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
Areas of focus
Each area anchors a distinct stream of writing within Artificial Intelligence.
Deep Learning
Foundations and scaling behaviour — including where returns genuinely diminish and what that implies for national investment.
Neural Networks
The architectures that carry real production workloads, as opposed to those that win benchmarks.
Transformers
The attention era: what it made possible, what it costs to run, and where its limits are now visible.
Physical AI
Robotics and embodied systems — intelligence that leaves the screen and enters the factory, the field and the clinic.
AI Infrastructure
Compute, energy, chips and the national stack. The least glamorous layer and the one that decides outcomes.
Sovereign AI Ecosystems
Nations owning their models, data and pipelines — and the institutional capacity that makes ownership real.
Consumer AI
Assistants and interfaces at population scale, priced and designed for the whole citizenry.
Enterprise AI
Adoption inside large regulated organisations, where the constraint is rarely the technology.
Large Language Models
Capability, cost and control of frontier LLMs — and the case for and against training your own.
Industrial Foundation Models
Sector-specific models trained on process data that no public corpus contains.
Next-Generation Transformers
The architecture roadmap beyond today's attention stack, read without enthusiasm or dismissal.
World Models
Systems that simulate environments rather than describe them, and what they change about planning.
Writing in Artificial Intelligence
Sovereign AI Is an Infrastructure Question, Not a Model Question
Nations racing to announce a national model routinely skip the harder problem: the compute, energy and data pipelines that make any model maintainable for a decade. A framework for sequencing the investment properly.
Industrial Foundation Models and the Return of Domain Expertise
General-purpose models plateau at the factory gate. The next tranche of value belongs to sector models trained on process data that exists only inside operating companies — which changes who holds the advantage.
Physical AI: When Intelligence Leaves the Screen
Embodied systems alter the economics of labour, safety and supply chains in ways that language models do not. What ministries and manufacturers should be planning for now, and what can safely wait.
After Attention — Reading the Architecture Roadmap Honestly
A practitioner's assessment of what succeeds the transformer, why the successor matters less than the infrastructure around it, and why most enterprise roadmaps should not wait for either.
Other domains
Geopolitics
Alliances, digital soft power and the credibility of institutions
5 topics · 3 posts 03Digital Economies
Inclusive growth, affordable access and the architecture of Vision 2050
5 topics · 3 posts 04Innovation & Technology
Quantum computing, blockchain, AGI and the frontier stack
5 topics · 3 posts 05Leadership & Strategy
Nation building, institutional reform and strategic resilience
5 topics · 3 posts 06My Journey
Engineer, founder, global strategist — and what changed at each step
3 topics · 3 posts 07Future Horizons
Speculative essays on convergence, scenarios and frontier predictions
3 topics · 3 posts 08Global Business Strategies
Fortune 2000 adoption, trade resilience and corporate diplomacy
6 topics · 3 posts