Morgan Stanley raises hyperscaler capex forecasts
Morgan Stanley sharply increased its capital spending outlook for five hyperscalers — Microsoft, Google, Amazon, Meta, and SpaceX — in a report dated July 12, 2026. TechFlowPost said the bank now expects their combined capex to reach about $1.2 trillion in 2027 and $1.4 trillion in 2028.
The report used a bottom-up cost model that split data center construction into in-rack IT hardware, including chips, memory, CPUs, and networking, and out-of-rack costs such as building materials and electrical and mechanical systems. Morgan Stanley said the cost to build each GW of data center capacity has risen about 20% because of higher memory prices and inflation in out-of-rack costs, with GPU and ASIC supply allocation also feeding into the revision.
For Meta, the bank raised its 2027 and 2028 capex forecasts by 29% and 22% to $225 billion and $250 billion. For Amazon, full-company capex estimates were lifted by 15% and 29% to $308 billion and $318 billion. Google had already seen a large increase in prior projections.
Compute capacity could quadruple by 2028
Morgan Stanley expects total available compute capacity across the five companies to approach 120GW by 2028, up from about 30GW in 2025.
AWS is projected to hold the largest capacity in 2028 at 35GW, followed by Google at 31GW. Meta is expected to expand from roughly 3.5GW at the end of 2025 to 14GW in 2027 and 21GW in 2028. The report said 55% of Meta’s added capacity in 2026 and 90% in 2027 would come from self-built infrastructure, with the rest from third-party leases.
Meta’s API business is the key call
Morgan Stanley made Meta its top pick and centered much of the thesis on API monetization. According to the report, Meta’s latest AI model, Muse Spark 1.1, is already available to third parties through an API, priced at $1.25 per million input tokens and $4.25 per million output tokens. That is 30% to 86% below comparable products in the market.
Under Morgan Stanley’s API revenue model, every 100MW of GB300 capacity assigned to API services could generate about $8 billion in revenue and add about $1.9 per share in EPS, equal to roughly 6% upside to 2028 EPS. On the paid-user side, the bank assumed that 25% of Meta’s 15 million advertisers — about 4 million — would pay around $200 a month, which would also translate into roughly $8 billion in annual revenue.
The report said Meta would need only 100MW out of its projected 21GW of total capacity at the end of 2028 to support that business, leaving substantial room for further monetization.
Amazon and Google remain Overweight
Morgan Stanley also described Amazon and Google as infrastructure beneficiaries. It forecast AWS revenue growth at 40% in 2027 and 36% in 2028, and estimated Amazon EPS at $11.53 in 2027 and $15.05 in 2028, with a $330 price target. Google, meanwhile, was described as the company with the largest increase in compute capacity, adding 9GW in 2027 and 11GW in 2028. The bank kept both names at Overweight.
TechFlowPost said the report’s main takeaway is that the competitive focus is shifting from how much capacity companies can build to how much revenue they can generate from it. The article also noted that it was a整理与解读 of a third-party brokerage report from Morgan Stanley dated July 12, 2026, and that the ratings, price targets, earnings forecasts, and related judgments cited were the views of the bank’s analysts, not investment advice.
