Morgan Stanley’s 2027 AI infrastructure model puts Nvidia at the center of a spending wave that could account for a large share of S&P 500 earnings growth, while also increasing pressure on data-center power supplies and corporate credit markets. The bank estimates that roughly $690 billion of supplier revenue tied to AI capital spending would flow to Nvidia, equal to 67% of the approximately $1.03 trillion expected to reach S&P 500 companies.
The estimate leaves the other 499 S&P 500 constituents sharing about $339 billion in AI-linked supplier revenue. Morgan Stanley calculates that the full spending cycle would contribute roughly 25% of the index’s sales growth and 78% of its earnings growth in 2027, placing unusual weight on one chip designer’s ability to meet demand and preserve margins.
The projection covers a $1.58 trillion AI budget across Alphabet, Amazon Web Services, Microsoft, Meta, Oracle, SpaceX, CoreWeave, Nebius and a pool of leased computing capacity. It tracks spending through semiconductors and servers, networking equipment, optical components, data-center infrastructure, storage systems and overseas suppliers.
Nvidia dominates the projected S&P 500 revenue flow
Semiconductors and servers account for the largest category in Morgan Stanley’s model, receiving an estimated $672 billion in 2027. Nvidia-linked demand represents the bulk of that spending, reflecting its position as the leading supplier of advanced accelerators used to train and run large AI models.
Morgan Stanley estimates that 91.7% of Nvidia’s 2027 revenue would be linked to AI infrastructure buildouts. Several companies serving adjacent parts of the data-center supply chain also show high exposure in the bank’s model: Erock at 91%, Vertiv at 90%, Applied Optoelectronics at 83%, Broadcom at 81% and Bloom Energy at 80%.
The concentration creates a narrower market outcome than headline AI spending figures may imply. Morgan Stanley estimates that stripping Nvidia from its calculations would reduce the AI boom’s contribution to S&P 500 sales growth to 1.6%. Companies supplying power systems, networking, cooling, storage and optical equipment would still receive substantial orders, but their combined revenue contribution would be much smaller than Nvidia’s.
A large portion of the spending also bypasses the S&P 500 altogether. Morgan Stanley assigns $511 billion to an “international” category that includes Samsung and SK Hynix memory products, Taiwan-based original design manufacturers and infrastructure related to Taiwan Semiconductor Manufacturing Co. These suppliers sit at critical points in the AI hardware chain but fall outside the U.S. large-cap index.
Leases add nearly $200 billion to the 2027 spending estimate
Morgan Stanley’s $1.58 trillion estimate includes $1.39 trillion in projected cash capital expenditures and an additional $195 billion associated with leases. The bank treats those leases as AI-related outlays with debt-like economic characteristics, using signed-but-not-yet-started commitments, capacity forecasts and estimated spending per added gigawatt of data-center capacity.
The lease adjustment raises the projected 2027 spending growth rate to about 60% year over year. For 2028, Morgan Stanley estimates $1.81 trillion in AI spending, including $1.57 trillion in cash capex and $246 billion in lease-related commitments. Growth would slow to roughly 14% in 2028 under the bank’s model.
Oracle has the largest disclosed 2027 lease commitments in the analysis, at $57 billion, followed by Microsoft with $45 billion, Meta with $37 billion, Amazon with $28 billion and Alphabet with $27 billion. Morgan Stanley said the signed-but-not-started lease pipeline reached $1.118 trillion by the second quarter of 2026, after more than doubling over two quarters. Microsoft accounted for $329 billion, Oracle for $288 billion and Meta for $279 billion.
Those commitments point to an expanding financing burden that does not appear fully in standard capital-expenditure lines. Morgan Stanley’s credit team expects free cash flow at hyperscalers and emerging cloud operators to reach its weakest point in 2027 as construction, equipment purchases and lease payments converge.
Debt markets face a larger role in funding AI buildouts
According to Morgan Stanley, hyperscalers had issued about $255 billion in unsecured bonds so far in 2026, equivalent to around 30% of their estimated capital spending for the year. Amazon had issued $98 billion and Alphabet $81 billion, while all-in borrowing yields had increased by 100 to 180 basis points from the start of the year.
The cost of debt is becoming more relevant to the economics of AI infrastructure. Morgan Stanley said Oracle’s 10-year bond spread had widened by 180 basis points. In one stress test, the bank assumed AA-rated hyperscalers issued $400 billion in debt at a 7% rate, reducing interest coverage to 11.9 times from 30.7 times.
A separate scenario assumes Oracle and SpaceX raise $120 billion at 8.5%. Morgan Stanley estimated their combined interest coverage would fall to 1.8 times from 4.0 times. Such scenarios do not represent the bank’s base case, but they illustrate how quickly financing conditions could alter the economics of data-center expansion when debt, leases and equipment costs rise together.
Power competition reaches Bitcoin mining facilities
The physical constraint may prove as consequential as the financial one. Morgan Stanley estimates the United States needs 97 gigawatts of power for projected AI expansion. Even after accounting for rapid power solutions covering 34% of that requirement, the bank projects a 32-gigawatt net shortfall before 2028 and a roughly 72-gigawatt shortfall before 2029.
That shortage places AI data centers in direct competition with Bitcoin mining and other high-density computing operations for electricity, transmission access, cooling equipment and sites with existing grid connections. Publicly traded mining and data-center hosting companies with secured power capacity could have more options to shift facilities toward high-performance computing customers, particularly where long-term contracts offer steadier revenue than mining.
The transition would not be automatic. Bitcoin mining facilities require modifications to support AI workloads, including different networking, cooling, redundancy and customer-service standards. Yet the premium attached to grid-connected capacity gives mining operators a potential strategic asset as hyperscalers seek faster paths to new computing capacity.
Morgan Stanley also warned that power delays would affect more than GPU deliveries. Storage, networking, optical equipment and original design manufacturers remain exposed if data-center projects are paused or postponed. The bank cited the 2022 NAND downturn, when shipments fell about 5% while prices dropped 20% to 30%, as a reminder that hardware cycles can weaken in both volume and pricing when demand assumptions change.
For cryptocurrency markets, the immediate transmission channel is likely to be infrastructure rather than token demand. Rising borrowing costs, scarce grid capacity and competition for data-center equipment could reshape the economics of listed miners and hosting providers, especially those deciding whether to retain capacity for Bitcoin hashing or pursue AI computing contracts.
Want to act on the AI megatrend beyond Nvidia? Explore diversified opportunities on Toobit’s AI-focused market opportunities today.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
