Morgan Stanley: AI Infrastructure Spending to Reach $140 Billion by 2028

• According to a Morgan Stanley research report, the top five hyperscale cloud providers could spend up to $1.4 trillion on capital expenditures by 2028.
• The cost per GW has been driven up by memory, power, and construction, with computing capacity potentially increasing from 30 GW to 120 GW.
• META is listed as the preferred AI internet platform, with a $775 price target contingent on the successful monetization of APIs, advertising, and subscriptions.

Morgan Stanley raised its capital expenditure estimates for major hyperscale cloud providers in a sell-side research report, projecting that the total capex of the five major platforms will reach $1.2 trillion in 2027 and $1.4 trillion in 2028, while maintaining META as its top pick for AI-driven internet companies, with a target price of $775.

These figures are based on the research report’s methodology and do not equate to the company’s official guidance. The publicly available Morgan Stanley materials have noted that global investment in AI-related infrastructure is expected to approach $3 trillion by 2028, with data center capital expenditures totaling approximately $2.9 trillion. The细分 item of $1.4 trillion for the five major platforms is primarily derived from sell-side analysts’ breakdowns and estimates for leading cloud and internet platforms.

The most newsworthy change in this report is the continued upward revision of AI infrastructure spending. By 2028, the available computational capacity on major platforms is projected to reach nearly 120 GW, approximately four times the 30 GW estimated for 2025. The cost per GW has also been raised, as next-generation platforms such as the GB200, GB300, and Vera Rubin require greater investments in memory, power, racks, and engineering.

For investors, the question has shifted from “Will AI giants spend money?” to “How quickly will that spending turn into revenue?” Meta is positioned prominently because it faces greater AI capital expenditure pressures while also having more direct monetization channels, such as advertising, consumer apps, model APIs, and subscription tools.

The research report raised the projected capital expenditures for the five major hyperscale cloud providers in 2027 and 2028 by 9% and 10%, respectively, to $1.2 trillion and $1.4 trillion. This estimate includes AI infrastructure spending by Amazon, Google, Microsoft, Meta, and SPX.

Capacity expansion is one of the primary drivers of increased spending. Under this model, available computing capacity on major platforms is projected to rise from approximately 30 GW in 2025 to nearly 120 GW by 2028. Amazon is expected to reach about 35 GW by 2028, Google will add the most capacity in 2027 and 2028, and Meta will increase from around 3.5 GW at the end of 2025 to 14 GW in 2027 and 21 GW in 2028.

Projected capital expenditures for the five major cloud providers: $1.4 trillion in total by 2028, with increases of 9% and 10% for 2027 and 2028, respectively, compared to previous estimates.

Available computing capacity is expected to rise from approximately 30 GW in 2025 to nearly 120 GW in 2028, with Meta reaching 21 GW and Amazon totaling approximately 35 GW.

The calculation methodology for Meta’s capital expenditures must retain differences in reporting standards. In the research report model, Meta’s capital expenditures for 2027 and 2028 have been raised to $225 billion and $250 billion, respectively. Some publicly available secondary reports cite Morgan Stanley’s estimate of approximately $380 billion in total capital expenditures for Meta from 2027 to 2028, which may reflect different definitions such as total capex, AI infrastructure spending, aggregate figures, or inclusion of off-balance-sheet financing.

These differences do not alter the main point: AI data center spending continues to weigh on free cash flow, depreciation, and near-term EPS, and will determine whether future revenues from cloud, advertising, search, APIs, and enterprise tools can be realized. Whoever can convert more computing power into billable products will find it easier to justify today’s capital expenditures.

The increased expenses stem not only from "building more data centers" but also from "higher costs per GW."

In the bottom-up cost model of the research report, the construction cost per GW for the GB200 is approximately $35 billion, an increase of 16% from previous assumptions. The GB300 is approximately $39 billion, up 19%. Vera Rubin is approximately $49 billion, up 20%. Google’s TPU v7 is approximately $27 billion, and Amazon’s Trainium3 is approximately $21 billion.

Updated deployment costs for GPU and ASIC GW-scale data centers: GB200 at approximately $35 billion, GB300 at approximately $39 billion, and Vera Rubin at approximately $49 billion.

Cost pressures stem primarily from two sources. The proportion of memory in high-end AI systems continues to rise, and external data center costs—including power, land, cooling, power distribution, and construction—are also increasing. The research report assumes these related costs will rise from approximately $10 million per MW to between $11 million and $19 million per MW.

This is also why the spending curve of AI giants is unlikely to decline in the short term. While improved chip supply can alleviate some pressure, constraints in power access, rack availability, construction, skilled labor, and local approvals will continue to extend project timelines. Some projects may be stretched to around three years, and the larger the capital expenditure, the more urgently revenue must demonstrate a return.

META is listed as the top choice, primarily because its AI revenue options are more concentrated than those of most internet companies.

The research report breaks down Meta’s potential upside into areas such as Meta AI search, new cloud services, API revenue, subscription tools, and advertising upgrades, collectively contributing approximately $10 to EPS by 2028. Under the base case, Meta’s 2028 EPS is projected at $33.41. If some options are exercised, EPS could see additional upside.

Cumulative contribution of five categories of META AI upside options to 2028 EPS, with a base EPS of $33.41, totaling an upside of approximately $10.

This estimate does not fully align with certain publicly reported secondary sources mentioning “four products or catalysts” or an “EPS upside of $1 to $3 by 2028,” and is better understood as a scenario analysis within this research report. The actual impact on financial statements will depend on product adoption rates, pricing power, and computing utilization.

The API is the most direct entry point. On July 9, Meta announced the public preview of the Meta Model API. Third-party pricing trackers, such as Artificial Analysis, indicate that the input and output prices for the Muse Spark 1.1 API are $1.25 and $4.25 per million tokens, respectively, lower than those of some leading competitors.

The research model further assumes that 100 MW of GB300 capacity allocated to APIs corresponds to approximately 53,300 GPUs operating at 75% utilization, generating around $8.59 billion in revenue, $640 million in incremental EBIT, and an incremental $1.91 in 2028 EPS. This estimate relies on high utilization and sustained demand; low pricing alone can help attract customers but cannot guarantee profitability by itself.

The subscription tool is also a potential entry point. The model assumes that 25% of Meta’s 15 million advertisers pay approximately $200 per month for tools such as business agents and coding assistants, contributing about $8 billion in revenue and approximately $2 in 2028 EPS. Whether advertisers are willing to continue paying ultimately depends on whether these tools can deliver higher conversion rates, lower production costs, or greater automation capabilities.

Amazon and Google are also key players in this round of increased capital spending, though they serve more as background references in this main narrative.

Regarding Amazon, the research report raised its revenue growth outlook for AWS, projecting 40% growth in 2027 and 36% in 2028. It also estimated that AWS’s backlog increased by approximately $110 billion quarter-over-quarter in the second quarter to about $475 billion. Since Amazon has not yet released its official second-quarter financial results, this backlog figure should be considered a sell-side estimate. Official filings have confirmed that AWS achieved 28% year-over-year sales growth in the first quarter of 2026, OpenAI added a $100 billion multi-year commitment, and cash capital expenditures continue to rise.

Google's advantage lies in its full-stack capabilities with the Gemini model, TPU, and cloud business. Analyst models show that Google will add the most new capacity among major platforms in 2027 and 2028. Short-term pressure stems from computing resources potentially constraining product scaling, especially when search, cloud services, and model APIs compete for compute power simultaneously.

These clues point to the same real-world issue: AI spending has reached the trillion-dollar level, and the market will increasingly demand direct answers to “how much revenue does each dollar of capital expenditure generate?” Cloud services, AI search, APIs, advertising tools, and enterprise subscriptions will all serve as entry points to validate the return on spending.

This round of capital expenditure increase has clear boundaries.

The first constraint is supply. Chips, HBM memory, racks, power access, and skilled labor all affect construction speed. AI data centers must navigate local approvals, grid upgrades, and construction timelines between planning and commissioning—they cannot be deployed linearly as models assume.

The second constraint is political and regulatory. Large data centers' consumption of electricity, water, and land may trigger local opposition. Energy policies and the pace of local approvals may also shift around the 2026 U.S. midterm elections and the November 2028 presidential election.

The third constraint is demand. META’s API, subscriptions, and ad upgrades remain optimistic scenarios; revenue realization depends on actual customers paying and consistently using the services. Pricing below competitors helps attract customers, but long-term profitability hinges on usage volume, gross margin, and tool ROI.

A $1.4 trillion capital expenditure plan depicts a high-cost growth trajectory. Giants are securing AI computing power in advance, and the market will continue to question when this computing capacity will translate into revenue and profits. Meta’s $775 price target is based on the gradual realization of AI monetization—the hardest step being the conversion of upward EPS revisions in models into actual cash flow on financial statements.