The GPU Pipeline Visualized

LLMs are here to stay, and the GPU pipeline — the picks and shovels powering this surge in novel compute demand — is the clearest lens on who controls who, and therefore, who controls the profit.

Takeaways

Compute is the choke point — GPUs flow from TSMC → Nvidia → cloud providers, and whoever sits on those pipes dictates the pace of AI.

Cash alone isn’t enough — billions in financing (SoftBank, Microsoft, Oracle) only matter if they secure long-term contracts for scarce compute.

Contracts shape power — exclusivity (Azure), mega-deals (Oracle $300B), and capacity guarantees lock in who profits and who’s locked out.

The Map

This map was inspired by a version that circulated on X after OpenAI’s $300B Oracle deal. This one attempts to refine it, distilling relationships down to the three real currencies: compute, cash, and contracts.

5 Related Facts

💰 SoftBank is now OpenAI’s biggest outside backer, pledging up to $40B and tying the deal to future infrastructure projects.

🖥️ Oracle landed one of the largest corporate contracts in history, a $300B, five-year commitment to supply OpenAI compute.

⚙️ CoreWeave quietly secured a $6.5B deal with OpenAI, cementing its role as a key overflow provider of GPU capacity.

🔋 Nvidia depends on TSMC’s fabs for every advanced GPU, making wafer capacity in Taiwan a hidden choke point in the pipeline.

📈 OpenAI’s valuation hit $500B in October 2025, overtaking SpaceX and putting it among the world’s most valuable startups.