MS The Model Race Gets More Complicated Chinese Model Pricing & Open Weights Adoption

M August 12, 2026 04:01 AM GMT

Global Thematics | North America Morgan Stanley & Co. LLC Idea

Adoption We address three key open-weights model developments from the past week. The landscape is evolving quickly, but evidence increasingly points to a multi-model world, with open-weights driving tangible enterprise ROI. We see the US/China model price differential as a key swing factor to watch.

Key Takeaways Investors remain hyper-focused on what model type prevails – closed, open, or hybrid – and what it means for AI adoption, capex, and winners/losers.

This is an evolving debate, so we address three key developments from this past week.

Chinese model price increases show we're entering an era of better monetization. We think the US/China pricing gap will narrow, as US players cut prices.

Recent US enterprise data points show the world is multi-model. Open models are driving ROI/saving costs, helping to dispel fears of a tokenmaxxing reduction.

Meta re-entering the open-weight game through Muse Spark 1.2 reads positively for the US ecosystem in the case of a Chinese model ban/sovereign AI world.

The open vs. closed-weight model plot thickens. As we wrote here last week, a key debate is whether closed or open-weight models win, or we live in a hybrid world. We believe this is one of the most important questions right now, as it has implications for 1) token costs, 2) Jevons Paradox / the speed of enterprise adoption, 3) hyperscaler capex, and 4) what segments of the tech stack benefit most ( Exhibit 2 ). Investors are still hyper-focused on the end-state of the world and remain Morgan Stanley does and seeks to do business with companies covered in Morgan Stanley Research. As a result, concerned that compute and AI infrastructure demand will fall with the proliferation investors should be aware that the firm may have a conflict of of open-weight Chinese models. Since publishing our note last week, we've seen interest that could affect the objectivity of Morgan Stanley Research. Investors should consider Morgan Stanley three important developments: 1) rising Chinese open-weights model prices and Research as only a single factor in making their investment decision. commercial licensing agreements, 2) greater open-weights adoption data points For analyst certification and other important disclosures, from US enterprises, and 3) Meta re-entering the open-weights game, each of which refer to the Disclosure Section, located at the end of this we address below. Our view continues to be that open-weights models are good for report.

competition and can enable faster AI diffusion, supported by Jevons Paradox. We are seeing that the world is increasingly multi-model, with open-weight models starting to drive tangible ROI and save costs, but the US/China model price differential will be a key swing factor to watch.

First, we are in a unique moment where prices for Chinese models – which were historically viewed as cheap and small – are rising, while frontier model prices are declining. As our China Internet team led by Gary Yu outlined, we are entering an era where Chinese models are becoming larger, with better monetization, higher barriers to entry, and vendors offering both high-end and low-end model options (read more here: China AI Foundation Models: Intelligence War Over Price War (9 Aug 2026)). These Chinese model price increases are occurring through two avenues:

• Direct price increases from players like DeepSeek to become more price rational (price increase is undisclosed, but reported to be "significant"). To contextualize, our analysis indicates that a 2x increase in DeepSeek V4-Flash prices (from $0.14 per 1M input tokens / $0.28 per 1M output tokens) would lift margins for a 200MW data center running DeepSeek V4-Flash in-line with Kimi K3 and other frontier models, which we see as an effort to get closer to industry margin profiles. We'd note that the DeepSeek price increases don't apply to customers who download the model and fine-tune it, only to customers consuming via API. • Commercial licensing agreements from companies like Moonshot and Alibaba. These require enterprise customers making $20M+ off these models to negotiate…