Morgan Stanley's $1.58 Trillion AI Spending Forecast Names Politics, Power and Funding as Risks as September AI Bond Sales Fell to $23 Billion

Since Morgan Stanley's Oct. 15 report put 2027 AI computing spending at $1.58 trillion, more details from the 65-page note have surfaced. They cover how the plan could fail, who it depends on, and how it gets paid for.

The downside section: people, power and politics

Morgan Stanley public policy strategist Ariana Salvatore frames the main constraints on the buildout as "people, power, and politics." The bank still expects the spending to happen. It attaches a growing list of conditions.

Salvatore writes that political pushback "has become the wedge issue in the midterm elections," which the bank sees as a catalyst for AI sentiment. The midterms are about two weeks away.

The bank does not expect Washington to halt the buildout. Salvatore cites the national-security framing and a likely presidential veto of restrictive legislation. She sees the midterms laying "the groundwork for potential legislation toward that end in 2029."

The levers that bind now are local. Interconnection, large-load rate design, siting, permitting and tax abatements are decided at the state level. That is "why governors and utility commissions matter more than Senate races," in Salvatore's words.

Morgan Stanley also ranks outcomes by sentiment. Non-incumbent candidates in key races tend to propose tougher restrictions. The bank therefore sees sentiment as most constructive under a Republican sweep, broadly neutral under divided government and least constructive under a Democratic sweep.

ZeroHedge, which obtained the section, notes that a JPMorgan equity basket called the "Democrat Factor" has risen about 20%. It reads that as Wall Street pricing a Democratic sweep. That is ZeroHedge's interpretation. Morgan Stanley's report does not make it.

September bond sales dropped

The funding side showed strain last month. Global AI-related bond issuance fell to roughly $23 billion in September, the second-lowest monthly total this year, according to Morgan Stanley's data. The U.S. investment-grade segment posted a rare zero-issuance month.

Through September, cumulative AI-related bond issuance reached $466 billion. That is more than double last year's full-year total of $216 billion. June was the peak at $113 billion.

Morgan Stanley calls the September slowdown "a pause, not a retreat." It gives three overlapping causes: financing front-loaded earlier in the year, data center construction bottlenecks with regulatory and political headwinds, and rapidly rising interest rates forcing some project financing terms to be renegotiated.

The bank says the freeze did not reflect weaker AI fundamentals or a shortage of capital. It expects high-rated hyperscale cloud providers to return to the U.S. investment-grade market in the fourth quarter, though not at the first-half pace. Its biggest year-end test for credit markets is the macro and rate backdrop, not an oversupply of AI bonds.

Higher interest rates matter more when the borrowing is this large. Hyperscalers have already sold about $255 billion of unsecured bonds this year, roughly 30% of estimated 2026 spending. The bank expects 2027 to bring the largest cash-flow deficit of its forecast period and more borrowing.

In a theoretical borrowing exercise, Morgan Stanley estimates combined interest coverage for Amazon, Alphabet, Meta and Microsoft, all rated AA or higher, would drop from about 31 times to 12 times. For Oracle and SpaceX, it would fall from 4 times to 1.8 times.

Leases and the spending that never hits a cash flow statement

About $195 billion of the 2027 figure is data center leases. The report calls this "synthetic capex" because it never appears as capital spending on a cash flow statement. Hyperscalers have disclosed more than $1.1 trillion in lease commitments that have not yet started.

The bank's credit strategists say the lease obligations are economically equivalent to debt financing. The report also says higher component prices explain a meaningful share of recent capex increases.

Nvidia's exposure and its two biggest buyers

Morgan Stanley counts 91.7% of Nvidia's 2027 revenue as exposed to AI infrastructure demand, the highest on its list. Vertiv is at 90% and Broadcom at 81%. The same report calls Nvidia the best risk/reward in the sector.

About $1 trillion of the 2027 spending lands as S&P 500 revenue, the bank estimates, and nearly $700 billion of that goes to Nvidia. Roughly two-thirds of the beneficiary revenue goes to one company. The other 499 index members split about $339 billion.

Hyperscaler and neocloud customers supply 5% of S&P 500 revenue in 2027 but drive about 25% of its revenue growth. AI-related companies contribute 78% of S&P 500 profit growth by 2027.

Nvidia's customer base is narrower still. In the bank's model, SpaceX and Amazon Web Services together account for about 54% of Nvidia's 2027 hyperscaler revenue and 72% of its growth.

Morgan Stanley's case for the stock rests on valuation and returns. Nvidia trades at 15 times fiscal 2028 earnings, and the bank says multiple expansion "is also not required for the stock to work as numbers continue to go higher." It expects the spending to earn 25% to 50% on invested capital.

The bank argues Wall Street has underestimated compute cycles by 38% on average across four past cycles: PCs, the internet, mobile and cloud. It calls its SpaceX and Amazon estimates conservative, "a low bar for remaining hyperscalers." Its model has Nvidia's share of other hyperscalers' spending slipping from 18% in 2026 to 16% in 2027, while Nvidia expects to gain share.

Morgan Stanley's own investment committee sounded a more cautious note in March. It said the bull market "rests, more than anything, on the durability of AI capital expenditures."

What to watch

Two tests come first. One is whether the fourth-quarter investment-grade issuance that Morgan Stanley expects materializes after September's freeze, with rates still rising. The other is the midterm result, which will shape how much local resistance data center projects face.