With tech giants pouring so much money into AI, is the return on investment actually good?

What is the actual return on investment for tech giants in this round of AI "money burning"? Morgan Stanley's answer is not pessimistic: model companies providing API services using their own computing power can achieve a return on investment (ROIC) of up to 46%; hyperscale cloud providers renting out GPUs have an ROIC of about 31%, and even model companies relying on third-party infrastructure have an ROIC of about 25%. From the perspective of return on capital, this trillion-dollar AI arms race is not simply a matter of "burning money".

More importantly, AI investments are gradually becoming self-sustaining. Morgan Stanley predicts that the combined operating cash flow of the four giants—Amazon, Google, Microsoft, and MetaQuotes—will increase from $739 billion in 2026 to $1.23 trillion in 2028, while the demand for new debt financing will decrease from $238 billion to $90 billion during the same period. By 2028, the giants' new debt demand will only be equivalent to about 7% of their operating cash flow, and their financial pressure will not worsen in tandem with the expansion of capital expenditures.

However, the frenzied growth in capital expenditures may be nearing its end. The four major hyperscale cloud vendors' data center capital expenditures are projected to rise from $917 billion in 2026 to $1.47 trillion in 2027, a year-on-year increase of approximately 60%, but will only grow by about 12% in 2028. This means that the next phase of the market will no longer be about who can build more data centers, but rather who can convert the already built computing power into revenue and profit.

This is precisely what Morgan Stanley is most optimistic about: as capital expenditure growth slows and AI application penetration increases, funds may gradually shift from hardware, semiconductors, and memory to models, cloud platforms, and software applications. The "return on investment test" for AI is moving from computing power construction to commercialization.

Capital expenditure: peaks in 2027, then slows significantly in 2028.

Morgan Stanley projects that hyperscale cloud vendors' data center capital expenditures will increase from approximately $466 billion in 2025 to approximately $917 billion in 2026, further rising to approximately $1.47 trillion in 2027 and approximately $1.64 trillion in 2028. However, the growth rate of capital expenditures will plummet from approximately 60% in 2027 to approximately 12% in 2028.

Among them, Google's expansion is the most aggressive, with data center capital expenditure expected to increase by 83% year-on-year in 2027; Amazon, Meta and Microsoft's growth rates are about 50%, 55% and 43% respectively, but all slow down significantly by 2028.

Morgan Stanley believes that real-world constraints such as chips, racks, land, electricity, and labor are limiting further expansion, while giants have already pre-built a large number of data centers to meet demand from 2027 to 2029, leaving limited room for further upfront capital expenditures.

Therefore, the key change in 2028 is not that AI demand will peak, but that infrastructure investment will enter a digestion period after a period of rapid expansion.

Computing capacity: nearly quadrupled in three years

Despite a slowdown in capital expenditure growth, computing power will continue to expand. Morgan Stanley predicts that the total computing power of the four major hyperscale cloud providers will increase from approximately 36GW in 2025 to approximately 144GW in 2028, nearly quadrupling.

Google is expected to be one of the companies adding the most capacity, with approximately 9GW and 11GW added in 2027 and 2028 respectively. This capacity will be mainly used to train Gemini, drive GCP growth, and support generative AI features for Search and YouTube.

At the same time, the computing power structure is also changing. Morgan Stanley predicts that the share of custom ASICs in new computing power will rise from 34% in 2025 to 66% in 2028, with Google TPU and Amazon Trainium becoming the main drivers.

In other words, the demand for AI computing power is still growing, but the industry is shifting from "grabbing GPUs" to "improving the efficiency of unit computing power." Nvidia still occupies a core position, but the importance of cloud vendors' self-developed chips continues to rise.

AI Returns on Investment: Own Infrastructure + Model Layer is the Most Profitable

Morgan Stanley calculated the return on capital for three GenAI business models. Under the IaaS model where hyperscale cloud providers lease GPUs, based on GB300, each GW corresponds to approximately $23 billion in revenue and a ROIC of approximately 31%.

Model companies that leverage their own infrastructure to provide API services have the highest return on investment, generating approximately $30.4 billion in revenue per GW and $17.9 billion in after-tax operating profit, corresponding to a ROIC of about 46%. If they rely on third-party infrastructure to provide APIs, although the revenue per GW reaches approximately $40.5 billion, the ROIC after deducting computing power leasing costs is only about 25%.

This means that the most attractive link in the AI industry chain is not necessarily simply renting out computing power, but rather companies that possess modeling capabilities, infrastructure, and commercialization capabilities . Morgan Stanley believes that Meta and Google are prime examples of this model.

Market Potential: AI Commercialization is the Key to the Next Stage

Morgan Stanley estimates that the global GenAI total market size (TAM) will reach $50 trillion to $60 trillion, of which the knowledge work market will be about $20 trillion to $30 trillion and the consumer market will be about $30 trillion.

Based on the public cloud adoption curve analogy, enterprise AI spending is projected to reach approximately $812 billion by 2027, equivalent to about 4% penetration of knowledge work TAM (Total Asset Management). Morgan Stanley believes that AI's diffusion rate may be faster than that of the public cloud because enterprises do not need to undertake large-scale infrastructure migrations, and AI can generate quantifiable productivity gains more quickly.

More importantly, AI has begun to show real signs of commercialization. As of the second quarter of 2026, approximately 25% of S&P 500 companies were able to quantify the benefits brought by GenAI, a significant increase from 14% a year earlier.

This means that AI investment is gradually shifting from "building first and waiting for demand" to a cycle of "computing power investment - application growth - revenue realization". Whether the application layer can continue to contribute revenue and profit will become the core of the next stage of valuation.

Risk warning and disclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.