Everyone talks about agi timelines, hardly anyone talks about compute gatekeeping

Most AI discourse fixates on when AGI will arrive or which breakthrough will unlock superhuman intelligence. But a viral Reddit post from srodland01 shifts the lens to a more structural constraint: compute gatekeeping. While algorithmic progress is often open and discussed, the hardware that runs these models—specifically Nvidia's H100 GPUs—is heavily concentrated in a tiny set of players. OpenAI, Microsoft, and a few other labs together own the overwhelming majority of these chips, creating a bottleneck that no amount of open-source ideas can bypass. The post argues that even if the math behind a breakthrough model is fully published, only those with access to massive compute clusters can actually train or run it.

This centralization of compute, rather than of ideas, leads to a future where the "god model" exists but is accessible only to three companies. For professionals building on AI, this means that open-source models like Llama 3 or Mistral may be free to use, but they cannot scale to the frontier without hardware that is essentially locked down. The implications are profound: AGI timelines become less about discovery and more about who controls the physical infrastructure to realize those discoveries. As compute costs rise and supply remains tight, the concentration of hardware could become the single greatest gatekeeper of AI progress—outweighing any advances in algorithms or open research.