This is the clearest articulation of the fork I've seen. "The danger was never in the weights, it was in the wiring" — that's exactly right.
I'd push one step further. The wiring doesn't get decided by market dynamics alone. In your "Grid" scenario, the pooled training runs, the open benchmarks, the governments buying in—those are all policy choices, not just competitive ones. They need a public buyer to make them stick.
I just wrote about what that could look like: an asset-light consortium that rents compute from the hyperscalers on competitive terms, but owns the science—data, code, methods, evaluations—as a public good. Less a market outcome, more a deliberate institution.
This is the clearest articulation of the fork I've seen. "The danger was never in the weights, it was in the wiring" — that's exactly right.
I'd push one step further. The wiring doesn't get decided by market dynamics alone. In your "Grid" scenario, the pooled training runs, the open benchmarks, the governments buying in—those are all policy choices, not just competitive ones. They need a public buyer to make them stick.
I just wrote about what that could look like: an asset-light consortium that rents compute from the hyperscalers on competitive terms, but owns the science—data, code, methods, evaluations—as a public good. Less a market outcome, more a deliberate institution.
Same destination, different vehicle. Would value your read: https://mwkorver.substack.com/p/rent-the-compute-own-the-science