Anubhav Sachar
Skip to content
Navigate
HomeAboutAll InsightsContact
Insights — eight domains
Artificial Intelligence12 topicsGeopolitics5 topicsDigital Economies5 topicsInnovation & Technology5 topicsLeadership & Strategy5 topicsMy Journey3 topicsFuture Horizons3 topicsGlobal Business Strategies6 topics
Ventures
World AcceleratorVisit site Global Development 50Visit site
Connect for partnerships
Home/Insights/Artificial Intelligence
Domain 01 of 08

Artificial Intelligence

From deep learning foundations to sovereign national AI ecosystems

Overview

What this domain covers

Artificial intelligence has stopped being a laboratory discipline and become national infrastructure. That shift changes the questions worth asking. The interesting problem is no longer whether a model can be trained, but who owns the substrate beneath it — the compute, the energy, the data pipelines, and the institutional capacity to maintain all three across a decade rather than a funding cycle.

This domain follows that thread from the mathematics upward. It covers the architectures that actually carry production workloads, the industrial and consumer applications where value accrues, and the sovereignty question that increasingly determines which nations set terms and which accept them.

The writing is deliberately practical. Where a technology is genuinely close, the essays say so and describe what to do about it. Where it is not, they say that too.

Topics inside this page

  • 01Deep LearningFoundations and scaling behaviour — including where returns genuinely diminish and what that implies for national investment.
  • 02Neural NetworksThe architectures that carry real production workloads, as opposed to those that win benchmarks.
  • 03TransformersThe attention era: what it made possible, what it costs to run, and where its limits are now visible.
  • 04Physical AIRobotics and embodied systems — intelligence that leaves the screen and enters the factory, the field and the clinic.
  • 05AI InfrastructureCompute, energy, chips and the national stack. The least glamorous layer and the one that decides outcomes.
  • 06Sovereign AI EcosystemsNations owning their models, data and pipelines — and the institutional capacity that makes ownership real.
  • 07Consumer AIAssistants and interfaces at population scale, priced and designed for the whole citizenry.
  • 08Enterprise AIAdoption inside large regulated organisations, where the constraint is rarely the technology.
  • 09Large Language ModelsCapability, cost and control of frontier LLMs — and the case for and against training your own.
  • 10Industrial Foundation ModelsSector-specific models trained on process data that no public corpus contains.
  • 11Next-Generation TransformersThe architecture roadmap beyond today's attention stack, read without enthusiasm or dismissal.
  • 12World ModelsSystems that simulate environments rather than describe them, and what they change about planning.

All topics above are covered within this domain page — they are not separate pages.

A nation does not own its artificial intelligence because it owns a model. It owns it when it controls the compute, the data pipelines and the people who can rebuild the model when it breaks. Anubhav Sachar
Topic Guide

Areas of focus

Each area anchors a distinct stream of writing within Artificial Intelligence.

01

Deep Learning

Foundations and scaling behaviour — including where returns genuinely diminish and what that implies for national investment.

02

Neural Networks

The architectures that carry real production workloads, as opposed to those that win benchmarks.

03

Transformers

The attention era: what it made possible, what it costs to run, and where its limits are now visible.

04

Physical AI

Robotics and embodied systems — intelligence that leaves the screen and enters the factory, the field and the clinic.

05

AI Infrastructure

Compute, energy, chips and the national stack. The least glamorous layer and the one that decides outcomes.

06

Sovereign AI Ecosystems

Nations owning their models, data and pipelines — and the institutional capacity that makes ownership real.

07

Consumer AI

Assistants and interfaces at population scale, priced and designed for the whole citizenry.

08

Enterprise AI

Adoption inside large regulated organisations, where the constraint is rarely the technology.

09

Large Language Models

Capability, cost and control of frontier LLMs — and the case for and against training your own.

10

Industrial Foundation Models

Sector-specific models trained on process data that no public corpus contains.

11

Next-Generation Transformers

The architecture roadmap beyond today's attention stack, read without enthusiasm or dismissal.

12

World Models

Systems that simulate environments rather than describe them, and what they change about planning.