Morgan Stanley recently released its report *Tracing the $1.6 Trillion AI Capital Flow*. For the first time, it traces the hyperscale cloud providers’ 2027 capital expenditures down to individual companies’ income statements, mapping out the complete flow of funds.

Morgan Stanley recently released its report *Tracing the $1.6 Trillion AI Capital Flow*. For the first time, it traces the hyperscale cloud providers’ 2027 capital expenditures down to individual companies’ income statements, mapping out the complete flow of funds.

Its conclusion is straightforward: the capital pouring into this AI boom is heavily concentrated in a tiny handful of firms. Underpinning this frenzy is also a large pool of “hidden debt” that the market has long overlooked.

I. Where Is the Money Going?

The report breaks down the combined $1.58 trillion in 2027 capital expenditures across nine major buyers — Google, Amazon, Microsoft, Meta, SpaceX, Oracle, CoreWeave, Nebius, plus a “lease pool” — into tiered buckets:

- Semiconductors and servers take the largest single slice at $672 billion;

- International suppliers account for $511 billion, primarily Samsung, SK Hynix, TSMC-related facilities and Taiwanese ODMs, most of which sit outside the S&P 500;

- Networking and optical communications: $215 billion;

- Data center infrastructure: $119 billion;

- Storage: $217 billion.

II. Nvidia Captures Two-Thirds

The most striking takeaway from the map is the extreme concentration of gains.

The S&P 500 stands to capture roughly $1.03 trillion in revenue from AI capital spending, and 67% of that total (around $690 billion) accrues to Nvidia alone. Put another way, 44% of the $1.58 trillion budget lands on one company’s income statement.

Excluding Nvidia, the AI Capex boom contributes only 1.6% to S&P 500 sales. The remaining 499 constituent companies split roughly $340 billion, while another $550 billion flows outside the index.

The concentration is even sharper at the single-stock level: 91.7% of Nvidia’s revenue is tied to AI infrastructure. The figures stand at 90% for Vertiv, 83% for Applied Opto, 81% for Broadcom, and even 21% for Caterpillar.

For a dozen or so giants, “AI capital expenditure” has become nearly synonymous with revenue. It supercharges results on the way up, but when Capex is revised downward, earnings follow suit — with few other business lines to act as a buffer.

III. 25% Sales Growth, 78% Earnings Growth

Morgan Stanley’s core thesis: AI capital spending will directly drive around 25% of S&P 500 sales growth, and AI-linked companies will contribute 78% of the index’s earnings growth in 2027.

In plain terms, the performance of the entire index is increasingly riding on this single capital-spending cycle.

IV. The Underappreciated “Hidden Debt”

The report’s key methodological adjustment is bringing leases into the calculation.

As of Q2 2026, the pipeline of signed but not yet activated leases has ballooned to $1.118 trillion, more than doubling in two quarters. This sum does not show up under capital expenditures on the cash flow statement, yet it is economically equivalent to debt financing.

With leases included, projected AI Capex rises from $1.39 trillion to $1.58 trillion for 2027, and from $1.57 trillion to $1.81 trillion for 2028.

Credit strategists expect 2027 to mark the peak free cash flow deficit for hyperscale cloud operators.

Emerging cloud providers are the most vulnerable: they fund nearly all their capital spending via debt issuance. Names such as CoreWeave have already seen their share prices weaken.

V. Slowing Growth and the “Power Wall”

Morgan Stanley forecasts that AI Capex growth will plunge from roughly 60% in 2027 to around 14% in 2028.

Power supply forms another hard constraint. Even accounting for fast-track power solutions, the U.S. still faces a net power shortfall of 32 GW by 2028, widening to approximately 72 GW by 2029.

On downside risks, the report is unusually blunt:

The market is currently gripped by a “fear of missing procurements” mindset. Buyers are paying premium prices today because they expect prices to rise further tomorrow. Once supply catches up with demand, this narrative will reverse.

It draws a parallel to the 2022 NAND memory cycle: shipments fell only about 5%, yet prices dropped 20%–30%. The market may be underpricing the risk of simultaneous declines in both volume and pricing.

VI. What This Means for International Students

Turning to employment, the report delivers three practical signals for international students and their families.

First, opportunities cluster within the narrow “AI + infrastructure” corridor.

High-paying roles are not mainly for people who can use AI tools, but for those who build the AI stack: semiconductors, data centers, power equipment, and optical networks — the hardest engineering tracks in STEM.

Second, the moat for AI jobs lies in irreplaceability.

With 91.7% of Nvidia’s revenue tied to AI, firms like this are deeply cyclical. They hand out raises readily in upswings, but also cut jobs quickly in downturns. When picking a career track, ask whether the role remains mission-critical when the cycle turns.

Third, hybrid hardware-software profiles are more recession-resistant.

When pure algorithm roles cool off, cross-functional talent who understand models alongside industry use cases (energy, manufacturing, healthcare) becomes scarcer.

For international students preparing for U.S. job searches, rather than chasing fleeting trends, map your position along the Capex value chain: can you land in the core (compute, infrastructure), or fill gaps in industry deployment?