The popularity of AI agents or limited by computing power bottlenecks is still optimistic that Meta (META.US)'s long-term commercialization prospects will raise the target price to $820 and maintain an “gain” rating

The Zhitong Finance App learned that Morgan Stanley's latest research report indicates that with the gradual spread of artificial intelligence (AI) smart applications, the GPU computing power required to support their operation may become an important factor limiting the expansion of the number of users. The bank estimates that for every additional 100 million daily active users, Muse, an AI agent under Meta Platforms (META.US), may require about 1 to 4 gigawatts (GW) of GPU computing capacity. Although huge infrastructure demand means that AI agents may become more popular than market expectations, Morgan Stanley is optimistic about Meta's long-term commercialization potential, raised its target price from $775 to $820, and maintained an “gain” rating.

Morgan Stanley said in a report released on October 18 that the market is currently paying more attention to Muse's demand for CPU resources, but as the scale of smart device usage expands, GPUs actually consume more electricity, so they are more likely to become a key constraint in determining the speed of product expansion.

Since the specific parameter scale, interaction frequency, and token processing speed of the Spark and Wonder models behind Muse are still unclear, the bank carried out sensitivity analysis through different scenarios. Assuming that 100 million daily active users use an average of 25% of the weekly free token quota, the GPU computing capacity required for Muse may reach about 1 to 4 gigawatts, depending on how efficiently the model processes tokens per second.

Morgan Stanley pointed out that the efficiency of model inference will directly affect the infrastructure costs of AI agent services. If a cutting-edge model with huge parameters is used, the computing power requirement will be significantly higher than in the scenario where a lightweight model is used. Different agent tasks may also be handled by models of different scales, so actual requirements depend on Meta's future model architecture and task allocation method.

This estimate means that AI agents still face considerable infrastructure challenges to achieve large-scale applications for hundreds of millions or even billions of users.

Morgan Stanley expects Meta to add about 13 gigawatts of total computing capacity from 2027 to 2028 to support its various AI products and services. However, based on the assumption that each GPU processes about 10,000 tokens per second, if Muse has 500 million daily active users and uses 50 million tokens per week, the corresponding service may require about 13 gigawatts of computing capacity, which is equivalent to the total computing capacity Meta expects to add in the next two years.

This means that even if Meta continues to invest in data centers on a large scale, the rapid spread of single smart products may consume a large amount of additional computing power resources, thereby limiting the development space for other AI businesses.

Morgan Stanley believes that improving model efficiency and reducing service costs per user will be the key to promoting the popularization of smart devices. Tech giants such as Google (GOOGL.US) and Amazon (AMZN.US) also need to balance computing power investment and return on investment, while smaller competitors may face greater pressure to expand infrastructure.

At the same time, the bank reminds investors that the user growth curve for AI agents may be longer than expected, and may even flatten out at some stages.

Using the development history of e-commerce as a reference, the report points out that in 2000, when e-commerce reached its fifth year, it accounted for only about 4% of adjusted retail sales; even by 2026, this ratio was only about 28%.

Although AI agents do not need to build physical logistics infrastructure on the same scale as e-commerce, they still need to continuously increase computing capacity, develop more efficient models, and push enterprises to change existing business systems and user interaction methods.

However, the huge demand for computing power also means that global AI capital expenditure may remain high for many years to come. Morgan Stanley pointed out that the chip and memory production capacity that major technology companies can obtain in 2027 has basically been locked in advance. Although the annual growth rate of AI capital expenditure may gradually slow down due to factors such as base expansion, supply chain, and actual construction capacity, the long-term expansion demand for smart applications is expected to support a high level of investment for several years.

From a commercialization perspective, Morgan Stanley believes that AI agents still have huge long-term market space. The bank estimates that around $30 trillion in global consumer spending is expected to be digitized through smart services, of which about 55% is related to e-commerce and travel.

As AI agents gradually become capable of product search, shopping decisions, travel planning, and transaction execution, related technologies are expected to change the way consumers interact with businesses. To achieve this goal, enterprises also need to upgrade existing Internet systems and establish technical architectures that support secure interaction and transactions between agents. Morgan Stanley believes this will create commercial opportunities for the Meta Enterprise Platform being developed by Meta, enabling enterprise customers to use agent-related tools through the platform.

In terms of short-term product performance, Muse still faces the challenge of insufficient user retention. According to Sensor Tower data and Morgan Stanley's analysis, Muse currently still receives about 300,000 new downloads every day in the US, but there are only about 2.5 million daily active users, and the retention rate after 14 days of installation is about 20%, which is significantly lower than mature apps such as Instagram, Facebook, and Google Search.

Morgan Stanley believes this reflects that although Muse can attract users to try it, it has not yet established a stable enough daily usage habit. Whether it can add practical scenarios, launch more convenient application connection functions, and support commercial activities such as shopping and travel planning in the future will be an important factor driving the growth of daily active users.

Although commercialization of Muse is still in its early stages, Morgan Stanley is more optimistic about Meta's core advertising business and emerging AI revenue opportunities. The bank raised the Meta target price from $775 to $820, calculated at a closing price of $709.33 on October 16, which meant a potential increase of about 15.6% and maintained an “overbearing” rating.

The bank said that this price increase is mainly based on improved performance in Meta's core advertising business and the incremental revenue that Muse and corporate AI subscription services may bring in the future. The bank raised Meta's 2027 and 2028 revenue forecasts by about 1% and 2%, respectively, and raised the US GAAP (GAAP) earnings per share forecast by about 1% and 3%, respectively, to $34.04 and $35.76, respectively. Among them, AI-driven advertising products are expected to continue to improve the return on investment of advertisers and drive Meta advertising revenue growth. Morgan Stanley raised its 2027 and 2028 ad revenue forecasts by about 1%, respectively, while raising payment and other business revenue expectations to reflect the potential contributions of Muse and enterprise AI subscription services.

Although Meta is expected to increase sales and related expenses to promote corporate AI platforms, Morgan Stanley believes that the company can still achieve profitable growth while maintaining large-scale AI investments. The bank currently expects Meta's total revenue in 2027 and 2028 to reach approximately US$36.1 billion and US$381.8 billion, respectively, with GAAP earnings per share of approximately US$34 and US$36, respectively.

In terms of valuation, the target price of $820 corresponds to the price-earnings ratio of about 23 times Meta's average expected earnings per share in 2027 and 2028. Morgan Stanley believes this valuation is reasonable considering the long-term commercialization potential of new businesses such as Muse, subscription services, and APIs.

Under an optimistic scenario, if Muse continues to launch more useful features to drive growth in user size, usage frequency, and commercial revenue, Meta's stock price is expected to reach 1,000 US dollars, and there is a potential increase of about 41% compared to the closing price on October 16.

However, Morgan Stanley also cautioned that large-scale adoption of AI agents will not happen overnight. Whether Meta can reduce computing power costs by improving model efficiency in the future and turn user growth into actual revenue will be a key factor in determining its AI return on investment and long-term valuation level.