DeepSeek’s R1 Launch Shows There Are No Moats Among Large Language Models

We believe that as the price of LLMs goes down, the value and usage of cloud infrastructure increase.

Dan Romanoff, CPA 28 January, 2025 | 10:32AM
Facebook Twitter LinkedIn

Illustration of AI depicted by a robot in thought with red and green circuit wires extending from its head, representing the robot's cognitive process

DeepSeek, a Chinese artificial intelligence company, released R1, its open-source reasoning model, earlier this month. The model’s capabilities roughly match those of advanced models from OpenAI, Anthropic, and Alphabet GOOGL while having materially lower training costs.

Why it matters: R1’s impressive performance/cost dynamics have raised investor concerns about the necessity of the billions of dollars in capital expenditures that large US tech companies have made (and the billions more they plan to spend) on generative AI.

R1’s launch and its dramatically lower pricing (more than 90% below OpenAI’s latest reasoning model) go hand in hand with our broader “commodification of complements” view of the large language model space.

We believe that as the price of LLMs (the complementary good) goes down, the value and usage of the public cloud vendors’ primary good, cloud infrastructure, increase. To that end, we believe Amazon AMZN, Microsoft MSFT, and Google benefit from reduced LLM pricing in the long run.

The bottom line: We maintain our fair value estimates for Microsoft ($490 per share), Amazon ($200), and Alphabet ($220), and see these wide-moat firms as benefiting from a commodified LLM layer, with increased spending on AI creating tailwinds for their public cloud businesses.

While we expect capital expenditures at public cloud vendors to remain elevated in the near term, we see them as primarily geared toward serving generative AI demand, which should flourish with lower pricing of LLMs, as opposed to training leading-edge models.

At the same time, we expect these large US tech companies to replicate some of the AI techniques DeepSeek leveraged to drive down the cost of R1 to reduce the costs of their model training and inference, potentially lowering their medium- and long-term capital expenditures.


The author or authors do not own shares in any securities mentioned in this article. Find out about Morningstar's editorial policies.

The information contained within is for educational and informational purposes ONLY. It is not intended nor should it be considered an invitation or inducement to buy or sell a security or securities noted within nor should it be viewed as a communication intended to persuade or incite you to buy or sell security or securities noted within. Any commentary provided is the opinion of the author and should not be considered a personalised recommendation. The information contained within should not be a person's sole basis for making an investment decision. Please contact your financial professional before making an investment decision.

Facebook Twitter LinkedIn

About Author

Dan Romanoff, CPA  is an equity research analyst on the technology, media, and telecommunications team for Morningstar in Chicago.

© Copyright 2025 Morningstar, Inc. All rights reserved.

Terms of Use        Privacy Policy        Modern Slavery Statement        Cookie Settings        Disclosures