Twitter co-founder Jack Dorsey publicly backed venture capitalist Chamath Palihapitiya’s call to fully embrace open source artificial intelligence (AI). <\/p>\n
Dorsey replied “yes” to Palihapitiya’s post warning that US restrictions on open models would be economically ruinous. <\/p>\n
Palihapitiya argued that closing off open source AI would force American firms to pay $26 to $56 per million tokens for intelligence that rivals abroad can buy for $0.50 to $1. He called that gap unsustainable. <\/p>\n
The AI Pricing Gap Behind the Debate <\/p>\n
Palihapitiya’s argument rests on a widening split between cost and capability. Open weight models have closed much of the performance gap with proprietary systems, yet the price difference remains enormous. <\/p>\n
Chinese labs have driven that shift. Beijing-based Moonshot AI’s new model Kimi K3 topped coding benchmarks this month, rattling US chip stocks. <\/p>\n
Other releases point to a broader narrowing capability gap between Chinese and American systems. <\/p>\n
Palihapitiya frames this as untenable if AI truly underpins future economic activity, since American businesses would face a structural cost disadvantage against global competitors. <\/p>\n
He posed the dilemma bluntly, saying intelligence cannot be the engine of the economy and a premium import at the same time. <\/p>\n
A Military Argument, Not Just an Economic One <\/p>\n
Palihapitiya extended his case to national defense. Paying dozens of dollars per million tokens to defend US systems, while adversaries attack for far less, carries the same imbalance, he said. <\/p>\n
David Sacks echoed that view. He agreed with researcher Sebastian Mallaby that dangerous capability will soon spread freely regardless of policy. <\/p>\n
Mallaby pointed to the same Mythos-level cyber capability concerns already flagged around Anthropic’s Claude Mythos model, arguing the world moves quickly from almost nobody holding that power to nearly everyone holding it. <\/p>\n
Sacks had predicted Chinese models would reach advanced cyber capability within months. He noted that Washington itself staggered its GPT-5.6 release over similar security worries, yet gatekeeping still failed to slow foreign progress. His answer is AI-powered cyberdefense rather than restriction. <\/p>\n