The desk's latest note argues AI is shifting from scarcity to abundance, and one chart captures the transition in real time
Goldman Sachs’ 1-Delta Desk, the firm’s macro trading and flow commentary unit led by Rich Privorotsky, just told clients to pay very close attention to a single chart. The chart in question tracks the competitive dynamics between open-source and closed-source AI models, and the desk believes it functions as a leading indicator for where the entire AI infrastructure trade is heading.
The note, dated June 9, 2026, carries a deceptively simple thesis: AI is transitioning from an era of scarcity to one of abundance. And when that happens, the companies charging premium prices for proprietary models are going to have a problem.
Open-source models are closing the gap, at half the cost #
The Goldman desk is pointing to experimental results showing that a combination of open-source and lower-cost models, specifically Gemini 3 Flash, Kimi K2.6, and DeepSeek V4 Pro, are now matching or outperforming closed models like GPT-5.5. The kicker: they’re doing it at roughly 50% of the operational cost.
The note emphasizes tracking real-world compute pricing and model performance metrics rather than getting swept up in the prevailing hype cycle.
DeepSeek has reportedly slashed its token prices by 75%. The Goldman desk frames this as an acceleration of “token price wars,” a dynamic where AI providers compete by undercutting each other on the cost of inference.
Supply-side risks are piling up #
The desk identifies two supply-side pressures that could weigh on multiples even as AI adoption continues to accelerate. First, equity issuance. AI companies have been tapping capital markets aggressively, and the resulting dilution creates headwinds for per-share valuations regardless of how quickly revenue grows. Second, local inference. As models become smaller and more efficient, more compute shifts from centralized cloud providers to on-device or on-premise deployment.
What this means for investors #
During scarcity, the winners are whoever controls the bottleneck: GPU manufacturers, cloud providers with exclusive model partnerships, and the handful of labs with frontier capabilities. During abundance, the winners shift to companies that can deploy AI at scale, reduce costs for enterprise customers, and build defensible positions through distribution and data advantages rather than raw model performance.
For traders and portfolio managers, the practical takeaway is to monitor the specific metrics the Goldman desk highlights: real-world compute pricing trends, benchmark performance gaps between open and closed models, and the pace of token price compression across major providers. The re-rating risk is asymmetric. If open-source models continue to close the gap, infrastructure premiums could compress meaningfully. If closed models manage to reestablish a clear performance lead, the current multiples might hold. The question is no longer whether open-source AI can compete. It’s whether closed-model providers can justify charging twice as much when it does.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our