AI is no longer a software story. It is a story about buildings, grids, straits, and alliances. KJ maps the chokepoints — TSMC, ASML, HBM memory, gigawatt-scale power — and the three blocs now forming around compute: the US hyperscaler axis, China's domestic stack, and the Gulf-led swing states buying their way in.
01Opening
The New Oil Wells
What you need to know
- 1
Compute has become a strategic resource. Nations now treat GPU clusters and hyperscale data centres the way they once treated oil refineries and naval bases.
- 2
The supply chain has three hard chokepoints: TSMC in Taiwan (advanced fabrication), ASML in the Netherlands (EUV lithography), and SK Hynix and Samsung in South Korea (HBM memory).
- 3
The real bottleneck is no longer chips. It is electricity, water, land, and grid interconnects. AI data centres are projected to consume roughly 945 TWh globally by 2030 — more than Japan uses today.
- 4
Three blocs are forming: a US-led hyperscaler axis, a Chinese domestic stack (Huawei Ascend, SMIC, CXMT), and a group of swing states — the UAE, Saudi Arabia, India, Malaysia, Indonesia — selling geography, power, and capital.
- 5
Europe is regulating what it does not build. Its share of frontier AI compute is a rounding error.
- 6
The 21st century will not be decided by who writes the best model. It will be decided by who owns the buildings the model runs in.
Read the full intelligence report — free
Enter your email to unlock the complete analysis, forecasts and sources, and download the PDF.
- 02GPUs Are the New Refineries
- 03The Five Layers of AI Power
- 04The Three Points of Failure
- 05The Real Bottleneck Is Watts, Not Weights
- 06How the World Is Splitting
- 07Why Riyadh and Abu Dhabi Are Buying Compute
- 08Regulating What You Do Not Build
- 09Compute Becomes a Defence Asset
- 10What Comes Next
- 11The Rank Order
- 12What to Actually Watch
- 13What This Means for You
- 14The Honest Bottom Line
Already a member? Sign in to read everything instantly.