The AI Boom Is Turning Into a Financing Story, and Bond Buyers, Tax Collectors and Insurers Are Doing the Math
Oracle’s credit sits one notch above junk, states are pulling back data-center tax breaks and insurers are pricing tornado risk on campuses worth $20 billion or more. The build-out’s next judges aren’t engineers — they’re the people who lend, tax and insure.
By Sean Anees Saifi · Capital Wealth · Published Thursday, September 10, 2026 · Source: The Wall Street Journal, September 8, 9 and 10, 2026 editions
Key Points
S&P rates Oracle (ORCL) debt one notch above junk, and insuring that debt against default costs about four times as much as for Microsoft (MSFT).
Oracle spent about $56 billion on capital projects in the fiscal year ended in May and plans about $70 billion this year; its stock trails the S&P 500 by about 51 percentage points over 12 months.
Capital spending by Oracle and the four other biggest cloud builders is about $800 billion this year and is projected at more than $1 trillion a year for the next four years.
Ohio’s data-center sales-tax break cost more than $1.5 billion last year, over 10 times the estimate, and more than 10 states have pulled back similar breaks, with New Jersey canceling its remaining $250 million.
Swiss Re projects data-center insurance premiums of $20 billion to $30 billion a year by 2030, while 40% of U.S. data-center capacity sits in tornado-prone areas and 62% of Michigan voters oppose data-center development.
$800B
Capex by Oracle and four peers this year
4x
Oracle default-insurance cost vs. Microsoft
$1.5B+
Ohio data-center tax break cost last year
40%
U.S. data-center capacity in tornado zones
Data-center campuses now run into the tens of billions of dollars apiece, and their financing increasingly depends on lenders, state tax policy and insurers willing to take a slice of the risk.
In one line: The AI build-out now needs lenders, legislators and underwriters to keep saying yes, and many portfolios own that bet more times than their owners realize.
Larry Page reportedly told Google staff he’s “willing to go bankrupt rather than lose this race.” Alphabet (GOOGL) can afford to talk like that. The race is the AI data-center build-out. Capital spending by Oracle (ORCL) and the four other biggest cloud builders runs about $800 billion this year, with projections above $1 trillion a year for the next four. At that scale, the interesting question isn’t who builds fastest. It’s who pays, and on what terms.
The weakest link has a credit rating
Oracle is where that question gets sharp. S&P has cut its debt to one notch above junk since June, and the cost of insuring its bonds against default hit a multiyear high this summer and still runs about four times the price for Microsoft (MSFT). Oracle spent roughly $56 billion on capital projects last fiscal year and plans about $70 billion this one, and its stock trails the S&P 500 by some 51 percentage points over 12 months. It reported after Thursday’s close, but one quarter won’t settle whether a borrower a notch from junk can keep pace with giants that can absorb poor returns.
The tax collectors and insurers want a say
Bond investors aren’t the only ones sharpening pencils. Ohio’s sales-tax break for data centers cost more than $1.5 billion last year, over 10 times the state’s estimate, and more than 10 states have pulled back similar breaks; New Jersey canceled its remaining $250 million. In Michigan, 62% of voters oppose data-center development. Insurers smell opportunity — Swiss Re projects $20 billion to $30 billion a year in global data-center premiums by 2030 — but single campuses carry insured values of $20 billion to $30 billion, and 40% of U.S. capacity sits in tornado-prone areas. Each of those is a price, and each tends to push the cost of building higher.
Here’s the portfolio question: how many ways do you already own this build-out? Plenty of portfolios own it several times over — through the S&P 500 index fund, the growth fund, the big tech names and the power companies wiring the sites. Vertiv (VRT), a data-center equipment maker, fell 9.6% Wednesday, and the paper gave no reason. That’s the catch with concentration: when one theme runs through four corners of a portfolio, a rough day rarely stays in one corner.
So count. Pull the statements, list every fund and stock whose story depends on the build-out continuing at this pace, and add up the percentage. If the total surprises you, that’s the useful part. The boom may well keep going — but it’s now being priced by lenders, legislators and underwriters, and none of them get paid to be optimistic. The check takes fifteen minutes, a statement and a calculator.
What It Means For Your Portfolio
Watch — the AI build-out is now a credit story
The AI build-out is now priced by lenders, state legislatures and insurers, not just by stock buyers — and a portfolio that owns the theme through index funds, growth funds and power names owns that financing risk several times over.
General planning principles, not advice for anyone in particular: count how many ways a portfolio owns one theme — index funds, growth funds, individual tech stocks and power or electrical names often overlap. Concentration rarely announces itself on a single statement. Adding up the combined exposure, then deciding whether it’s intentional, is the review worth doing before a financing scare does it for you.
In the book, Oracle (ORCL) and Vertiv (VRT) are held and on watch — Oracle for credit one notch above junk, Vertiv after Wednesday’s 9.6% drop. GE Vernova (GEV), Quanta Services (PWR), Nvidia (NVDA), Microsoft (MSFT), Broadcom (AVGO) and Alphabet (GOOGL) are held. Nothing changes before Friday’s inflation report, and one Oracle quarter won’t settle the financing question.
