Over $650 billion has been deployed into AI infrastructure. Here’s where the money is going and who’s winning the race for compute sovereignty.
Mega Projects
| Project | Investment | Players | Strategic Purpose |
|---|---|---|---|
| Stargate | $500B | OpenAI + SoftBank + Oracle | US-based compute independence |
| xAI Memphis | $6B | xAI | 100K H100 GPUs, built in 4 months |
| Google TPU | 7GW capacity | Custom silicon at scale | |
| Amazon Trainium | Undisclosed | AWS | Escape NVIDIA tax |
| Microsoft Maia | Undisclosed | Microsoft | OpenAI hedge |
Compute Providers: The “AWS for AI” Race
| Company | Valuation | Focus |
|---|---|---|
| CoreWeave | $12B | “AWS for AI” – IPO coming |
| Cerebras | $4.3B | Wafer-scale chips, IPO filed |
| Crusoe | $3B | Clean compute (stranded gas) |
| Lambda | $500M | GPU cloud |
| Together | $230M | Decentralized inference |
Energy Plays
AI needs power. Lots of it:
- GPT-4 query: 10x energy of Google search
- Data centers: 8% of US power by 2030
- Nuclear: 24/7 baseload power for AI
| Company | Deal | Purpose |
|---|---|---|
| Microsoft Constellation | Three Mile Island | Nuclear restart, 24/7 clean power |
| Amazon Talen Energy | $650M | PA nuclear campus |
| Google Kairos | SMR nuclear | 2030 deployment |
| Oklo | $850M | Sam Altman backed, small modular reactors |
The Consolidation Math
- 90%+ of AI compute controlled by 5-7 players
- $2.8T NVIDIA market cap (90%+ training chip share)
- 5-7 companies will control the infrastructure layer
Why Infrastructure Matters
- Permanent moat: Can’t be copied overnight
- Bottleneck = margin: Control the chokepoint, control pricing
- AI dependency: Every model needs compute
- Sovereignty: Nations investing for strategic independence
The insight: Physical infrastructure creates permanent moats. Whoever controls compute controls the AI economy.
This is part of a comprehensive analysis. Read the full analysis on The Business Engineer.
More Resources
