Morgan Stanley Warns Open-Weight AI Models May Pressure Pricing

Analysts at Morgan Stanley report that the proliferation of lower-cost open-weight AI models may pressure pricing and returns for model developers. These models, which allow users to customize and deploy weights on their own infrastructure, are expected to accelerate enterprise AI adoption and drive token prices lower. Meta Platforms Incorporated's Muse models are identified as a key indicator of this pricing pressure and market adoption.

Despite these competitive pressures, the bank estimates that model providers could still achieve returns on invested capital between 20% and 60% when utilizing Nvidia GB300 infrastructure. Hyperscalers renting GPU capacity are projected to see returns between 23% and 39%. Morgan Stanley suggests that low-priced model access may function as a loss leader to attract more profitable cloud spending on storage, security tools, and managed APIs.

The bank maintains Overweight ratings on Amazon.com and Alphabet Inc., noting that both companies use proprietary hardware—specifically Trainium chips and tensor processing units—to reduce compute costs and protect margins.