China playbook 101: don't build a bigger walled garden. Make walled gardens pointless. They did it to solar. China makes 86 to 96% of the world's polysilicon, wafers, cells and modules. Flooded the market, crashed prices over 30%, buried anyone trying to compete on premium tech. They did it to EV batteries. CATL and BYD alone are more than half the global market. Not cheap labor, structural cost advantage from owning the entire supply chain. Now watch an American infra company run the exact same play. Together AI (just raised $800M at an $8.3B val.): - Rents out GPU compute so companies can run open-source models instead of closed ones - Tasks moved from closed models to Together AI's infra cost one-fifth to one-seventh as much, per a founder cited in NYT reporting - Backed by Aramco, Nvidia, Vista Equity, General Catalyst, 500MW+ of compute secured - Saw the closed-model bottleneck coming and took the China route before it became one Closed labs and hyperscalers: - $650B+ from four hyperscalers in 2026 alone - Roughly 5,427 US data centers versus about 449 in China - Nearly half of planned US builds are getting delayed or canceled, power and grid bottlenecks - Staying closed, buying distribution with investor billions instead of earning it by opening up The closed side is raising more and showing less for it. $122B+ rounds building walls. $800M just made those walls a fifth as relevant. Call it "Borrowed Scale." Buying your reach with money instead of earning it by giving the model away. That's the closed labs and the hyperscalers. You can't out-spend a strategy you refuse to copy. When intelligence becomes zero cost, cost becomes everything.