Morgan Stanley Report: AI Capital Spending to Increase, Revenue Must Cover Costs

Written by: Rita

AI capital expenditures are still rising, but whether revenue can cover capital costs is the only question that matters. Bank of America believes that the weighted average cost of capital (WACC) for hyperscalers has risen from 4% in 2023 to 10%, and AI returns must surpass this threshold. The S&P 500’s risk-reward balance over the next 12 months favors a long-term bullish outlook on U.S. equities, with the best opportunities likely outside of technology.

According to a report released by Bank of America on October 16, 2026, AI capital expenditures this year exceeded expectations by 50%, while revenue exceeded expectations by only 20%. The projected trend for 2027 shows a clear shift. Computing demand through 2030 corresponds to approximately $10 trillion in spending—$2 trillion higher than current forecasts. Hyperscalers’ free cash flow is expected to turn positive in the second half of 2027, assuming accelerated AI revenue and reduced capital expenditures.

Large-scale vendors' capital expenditures continue to accelerate. Bank of America expects that the combined capital expenditures of the five major large-scale vendors and neoclouds will reach approximately $956 billion in 2026, $1.32 trillion in 2027, and $1.5 trillion in 2028. Capital expenditures among the S&P 500, excluding TMT, are also rising, indicating that AI infrastructure development has surpassed the momentum driven by fiscal stimulus.

From a demand perspective, approximately 212 gigawatts of data center capacity need to be added between 2026 and 2030. Conservatively estimating $40 billion per gigawatt, cumulative spending amounts to about $9.5 trillion—$2.1 trillion higher than the $7.4 trillion currently forecasted for capital expenditures by hyperscalers and new cloud providers. Bank of America believes capital expenditures are still rising.

Bank of America notes that the pace of AI capital expenditures is constrained by bottlenecks, limiting the speed of capacity expansion unless there is an overbuilding surge akin to the late-1990s telecom bubble. While current capital spending as a share of GDP appears late-cycle, it remains mid-cycle when measured against the profitability of spenders.

The WACC of hyperscale vendors rose from 4% in 2023 to 10%. The risk-free rate is more than 4 percentage points higher than its pandemic low, debt costs have increased, equity costs have risen, and the beta of hyperscale vendors has also increased. Bank of America believes that the doubling of capital costs directly raises the return threshold for AI investments.

Economic常识 is that investment continues until the return on incremental invested capital (ROIIC) falls below the WACC. Currently, the ROIIC-WACC for hyperscale manufacturers is negative, capital expenditures as a percentage of sales are rising, invested capital is increasing, and returns have yet to materialize. Bank of America believes that companies facing the heaviest future capital expenditure burdens must demonstrate returns.

Risks include inflation, rising real yields, credit events, increased regulation, and equity supply exceeding demand. Bank of America notes that U.S. household net worth stands at $196 trillion, exceeding estimated AI spending through 2030, with capital still flowing. Policy is more likely to provide support than act as a barrier, as evidenced by Intel’s prior backing.

Bank of America uses scenario analysis to calculate AI investment returns. The key variable is AI revenue per gigawatt. In the bear case, annual revenue per gigawatt is assumed at $90 billion, with a 3% annual change, 5% core revenue growth, AI cash operating costs of $3 billion per gigawatt, and a depreciation period of 6.2 years. In the base case, revenue per gigawatt is $120 billion, with a 2% annual change, 7% core growth, operating costs of $2 billion, and a depreciation period of 8.2 years. In the bull case, revenue per gigawatt is $150 billion, with a 5% annual change, 8% core growth, operating costs of $1 billion, and a depreciation period of 10.3 years.

Under the base case, cumulative excess returns from 2026 to 2030 amount to approximately $200 billion. Under the bear case, cumulative losses total around $500 billion, while under the bull case, cumulative excess returns approach $1 trillion. Bank of America believes that revenue per gigawatt is the key differentiator across scenarios, and the pace of monetization determines whether AI capital expenditures create or destroy value.

The enterprise segment represents the largest monetization opportunity. Bank of America estimates that if 10% of U.S. workers are displaced by AI, the $13 trillion in wages corresponds to a maximum annual AI total addressable market (TAM) of $3 trillion. Enterprise applications account for approximately 80% of the total AI TAM, while consumer subscriptions and digital advertising make up a smaller share.

Bank of America believes the best opportunities for AI capital spending may lie outside of technology. Semiconductors captured the first wave, with the top three semiconductor companies seeing a surge in free cash flow over the next 12 months, while hyperscalers’ free cash flow turns negative. The next wave is physical infrastructure construction: approximately 40% of the cost per gigawatt of AI capacity goes toward networks, power, cooling, and buildings—not servers and GPUs.

Electrification itself could drive copper demand growth comparable to that of all industries over the past decade. Bank of America believes that technology first enables the economy to produce more—such as power grids, power plants, data centers, and factories—allowing suppliers to charge regardless of which AI model or application prevails.

In terms of positioning, Bank of America sees a balanced risk-reward profile for the S&P 500 over the next 12 months and maintains a long-term bullish outlook on U.S. equities. It maintains overweight positions in both information technology and communication services, as valuations have compressed, leading to a more rational competitive landscape. Large-cap tech firms are trading at a 24% discount to the S&P 500, while semiconductors are trading at a 42% discount. Bank of America avoids chasing tech stocks and instead favors AI adopters, commodities, and physical infrastructure.

AI is the new fuel for efficiency-driven bull markets. Bank of America notes that non-farm productivity has surpassed pre-pandemic trends, with real revenue per employee in technology and communication services growing faster than the S&P 500. Financial services and healthcare are well-positioned for large-scale AI deployment; however, financial services are held back by regulatory burdens, while healthcare has a high proportion of clinical trial costs in categories significantly impacted by AI.

Two quantitative factors supported by Bank of America favor efficient investing: companies with high ROIIC-WACC have outperformed their low-spread peers by approximately 1.6 percentage points annually since 1990; low labor-intensity companies have outperformed high labor-intensity companies by approximately 3.6 percentage points annually over the past four decades. AI further rewards businesses that can expand revenue without increasing headcount.

Bank of America believes that upward productivity pressures have lowered the equity risk premium (ERP), supporting valuations. AI adopters, commodities, and physical infrastructure represent the current best opportunities, with greater balance within the technology sector. AI capital expenditures are still growing, but whether revenue generation can cover a 10% cost of capital will determine whether returns are positive or negative.

Disclaimer

This article is a compilation and interpretation by Chaoxiang Research of a third-party brokerage research report (Bank of America, October 16, 2026), combined with publicly available market information. The ratings, price targets, earnings forecasts, and related judgments cited herein reflect the views of the brokerage’s analysts and represent the position of their respective institution only; they do not reflect the views of Chaoxiang Research nor constitute any investment advice.

The market carries risks; make decisions independently. This article should not be used as a basis for buying or selling any securities.