Morgan Stanley raises 2028 cloud capex forecast to $1.4 trillion

The five largest cloud providers will spend $1.4 trillion on AI infrastructure in 2028, quadrupling available computing capacity to 120GW, Morgan Stanley estimates.

Morgan Stanley raised its 2027 and 2028 capital expenditure forecasts for Meta, Amazon, Microsoft, Google and SpaceX by 9% and 10%, projecting total spending of $1.2 trillion and $1.4 trillion, respectively. The upward revision reflects a roughly 20% increase in GPU construction costs and data center build timelines stretching to three years, the sell-side research report shows.

"Available computing capacity is modeled to approach 120GW by 2028, roughly quadrupling from 30GW in 2025," the Morgan Stanley analysts wrote in the July 14 report. "Construction cost per GW has been upgraded as newer generation platforms require more memory, electricity, racks and engineering investment."

Meta's capacity is projected to rise from about 3.5GW at the end of 2025 to 14GW in 2027 and 21GW in 2028, while Amazon's total capacity is expected to reach roughly 35GW. Google is forecast to add the most new capacity in 2027 and 2028. Per-gigawatt construction costs have climbed sharply: GB200 deployments now cost approximately $35 billion per GW, up 16% from prior assumptions, while Vera Rubin stands at roughly $49 billion per GW, a 20% increase. Google's TPU v7 is estimated at about $27 billion per GW, and Amazon's Trainium3 at around $21 billion.

For investors, the question has shifted from whether hyperscalers will spend to how quickly that spending translates into revenue. Meta is positioned as the top pick with a $775 price target, contingent on monetization from APIs, advertising upgrades and subscription tools that could contribute roughly $10 to 2028 earnings per share. Amazon and Google also stand to benefit, but revenue validation remains the core challenge as the industry enters a trillion-dollar investment cycle.

Cost Per GW Rises as Memory and Power Constraints Bite

The spending increase is not simply about building more data centers — each gigawatt is becoming more expensive. Cost pressures come primarily from two areas: high-bandwidth memory in advanced AI systems and the "shell" costs of data centers, including power, land, cooling and construction. The report assumes these related costs have risen from roughly $10 million per MW to a range of $11 million to $19 million per MW.

This dynamic makes it difficult for AI giants' expenditure curves to decline in the near term. While improved chip supply can ease some pressure, factors like power availability, skilled labor and local permitting continue to extend construction timelines. Some projects now take roughly three years from planning to commissioning.

Meta's AI Monetization Path Faces Its First Real Test

Meta's status as the top pick hinges on whether its massive computing investment can generate measurable returns. The report breaks down potential upside into five areas: Meta AI search, new cloud services, API revenue, subscription tools and advertising upgrades. Together, these could contribute approximately $10 to Meta's 2028 earnings per share, on top of a base case of $33.41.

APIs represent the most direct monetization channel. Meta opened the Meta Model API public preview on July 9, with pricing tracker Artificial Analysis showing input and output prices of $1.25 and $4.25 per million tokens for Muse Spark 1.1 — below some leading competitors. The report's model assumes every 100MW of GB300 capacity allocated to the API, corresponding to about 53,300 GPUs at 75% utilization, could generate roughly $8.59 billion in revenue and contribute about $1.91 to 2028 EPS. That calculation depends heavily on sustained high utilization and customer demand.

Subscription tools offer another potential revenue stream. The model assumes 25% of Meta's 15 million advertisers pay roughly $200 per month for tools like business agents and coding assistants, contributing about $8 billion in revenue and roughly $2 to 2028 EPS. Whether advertisers maintain that spending depends on whether these tools deliver higher conversion rates and stronger automation capabilities.

Amazon and Google Face the Same Revenue Question

Amazon and Google are also significant beneficiaries of this capex cycle, though they face similar pressure to demonstrate returns. Morgan Stanley raised its AWS revenue growth outlook to 40% and 36% in 2027 and 2028, respectively, and estimates AWS's backlog increased by roughly $110 billion quarter-over-quarter in the second quarter to about $475 billion. AWS sales grew 28% year-over-year in the first quarter of 2026, and OpenAI made an additional $100 billion multi-year commitment.

Google's advantage lies in its full-stack capabilities spanning the Gemini model, TPUs and cloud business. The report shows Google is expected to add the most new capacity among major platforms in 2027 and 2028. A near-term constraint is that computing resources may still limit product scaling, particularly when search, cloud services and model APIs compete for computing power simultaneously.

Supply, Regulation and Demand Cap the Spending Trajectory

This round of capital expenditure increases has clear boundaries. On the supply side, chips, HBM memory, racks, power access and skilled labor all constrain the speed of construction. On the regulatory side, large-scale data centers face local resistance over power, water and land use, and energy policy could shift around the 2026 US midterm elections and the November 2028 presidential election. On the demand side, Meta's APIs, subscriptions and advertising upgrades remain upside scenarios — revenue realization requires actual customer payments and sustained usage.

The $1.4 trillion in capital expenditure paints a picture of a high-cost growth curve. The hyperscalers are securing AI computing power ahead of time, and the market will continue pressing for answers on when that computing power translates into revenue and profit. Meta's $775 price target is built on the gradual realization of AI monetization, but the hardest step remains turning the earnings upside in the model into cash flow in the financial statements.

Meta shares, trading at roughly 23 times forward earnings, will need to demonstrate that its computing capacity can generate advertising revenue, API usage and subscription fees at scale. Nvidia, whose GPUs power much of this infrastructure buildout, stands to benefit from sustained demand regardless of which hyperscaler captures the most revenue. The broader market implications extend to data center REITs, power utilities and networking equipment providers that supply the physical layer of this expansion.

This article is for informational purposes only and does not constitute investment advice.