1. How much AI capex does Morgan Stanley expect in 2027?
Morgan Stanley projects that hyperscalers and so-called neocloud providers will invest about $1.58 trillion in AI computing in 2027, and its model assigns roughly $690 billion of the spending that reaches S&P 500 companies to Nvidia. The projection, reported on October 18, places the chipmaker at the center of what may be the largest corporate building cycle on record.
The $1.58 trillion total combines about $1.388 trillion of cash capital expenditure with roughly $195 billion of data center leases. Capital expenditure, usually shortened to capex, is money a company spends on long-lived assets such as buildings, servers and chips. Hyperscalers are the very large cloud operators that run enormous data centers, and neoclouds are newer providers that rent out AI computing capacity.
The bank expects the total to rise further, to about $1.81 trillion in 2028, before growth in the spending slows after 2027. The forecast therefore describes a boom that keeps climbing in dollar terms while the rate of increase begins to ease.
Forecasts of this kind are sensitive to what is counted. A narrow definition that includes only the cash spent by a few of the biggest cloud companies yields a much smaller number than one that adds neocloud providers and leased capacity. That is why published estimates of AI infrastructure spending vary widely between banks and between months, and why the scope of the $1.58 trillion figure matters as much as its size.
The projection is useful mainly as a map of where money is expected to go. It does not predict the performance of any individual stock, and it does not establish that the spending will prove profitable for those who make it, a question that depends on how much revenue AI products eventually generate.
2. Why does Nvidia sit at the top of the spending map?
Morgan Stanley estimates that around 65% of the 2027 total, about $1.03 trillion, will arrive as revenue at companies in the S&P 500. Within that share, Nvidia is the largest single recipient, with about two-thirds, or close to $690 billion.
The reason is straightforward. Nvidia designs the graphics processors that train and run most large AI models, and data center operators buy them in vast quantities. When a hyperscaler builds a new facility, a large part of the equipment budget goes to those accelerators and to the systems built around them, so spending on AI infrastructure flows disproportionately to one supplier.
The figure should be read as a modeled estimate rather than reported revenue. It comes from Morgan Stanley's allocation of expected spending and is not a number Nvidia has confirmed. Actual sales will depend on how much customers ultimately spend, what share of the budget goes to Nvidia hardware rather than to other suppliers, and whether chip supply can keep pace with orders.
It also helps to keep the arithmetic in view. Two-thirds of the roughly $1.03 trillion that is expected to reach S&P 500 companies comes to about $690 billion, which is the number associated with Nvidia. The remainder is spread across other chipmakers, server makers, networking suppliers and the owners of the buildings and power systems that house the equipment.
3. What does the spending mean for the S&P 500?
The bank puts AI-related investment at about 4.9% of projected S&P 500 revenue in 2027, which is close to 5%. The spending is also expected to account for roughly a quarter of the index's sales growth that year. Some analysts express the effect differently, estimating that direct AI investment adds about two percentage points to revenue growth, lifting it from roughly 6.6% to 8.6%.
The picture is more concentrated for profits. A separate analysis of 42 AI enablers and adopters is tied to about 78% of projected S&P 500 earnings growth in 2027. That number describes a share of profit growth, not a share of total profits, and the distinction is important: the index's overall earnings would still be much larger than the portion that depends on this theme.
Even so, the numbers suggest that a single investment cycle carries an outsized share of expected growth. When a quarter of sales growth and most of earnings growth come from the same trend, index investors are more exposed to that trend than the diversified appearance of 500 companies implies.
Capital spending flows through the economy in several steps. A cloud operator orders equipment, suppliers record that order as sales, and those suppliers in turn buy components, pay staff and invest in capacity. The effect on index-level revenue is the sum of those steps, which is why a large spending program by a handful of customers can register at the scale of the entire S&P 500.
4. Why do leases matter in the AI capex count?
Leases are the part of the forecast that is easiest to overlook. Data center leases function much like debt-financed purchases of assets, since a company commits to years of payments for capacity it will use, but they are not recorded as capex. As a result, traditional cash-flow statements can understate the real scale of the buildout.
Morgan Stanley included about $195 billion of leases in its 2027 figure. A hyperscaler that rents capacity from a developer or a neocloud still takes on a substantial financial obligation, even though the cost shows up in a different part of its financial statements than a purchase of servers would.
For readers who follow company filings, the practical lesson is that headline capex numbers alone may not capture the full commitment. Lease disclosures and borrowing trends provide a fuller view of how much an operator has promised to spend.
The accounting treatment is not a flaw in the data but a feature of how leases are reported. It does mean that observers comparing companies need to be careful to compare like with like, particularly when one operator buys its servers and another rents space from a developer.
5. How are borrowers positioned to fund it?
The bank's analysis flags credit as a point of vulnerability for some companies. Borrowers with lower credit ratings may be more exposed if capital markets tighten, because they depend more heavily on outside financing to pay for their buildouts.
Highly rated hyperscalers are in a stronger position. Their credit standing gives them access to debt on favorable terms, and they have meaningful capacity to borrow. Spending that is concentrated in the early years of a project can still strain free cash flow, which is the cash left after operating costs and investment.
This difference helps explain why the same forecast can look reassuring for the largest operators and more fragile for smaller ones. A company with strong ratings can absorb a slowdown in financing conditions far more easily than one that relies on investors continuing to lend at current prices.
The financing question will probably grow in importance as the spending grows. A buildout measured in the trillions of dollars cannot be paid for from operating cash alone, so borrowing and leasing take on a larger role, and the terms on which companies can raise money become part of the investment story.
6. What does the forecast mean for investors?
The main takeaway is concentration. Nvidia's projected share of the spending, the quarter of index sales growth tied to AI capex and the 78% share of earnings growth linked to enablers and adopters all point to a market whose near-term performance is closely tied to whether the spending materializes.
The forecast also shows why precise tracking matters. Because leases sit outside the capex line, two analysts looking at the same company can arrive at different views of how fast it is expanding. Estimates for the total can also shift as methods and assumptions change, which means the $1.58 trillion figure is best treated as a reference point and not as a fixed outcome.
Risk assets tend to respond to changes in the pace of AI investment, since many large technology holdings share the same drivers. A shift in expectations for hyperscaler spending can move the shares of chipmakers, cloud operators and the broader indexes that hold them.
For investors who want to look beyond the headline total, the most useful habit is to read lease and debt disclosures alongside capex. A company whose obligations are rising faster than its reported spending may be taking on more commitment than the usual metrics suggest, and the forecast makes clear that such obligations are a meaningful part of the overall wave.
7. What should readers watch next?
Several indicators will show whether the forecast holds. The first is capex and lease guidance from the largest hyperscalers when they report results, since any change in their plans feeds directly into the total. The second is how quickly Nvidia and its manufacturing partners can ship new platforms, because supply limits can cap sales even when demand is strong.
The third is whether Morgan Stanley and other forecasters revise their 2027 and 2028 numbers. The bank expects growth in spending to slow after 2027 even as the total reaches about $1.81 trillion in 2028, so the shape of the curve, and not only its height, will draw attention. Revisions upward or downward would indicate whether the current outlook is conservative or optimistic.
Because the projection is a model and not a report of results, each of these checkpoints offers a chance to test its assumptions against real data. Readers should expect the numbers to move, and should focus on the direction of revisions rather than on any single total.