Written by: Rita

Trends Guide

Morgan Stanley significantly raised its capital expenditure forecast for hyper-scale companies, estimating that in 2027/2028, the five major companies (Microsoft, Google, Amazon, Meta, SpaceX) will have a combined capital expenditure of about $1.2 trillion/$1.4 trillion. By 2028, available computing capacity is expected to increase from about 30GW in 2025 to nearly 120GW, a fourfold increase. In this report, Morgan Stanley explicitly lists Meta as the preferred target, citing five market options that have not been adequately priced in, with the most noteworthy being the API business, where the pricing of the Muse Spark 1.1 model is 30%-86% lower than its peers, potentially generating about $8 billion in revenue for every 100MW of computing power. The outcome of the computing power arms race is not who builds faster, but who sells better.

Capital Expenditure Continues to Increase, Expected to Reach $1.4 Trillion by 2028

Morgan Stanley employs a bottom-up cost model, assessing the construction cost of each GW data center from two dimensions: IT hardware cost within racks (chips, memory, CPUs, networking) and external costs (building materials, electrical and mechanical systems). Due to rising memory prices and inflation in external costs, the cost per GW has increased by about 20%. Combined with GPU/ASIC supply distribution, Morgan Stanley raised the total capital expenditure for the five hyper-scale companies in 2027/2028 to around $1.2 trillion/$1.4 trillion. Among these, Meta's capital expenditure for 2027/2028 was raised by 29%/22%, reaching $225 billion/$250 billion; Amazon's total capital expenditure was raised by 15%/29%, reaching $308 billion/$318 billion; Google had already raised its forecast significantly.

Computing Capacity Expected to Quadruple, AWS with the Largest Holdings

Morgan Stanley expects the total available computing capacity of the five hyper-scale companies to reach nearly 120GW by 2028, growing fourfold from about 30GW in 2025. AWS is expected to have the largest capacity in 2028 (35GW), followed closely by Google with 31GW, while Meta is projected to increase from about 3.5GW at the end of 2025 to 14/21GW in 2027/2028. In 2026/2027, 55%/90% of Meta's new capacity comes from in-house construction, with the remainder sourced from third-party leasing.

Meta Muse Spark Pricing is Aggressive, Huge API Revenue Potential

Morgan Stanley focused on analyzing the monetization opportunities of Meta's APIs. The latest AI model Muse Spark 1.1 has opened up API calls to third parties, with pricing set at $1.25 per million tokens for input and $4.25 for output, which is 30%-86% lower than comparable market products. Morgan Stanley constructed an API revenue model: using 100MW of GB300 computing power for API services could generate about $8 billion in revenue and an incremental EPS of about $1.9 per share, equivalent to approximately 6% upside to the 2028 EPS. On the side of paying users, Morgan Stanley assumes that 25% of Meta's 15 million advertisers (about 4 million) will each pay around $200 monthly, potentially generating about $8 billion in annual revenue. Morgan Stanley pointed out that Meta only needs to allocate 100MW from its total capacity of around 21GW by the end of 2028 to support this business, leaving substantial remaining capacity for monetization.

Amazon and Google: Beneficiaries of Infrastructure

Amazon AWS's revenue growth rate for 2027/2028 is expected to reach 40%/36%, with Morgan Stanley projecting an EPS of $11.53/$15.05 for 2027/2028, with a target price of $330. Google is expected to see the largest increase in computing capacity, with an additional 9GW/11GW in 2027/2028. Morgan Stanley maintains an overweight rating on both companies.

Trend Perspective

The decisive factor in the computing power arms race is shifting from "how much is built" to "how much is sold." Over the past two years, the market has focused on who hoarded the most GPUs; moving forward, it will be about who can convert computing power into monetizable revenue streams. Meta received the highest rating in this report. It may not be building the fastest, but it has the most monetization pathways among the five: core advertising, subscription income, NeoCloud computing power rental, API model calls, diffusion models, and business agents. The aggressive pricing strategy of Muse Spark itself is a strategy for maximizing computing utilization. Idle computing power is worthless, while low-cost volume can at least cover marginal costs and build an ecosystem. If previously only capital expenditure and capacity planning were considered for selecting targets, now a new dimension needs to be added: who has already built the sales channels while constructing.

Disclaimer

This article is an整理与解读 of a third-party brokerage research report (Morgan Stanley, July 12, 2026) by Trend Research. The ratings, target prices, earnings forecasts, and related judgments quoted in the text represent the opinions of the analysts from that brokerage firm, do not represent the views of Trend Research, and do not constitute any investment advice.

The market carries risks; decisions should be made independently. This article should not be used as a basis for buying or selling any securities.