Twitter Co-Founder Backs Open Source AI Warning: What’s at Stake?
Jack Dorsey’s public backing of venture capitalist Chamath Palihapitiya’s call for open source AI signals a growing schism among US technology leaders over how America should compete with lower-cost foreign models. The debate centers on whether restricting advanced AI models protects national security or instead handicaps American firms economically while failing to slow Chinese progress anyway.
- US firms pay $26 to $56 per million tokens versus $0.50 to $1 globally, creating an unsustainable cost disadvantage for American companies
- Beijing’s Moonshot AI recently topped coding benchmarks with its Kimi K3 model, signaling Chinese labs have narrowed the capability gap
- Dorsey and prominent venture investors now argue gatekeeping fails to slow foreign progress while imposing structural costs on domestic competitiveness
- $26, $56 Price per million tokens US firms currently pay versus foreign competitors
- July 18, 2026 Date Dorsey publicly endorsed Palihapitiya’s open source AI position
- $0.50, $1 Cost per million tokens available to overseas buyers in comparison
Jack Dorsey, the Twitter co-founder and current CEO of Block, replied “yes” to Palihapitiya’s assertion that US policy restrictions on open source artificial intelligence models would prove economically ruinous for American firms.
Palihapitiya had framed the issue in starkly economic terms: if US companies must pay $26 to $56 per million tokens for AI capability while rivals in other countries access equivalent models for $0.50 to $1, American businesses face an untenable structural disadvantage that no amount of engineering innovation can overcome.
Dorsey’s one-word endorsement carries institutional weight, Block, which Dorsey built from Square, has developed Goose, an open source AI agent that he has championed publicly for years, aligning his commercial interests with the pro-openness position.
Moonshot AI’s Coding Win Exposes Narrowing US-China AI Gap
Palihapitiya’s economic warning gains urgency from concrete evidence that Chinese AI labs have closed much of the performance gap with American systems.
In July 2026, Beijing-based Moonshot AI released Kimi K3, which topped coding benchmarks that month, a technical milestone that rattled US semiconductor stocks and signaled that open weight models now deliver capability comparable to proprietary systems at a fraction of the cost.
The shift reflects years of Chinese investment in open model research and deployment; Beijing labs have no incentive to match US pricing and instead compete on accessibility and cost.
Open weight models, those whose underlying architecture and learned parameters are publicly available, have historically lagged proprietary closed systems in raw performance. That gap has narrowed sharply as Chinese researchers deployed models trained on massive datasets at comparatively low operational cost.
Moonshot’s Kimi K3 result is not an isolated datapoint but rather a signal of a broader trend: the capability differential between Chinese and American systems has compressed to the point where the remaining advantage no longer justifies the price premium American firms face.
For institutional investors tracking AI infrastructure costs and competitive positioning, this benchmark result implies that future AI budgets will depend less on which country’s model a firm buys and more on how much it has to pay.
The asymmetry creates a strategic problem that Palihapitiya frames as unsustainable: if artificial intelligence will genuinely underpin future economic activity, American businesses cannot afford to treat it as a premium import.
Sacks and Mallaby Reframe Restriction as a Vulnerability, Not a Defense
Palihapitiya extended his argument beyond economics to national defense, arguing that paying dozens of dollars per million tokens to defend US systems while adversaries attack for far less replicates the same cost imbalance on the security front.
David Sacks, another prominent venture investor, amplified that reasoning by citing researcher Sebastian Mallaby’s observation that dangerous AI capability inevitably spreads from scarcity to near-ubiquity far faster than policymakers anticipate.
Mallaby flagged the same concern around Anthropic’s Claude Mythos model, a system with cyber capability implications, yet pointed out that the world moves rapidly from a state where almost nobody holds such power to one where nearly everyone does, regardless of official restrictions.
Sacks had previously predicted that Chinese models would reach advanced cyber capability within months. He observed that Washington itself staggered the GPT-5.6 release over similar security worries, yet gatekeeping still failed to slow foreign progress.
His conclusion inverts the traditional national security argument: instead of restricting model access to protect American defenses, the priority should be AI-powered cyberdefense to counter foreign threats.
That framing suggests restriction creates false confidence, the appearance of control, while leaving US firms structurally unprepared for adversaries operating on cheaper, openly available models.
For institutional investors in cybersecurity and defense technology, the implication is that the market opportunity may shift away from model restriction and toward defensive AI systems that operate atop whatever models happen to dominate globally.
Both Sacks and Mallaby argue that the spread of capability is inevitable; the question is whether America chooses to disadvantage itself economically while the spread happens anyway.
Washington’s Gatekeeping Debate and the Path Forward
Dorsey and Palihapitiya’s exchange lands as US policymakers actively debate how tightly to control advanced AI models. Officials have floated plans to vet AI models before release, hoping to manage security risk without ceding ground to China in the longer competition for AI dominance.
The tension is direct: tighter restrictions could slow dangerous capability spread in theory, but they also impose immediate cost penalties on American firms and may not slow foreign progress in practice.
Palihapitiya’s argument pushes the inverse conclusion from an economic angle, that restriction, not openness, poses the greater risk to US competitiveness, while Sacks argues the same point on defense grounds.
The Dorsey-Palihapitiya-Sacks coalition represents a significant voice among US technology leaders, but it does not reflect consensus in Washington or even in Silicon Valley. Anthropic and other safety-focused AI labs argue that unfettered openness creates genuine risks around biosecurity, cyberattacks, and other misuse scenarios. That debate remains unresolved at the policy level.
Institutional investors tracking AI infrastructure, cloud computing, and cybersecurity face genuine uncertainty about which regulatory path Washington will ultimately choose, a choice that will directly affect the cost structure and competitive positioning of American AI companies over the next 12 to 24 months.
The immediate test will be whether US regulators continue or accelerate model vetting restrictions, or instead follow the Dorsey-Palihapitiya recommendation to embrace open source development. That decision will determine whether American firms absorb a structural 26-to-56x cost disadvantage against foreign competitors, or whether they gain access to the same low-cost capability their global rivals already use. Washington has not yet signaled which direction it will move.
