KPMG on India's AI building blocks, and where F2A puts its money

What happened

A KPMG report, Agrasar Bharat: Turning India’s Aspirations into Action, says India must build capabilities and institutions before AI pays off at scale. It names five building blocks: skilled talent, research and development, computing and data infrastructure, technology enterprises, and digital public infrastructure. On AI it asks for secure, well-governed data ecosystems, adequate computing capacity and responsible governance.

The report uses UPI as its model. UPI processed more than 24,161 crore transactions worth around Rs 314 lakh crore in FY2025-26, and had 55.49 crore users onboarded by June 2026. KPMG suggests the same open-platform approach for AI and other technologies. It also warns that innovation alone will not guarantee results.

The same week, Analytics India Magazine set out how one new fund plans to invest. Fundamentum Frontier Advisors (F2A), anchored by Nandan Nilekani, has Rs 3,000 crore in subscriptions and a raised limit of Rs 3,500 crore. It plans about 40% for physical AI, 40% for digital consumer companies that use AI and 20% for enterprise tools, across 12-15 companies with cheques of Rs 100-200 crore.

Reported by The Economic Times (ANI); carried as story 5 in our 26 October 2026 brief.

The numbers

What it means for builders

1. Compute is a named constraint, so price every AI feature per task

KPMG lists computing capacity as a building block India still has to build. Until it exists, compute stays a cost you pay per call. Before you ship an AI feature, work out what one task costs at today's prices and at ten times today's volume. Use the smallest model that passes your own test set, and keep the large model for the cases that need it.

2. Clean, well-governed data is the part you can own

The report asks for secure, well-governed data ecosystems. For a product team that is a practical instruction: know what data you collect, get consent for it, label it, and keep an evaluation set in the languages your users speak. A model you call through an API is available to every competitor. A labelled dataset from your own users, collected with consent, is not.

3. Build on public platforms where they exist, as UPI-era fintechs did

KPMG's model is UPI: one public platform with 55.49 crore onboarded users that thousands of products built on. If similar public layers arrive for AI, the teams that plug in early will spend less on infrastructure. Track the public datasets, compute programmes and language models on offer, and design your stack so a model or data source can be swapped without a rewrite.

Who should act

• Founders of AI-first startups — F2A's split shows where this growth money goes: most of it to physical AI and consumer products, the smallest share to enterprise tools.
• Product managers adding AI features — Compute cost per task and data quality decide whether a feature survives at scale.
• Engineering leads choosing models and hosting — A stack where the model and data source can be swapped keeps you ready for public AI platforms if they arrive.

This week

• Write down the cost of one AI task in your product today, and what it would be at ten times today's volume.
• Run your top AI flow on a smaller model against your own test set and record where it fails.
• List the user data you collect for AI, the consent you hold for it, and who labels it.
• If you are raising growth money for an AI or deeptech company, check whether your stage and cheque size fit funds like F2A.

Read next

• LLM costs at scale — how to price an AI feature per task
• Building AI features for Indian apps — language, data and cost choices for Indian users
• How many people use UPI in India? — the official user count and its source
• Sarvam — Indian-language models for speech and text
• DigitalOcean — simple cloud hosting for self-run models
• Cloudflare — edge network with Workers AI
• The story in the 26 October brief — the summary and source

Sources

1. The Economic Times: Talent, compute and R&D key to India's AI ambitions: report, 26 October 2026 — ANI copy; our checker could not open ET, so figures were checked on The Tribune's copy of the same ANI story
2. The Tribune (ANI): Talent, compute and R&D key to India's AI ambitions: report, 26 October 2026
3. Dataquest: AI, semiconductors key to India's 2047 growth ambition: KPMG, 25 October 2026
4. Analytics India Magazine: Why Nandan Nilekani-backed F2A believes India's investment opportunity lies in edge AI, 22 October 2026
5. productgrowth.in: How many people use UPI in India?, checked 23 October 2026 — Ministry of Finance figure

Questions people ask

What does KPMG say India needs for its AI ambitions?

Five building blocks: skilled talent, research and development, computing and data infrastructure, technology enterprises, and digital public infrastructure. For AI specifically, secure and well-governed data, enough computing capacity and responsible governance.

What is F2A and what does it invest in?

Fundamentum Frontier Advisors is an investment platform run by Ashish Kumar and anchored by Nandan Nilekani. It has Rs 3,000 crore in subscriptions and plans about 40% for physical AI, 40% for consumer digital companies using AI and 20% for enterprise tools.

Is F2A an edge AI fund?

Not on the evidence published. Its stated split covers physical AI, consumer digital and enterprise tools. An edge AI platform is one of three deals it expects to close within three months.