Goldman warns declining token prices may hinder investment growth
Summary
The recent statement from Goldman Sachs highlights troubling trends in the hyperscaler market, specifically noting a significant drop in return on invested capital (ROIC) to an all-time low. Factors contributing to this decline include the impact of competitive open-source AI models from Chinese developers, such as Moonshot AI’s Kimi K3, which are not only narrowing the performance gap with leading frontier models but also driving down average token prices. This situation creates a pressing need for token demand growth to exceed the decreased token prices in order to sustain investment in hyperscaler compute infrastructure.
Analysis
Goldman: Goldman Sachs is a major global investment bank and financial services firm known for its equity research and market analysis across technology and infrastructure sectors. Its strategists have recently examined the economics of AI compute investments, noting how declining average token prices driven by open-source and competitive models may challenge sustained demand growth for frontier AI infrastructure. Token Economics: Competitive open-source alternatives are contributing to reduced average token prices, potentially requiring stronger demand growth to maintain investment momentum in hyperscaler AI infrastructure. AI Model Competition: Open-weight models from Chinese developers such as Moonshot AI’s Kimi K3 have narrowed the capability gap with leading closed frontier models while offering substantially lower costs. Infrastructure Returns Focus: Goldman Sachs research frameworks emphasize the need for hyperscalers to achieve sufficient revenue from AI services to support their ongoing capital investments in compute capacity.
Categories
cryptorwaai
Related sources
- https://news.futunn.com/en/post/1000215519/as-ai-capital-spending-surges-toward-1-7-trillion-what
- https://voxdev.org/topic/technology-innovation/will-open-source-ai-keep-long
- https://ground.news/article/goldman-sees-hyperscaler-ai-capex-rising-50-to-12-trillion
- https://finance.biggo.com/news/00d0de5b-5dfe-461d-969b-d3161848ec79
- https://www.bloomberg.com/news/articles/2026-09-24/goldman-sachs-is-underweight-hyperscalers-on-debt-supply-surge
- https://arstechnica.com/ai/2026/09/exclusive-open-chinese-models-close-gap-with-silicon-valleys-frontier-ai-models/
- https://raw.githubusercontent.com/huggingface/blog/main/state-of-open-models-summer-2026.md
- https://www.lpl.com/research/weekly-market-commentary/can-hyperscalers-earn-their-ai-ambitions.html
- https://www.gate.com/post/status/22102741
- https://www.bloomberg.com/news/articles/2026-09-25/goldman-sees-hyperscaler-ai-capex-rising-50-to-1-2-trillion
- https://www.thewealthadvisor.com/article/goldman-sachs-says-ai-can-keep-booming-doesnt-mean-todays-winners-keep-winning
- https://www.bitget.com/news/detail/12560605873053
- https://www.sls.guru/blog/how-open-weight-models-are-changing-the-ai-race
- https://openuk.uk/wp-content/uploads/2026/07/AI-Openness-Report-July-2026-1.1.pdf
- https://www.pymnts.com/big-tech/2026/goldman-sachs-projects-big-tech-ai-infrastructure-spending-to-hit-1-2-trillion-in-2027/