摩根士丹利：AI基建豪掷1.4万亿美元，META的“算力账”能回本吗？

TL;DR

• Morgan Stanley estimates that the five major hyperscale cloud providers could see CAPEX reach $1.4 trillion by 2028.
• Per-GW construction costs are being driven up by memory, power, and construction, with computing capacity potentially expanding from 30GW to 120GW.
• META is listed as the top pick for AI internet, with a $775 price target contingent on the realization of API, advertising, and subscription monetization.

Morgan Stanley, in a sell-side research report, has raised its capital expenditure estimates for major hyperscale cloud providers, projecting total CAPEX for the five major platforms to reach $1.2 trillion in 2027 and $1.4 trillion in 2028. It continues to list META as the top AI internet pick, maintaining a price target of $775.

These figures are based on the research report's model estimates and do not equate to official company guidance. Public versions of Morgan Stanley materials have already mentioned that global AI-related infrastructure investment could approach $3 trillion by 2028, with data center CAPEX around $2.9 trillion. The specific figure of $1.4 trillion for the five major platforms stems more from the sell-side's granular estimates for major cloud and internet platforms.

The most newsworthy change in this report is the continued upward revision of AI infrastructure spending. By 2028, the available computing capacity of major platforms in the model approaches 120GW, roughly four times the 30GW estimated for 2025. Per-GW construction costs have also been revised upwards, as next-generation platforms like GB200, GB300, and Vera Rubin require more memory, power, racks, and engineering investment.

For investors, the question has shifted from "Will AI giants spend the money?" to "How quickly will this spending translate into revenue?" META is positioned favorably because it faces higher AI CAPEX pressure while possessing more direct monetization channels like advertising, consumer applications, model APIs, and subscription tools.

$1.4 Trillion in Spending Bets on 120GW of Computing Power

The report raises CAPEX expectations for the five major hyperscale cloud providers by 9% and 10% for 2027 and 2028, respectively, reaching $1.2 trillion and $1.4 trillion. This estimate covers AI infrastructure spending by Amazon, Google, Microsoft, META, and related SPCX entities.

Capacity expansion is a primary driver of the increased spending. In this model, the available computing capacity of major platforms rises from approximately 30GW in 2025 to nearly 120GW by 2028. Amazon's total capacity is projected to be around 35GW by 2028, Google adds the most capacity in 2027 and 2028, while META's capacity grows from roughly 3.5GW at the end of 2025 to 14GW in 2027 and 21GW in 2028.

CAPEX forecast for the five major hyperscale cloud providers, totaling $1.4 trillion in 2028, with upward revisions of 9% and 10% for 2027 and 2028, respectively.

Available computing capacity rises from approximately 30GW in 2025 to nearly 120GW by 2028, with META reaching 21GW and Amazon totaling around 35GW.

Discrepancies in META's CAPEX estimates need careful consideration. In the report's model, META's CAPEX for 2027 and 2028 is raised to $225 billion and $250 billion, respectively. Some publicly cited secondary reports mention a Morgan Stanley estimate of META totaling about $380 billion from 2027 to 2028, which might involve different scopes like total CAPEX, AI infrastructure, gross amounts, or including off-balance-sheet financing.

These differences don't alter the main theme: AI data center spending continues to pressure free cash flow, depreciation, and short-term EPS, while also determining whether future revenue from cloud, advertising, search, APIs, and enterprise tools can be realized. The entity that can convert more computing power into chargeable products will have an easier time justifying today's capital expenditures.

Per-GW Costs Rise as Memory and Power Infrastructure Raise the Bar

The upward revision in spending isn't solely due to "building more data centers" but also because "each GW costs more."

In the report's bottom-up cost model, the per-GW construction cost for GB200 is approximately $35 billion, a 16% increase from the previous assumption. GB300 costs around $39 billion (up 19%), Vera Rubin about $49 billion (up 20%), Google's TPU v7 around $27 billion, and Amazon's Trainium3 about $21 billion.

Updated GPU and ASIC GW-level data center deployment costs: GB200 ~$35B, GB300 ~$39B, Vera Rubin ~$49B.

Cost pressures primarily stem from two areas. The memory portion of high-end AI systems continues to increase, while costs for the data center shell – including power, land, cooling, power distribution, and construction – are also rising. The report assumes related costs increase from roughly $10 million/MW to between $11 million and $19 million/MW.

This explains why AI giants' spending curves are unlikely to decline in the short term. While improved chip supply can alleviate some pressure, power access, rack deployment, construction, skilled labor, and local permitting will continue to lengthen build-out timelines. Some project timelines may extend to around three years, and the larger the CAPEX, the faster the revenue side needs to demonstrate returns.

META's Focus Shifts to AI Monetization

META is listed as the top pick primarily because its AI revenue optionality is more concentrated than most internet companies.

The report breaks down META's potential upside into areas like Meta AI search, new cloud services, API revenue, subscription tools, and advertising upgrades, which could collectively contribute approximately $10 to the 2028 EPS. In the base case scenario, META's 2028 EPS is $33.41. If some of these options materialize, there is further upside potential for EPS.

Cumulative contribution of META's five AI upside options to 2028 EPS. Base EPS is $33.41, with a combined upside of approximately $10.

This estimate may not perfectly align with publicly cited secondary reports mentioning "four products or catalysts" or "EPS upside of $1 to $3 in 2028." It is more appropriately viewed as a scenario estimate within this specific report. The actual impact on financial statements will depend on product adoption rates, pricing power, and compute utilization.

APIs represent the most direct entry point. On July 9, Meta announced the public preview of the Meta Model API. Third-party information from pricing trackers like Artificial Analysis indicates that the input and output prices for Muse Spark 1.1 API are $1.25 and $4.25 per million tokens, respectively, lower than some leading competitors.

The report's model further assumes that every 100MW of GB300 capacity dedicated to APIs (equivalent to ~53,300 GPUs at 75% utilization) could generate approximately $8.59 billion in revenue, $640 million in incremental EBIT, and contribute about $1.91 to META's 2028 EPS. This estimate relies heavily on high utilization and sustained demand. Low pricing alone can help attract customers but cannot guarantee profitability by itself.

Subscription tools are another potential entry point. The model assumes 25% of META's 15 million advertisers pay roughly $200 per month for tools like business agents and coding assistants, potentially contributing about $8 billion in revenue and approximately $2 to the 2028 EPS. Whether advertisers will consistently pay ultimately depends on whether these tools can deliver higher conversion rates, lower production costs, or stronger automation capabilities.

Amazon and Google Benefit, but Revenue Verification Must Follow

Amazon and Google are also key players in this round of CAPEX increases, though they serve more as background references in this narrative.

For Amazon, the report raises its AWS revenue growth outlook, projecting growth of 40% in 2027 and 36% in 2028. It also estimates that AWS's backlog increased by approximately $110 billion quarter-over-quarter in Q2 to roughly $475 billion. As Amazon has not yet released its corresponding official Q2 earnings report, this backlog figure should be treated as a sell-side estimate. Officially confirmed documents show AWS sales grew 28% year-over-year in Q1 2026, OpenAI made an additional $100 billion multi-year commitment, and cash CAPEX continues to rise.

Google's strength lies in its full-stack capabilities with the Gemini model, TPUs, and cloud business. The report's model indicates that Google will add the most capacity among major platforms in 2027 and 2028. A short-term pressure point is that computing resources may still constrain product scaling, especially when search, cloud services, and model APIs compete for compute power simultaneously.

These threads point to a common real-world issue: AI spending has entered the trillion-dollar realm, and the market will increasingly directly question "how much revenue does each dollar of CAPEX generate?" Cloud services, AI search, APIs, advertising tools, and enterprise subscriptions will all become entry points for validating spending returns.

Massive Spending Must Navigate Power, Approvals, and Real Demand

This round of CAPEX increases has clear boundaries.

The first constraint is supply. Chips, HBM memory, racks, power access, and skilled labor all impact construction speed. AI data centers, from planning to operation, must navigate local permitting, grid upgrades, and construction cycles, and cannot be linearly deployed as per model assumptions.

The second constraint is politics and regulation. The significant demands of large data centers on power, water resources, and land may encounter local resistance. Energy policy and local approval rhythms could also shift around the 2026 US midterm elections and the November 2028 presidential election.

The third constraint is demand. META's API, subscription, and advertising upgrades remain upside scenarios; revenue realization requires real customer payments and sustained usage. Lower prices than competitors help attract customers, but long-term profitability depends on usage volume, gross margins, and tool ROI.

The $1.4 trillion CAPEX scenario paints a picture of a high-cost growth curve. Giants are pre-locking AI computing power, and the market will continue to press for when this compute translates into revenue and profit. META's $775 price target rests on the gradual realization of AI monetization, and the hardest step remains turning model-driven EPS upside into actual cash flow on financial statements.