable, unlike API-only closed models) let firms keep sensitive data and fine-tuning in-house.
• Fit-for-purpose sizing: MeitY's India AI Governance Guidelines (5 November 2025) explicitly encourage smaller, resource-efficient models — a benchmark leader is often oversized for a routine task [2].
• Vendor and sovereignty risk: dependence on one foreign lab's pricing and roadmap is a strategic exposure, not merely a commercial one.
• Operational reality: downloading weights is easy; running them reliably at scale needs skills most firms lack.

Implications for India's governance strategy

• Validates the sovereign-model bet: indigenous models from Sarvam AI, BharatGen, Gnani and Soket, launched at the India-AI Impact Summit 2026, target Indic-language workloads where frontier models underperform [3].
• Regulating deployment, not just models: with self-hosting, risk sits in context of use — the Guidelines' techno-legal, whole-of-government design fits this [2].
• New institutions to build capacity: the AI Governance Group, Technology & Policy Expert Committee and AI Safety Institute must supply testing benchmarks and standards enterprises cannot generate alone [2].
• Shared compute as public good: pooled national capacity prevents duplicative, idle private infrastructure [1].

Workload-fit selection reframes AI from a race for the top rank into a question of appropriate technology. India should deepen assurance testing, open-weight licensing clarity and skilling so that choice translates into capability — advancing the Summit's own theme, "Welfare for All, Happiness of All."