AI Has America's Oldest Monopoly Problem – Part 1

6 hours ago

The article argues that AI giants are creating a monopoly problem similar to historical cases like Standard Oil, by controlling the entire AI stack including GPUs, cloud compute, and model access.
Vertical integration in AI allows hyperscalers (like AWS, Azure, Google Cloud) to operate at multiple layers, creating a conflict of interest where they control infrastructure and compete with startups using that same infrastructure.
Startups can fine-tune models or run inference on rented GPUs, but scaling to compete at the frontier is hindered by limited GPU availability and control by competitors, forming a 'tollbooth' for market entry.
Open-weight models (e.g., Llama, Qwen) address model access but not infrastructure access, as running them at scale still requires compute resources dominated by hyperscalers.
The author advocates for government intervention to ensure fair access to AI infrastructure, drawing parallels to historical antitrust actions and proposing a solution inspired by Switzerland's telecom unbundling model.
Counterpoints include that vertical integration may drive efficiency and investment, and open models provide some competition, but the structural risk of stifled innovation remains a concern.
The article previews a series exploring policy approaches, with Part 2 focusing on unbundling infrastructure layers to separate control from competition.

Hasty Briefsbeta