The borrowing behind the AI boom

By Curtis Travis

Special to Financial Independence Hub

Every boom finds its own way to borrow, and I have a habit of reading the footnotes to see how. This one raised about US$466 billion of AI-related debt in the first nine months of 2026, against about US$101 billion a year earlier. [All dollar amounts here are US unless specified C$.]

Those figures were reported by the Financial Times, citing Morgan Stanley data. I could not find the original note or its definition, so I treat the $466 billion as a reported total. The direction is not in doubt. Every source I read says borrowing is several times larger than it was, and that issuers are paying more to get deals done.

If you hold a U.S. or global index fund, you very likely own the borrowers. That is why the split matters, and the market has already made one.

Two markets in one boom

At the October 21 close, Oracle was 55 per cent below its intraday 52-week high and CoreWeave 56 per cent below. Meta, Nvidia, Amazon and Microsoft sat within 4 to 11 per cent of theirs, and Alphabet 18 per cent. The two that fell furthest carry the credit stress. S&P cut Oracle from BBB to BBB-minus, the lowest investment-grade rating, on July 9. CoreWeave had $47.5 billion of debt principal at September 30, up from $35.6 billion three months earlier, and its 9.75 per cent notes due 2031 were reported yielding about 15.6 per cent on October 20, which put them near 81 cents on the dollar.

Off the balance sheet is not off the page

Here is the number I find most useful, and it comes straight from the filings. Each of five large companies discloses leases that are signed but have not started. The liability goes on the balance sheet when each lease begins. In each company’s latest filing, as of November 2:

Microsoft: $364.6 billion (September 30)

Meta: $355.7 billion (September 30)

Oracle: $288 billion (August 31)

Amazon: $164.9 billion (September 30)

Alphabet: $130.5 billion (September 30)

That adds up to about $1.3 trillion, though the dates and definitions differ a little between filings. Microsoft, Amazon and Alphabet filed after my first count on October 22, and together they added more than $100 billion to the total. Meta’s figure alone went from $103.8 billion at December 31 to $355.7 billion at September 30, which is 3.4 times in nine months (the latest figure also counts energy infrastructure). I would love a TFSA that did that, but this is the other side of the ledger.

Two cautions. These amounts are disclosed in the notes, so calling them hidden is wrong. And what they cover varies. Microsoft and Oracle describe theirs as primarily for data centres, Alphabet adds energy infrastructure, Meta lists data centres alongside energy and network infrastructure, and Amazon does not say what its are for. Take Oracle: its leases run 15 to 19 years, while its servers have an estimated useful life of six. That is a long mortgage on a house full of phones. The real question is whether the computing pays for the building before it ages.

Subprime or sub-par?

A report published in October by the Open Markets Institute and Americans for Financial Reform Education Fund, written by Matthew Scherer and Maya Jenkins, says speculative data centre debt is being repackaged and sold to pension funds, endowments and other institutions “in a manner eerily reminiscent of the subprime bubble.” It is a policy paper written to argue for regulation. It says it is not predicting a crash, though it argues the ingredients for a serious one are in place.

The comparison holds in one respect. Securitization lets lenders pass default risk to buyers, and those buyers include retirement money. Elsewhere it is weaker, with a catch or two. Scale: about $61 billion of securitized data centre debt, according to Barclays research cited by the Structured Finance Association in July, is roughly 4 per cent of the $1.5 trillion subprime mortgage market the Chicago Fed described in 2007 (my arithmetic, across different years and different measures).

The catch is that the $1.3 trillion of unstarted leases is close to the size of that whole market, although a lease from a profitable tenant is not a mortgage. Borrowers: most of the tenants are among the largest, most profitable companies in the world, not stretched households, though Oracle, cut to the bottom rung of investment grade, shows the weaker end. Structure: the analysis below has banks higher in the capital structure and less exposed to the first losses, which is one scholar’s model, not a measurement.

A Chicago Booth Review analysis by Stefan Hepp, published in August, models a severe repricing and puts the first losses on private credit funds and AI infrastructure credit vehicles, then on their investors, mainly pension plans and endowments. I found no pension plan named as a holder of securitized data centre debt. Pension money is in the boom, though, through loans. CPP Investments put up C$225 million, half of a C$450 million construction loan for a data centre in Cambridge, Ontario, in July 2025, and La Caisse provided C$240 million of senior debt for a Montreal data centre in March.

Four things I’m watching

Conveniently, they spell DEBT.

Downgrades. S&P cut Oracle to the lowest investment-grade rating in July, so a further cut would take it below.

Earnings. The leveraged builders and the cash-rich tenants have drifted apart. If that gap narrows as earnings come in, it is a good sign.

Breaches. A covenant breach or a missed payment at a leveraged builder such as CoreWeave would change the story quickly.

Tenants. Leases that convert on schedule into rent that is actually being paid are the good news. Tenants cancelling or shrinking leases before they start would be the bad.

I give no crash probability, because the evidence cannot support one.

This is information, not advice. When I wrote the original workfile on October 22, I owned shares of Amazon and Nvidia and none of the other companies named.

Curtis Travis, B.Comm, PGCV, AACI (retired), is the founder of Travis Valuation (travisvaluation.ca), where he publishes a weekly plain-English valuation letter. His book, The Modern Value Investor, is available as an e-book and in paperback on Amazon.ca.

 

The Cost of Lower Mortgage Renewal Payments

Historic row houses on Sherbrooke Street in Montreal. Photo courtesy Philipp Hienstorfer / Wikimedia Commons

By Raj Singh

Special to Financial Independence Hub

A mortgage renewal offer can look reassuring when the monthly payment fits the household budget. But the amount still owing when that term ends deserves equal attention. For Canadian homeowners considering a longer repayment schedule, a manageable payment today can mean carrying substantially more debt into the next renewal.

A NorthScope News analysis published on October 3 illustrates the trade-off using a $400,000 mortgage. At a fixed 5% rate, extending the remaining amortization from 20 years to 25 cuts the monthly payment by about $302. After five years, the borrower still owes roughly $20,516 more than under the shorter schedule.

These are calculated examples, rather than a survey of borrowers or a prediction of available mortgage rates. They show how the same balance behaves under different repayment schedules. The figures below were also checked against a month-by-month calculation.

What the smaller payment leaves behind

The distinction between term and amortization is important. The term is the period covered by the mortgage contract. Amortization is the estimated time needed to repay the loan. A five-year term can therefore sit within a much longer repayment schedule, with another renewal needed before the debt is cleared.

In the $400,000 example, a 20-year amortization produces a monthly payment of $2,628.50. Extending it to 25 years lowers that to $2,326.42. Over 60 payments, the household keeps approximately $18,125 available for other expenses.

The remaining balances tell the other half of the story. After five years, the 20-year schedule leaves about $333,514 outstanding. The 25-year schedule leaves about $354,030. The longer schedule has reduced the principal more slowly, so a larger balance reaches the next renewal.

Calling the entire $20,516 difference an extra interest charge would be misleading. About $18,125 reflects payments postponed during the term. The additional interest paid over those five years is approximately $2,392. Together, those amounts explain why the outstanding balances diverge. Continue Reading…

What 8,000 paycheque checks reveal about how Canadians think about money

Graphic courtesy canpayinsights.ca

By Travis Zhang

Special to Financial Independence Hub

Ask a Canadian what they make and you’ll get a confident number: “Eighty-five.” Ask what actually hit their account last Friday and the confidence evaporates. “Uh… two-something?”

That second number is the one that pays the rent. It is also the one that most of us have never looked at closely.

Here is the gap. An Ontario salary of $85,000, at 2026 rates, divides into a tidy $3,269 every two weeks. What actually lands after income tax, CPP and EI is $2,449. The missing $820 is not a mistake on your pay stub. It is the difference between the number we brag about and the number we live on.

I run CanPay Insights, a free take-home pay calculator for every province, checked against the CRA’s own payroll tables. Since August 7 it has quietly logged 8,183 anonymous calculations: no accounts, no IP addresses, just the shape of what people asked.

One honest caveat before the fun part. These are people who went looking for a pay calculator, not a random sample of Canadians. Think of it less as a census and more as a window into who does this arithmetic, and when.

The people doing the math are not the ones with the corner office

Fifty-eight per cent of calculations were for full-time hourly work, and another 13% for part-time hourly work. Salaried employees? Just 21%. And 44% of the incomes entered were under $50,000.

It makes sense. If you’re on salary, your paycheque is boringly identical every two weeks. If you’re paid by the hour, with a different shift schedule every week, every cheque is a small mystery. The people closest to the edge of their budget are the ones checking the math most often.

Peak pay anxiety: 3 p.m.

Forty-four per cent of calculations happened between 9 a.m. and 5 p.m., local time, with the busiest hour around 3 p.m. Translation: a lot of people are working out their pay at work. Possibly while looking at a job posting in another tab. We won’t tell.

The other end of the clock is more telling. Twenty-one per cent of calculations happened between 10 p.m. and 6 a.m. The data can’t say why, but I have a guess, and so do you: nobody runs payroll math at 2 a.m. for fun. And 71% of all checks came from a phone, which fits both the lunch-break and the lying-awake theories.

Your savings rate is probably better (and harder) than you think

Most savings rates are quoted against gross income. But you can only save from what reaches your account.

Take an Ontario earner on $60,000 who puts away $500 every paycheque. Against gross pay, that’s a respectable 21.7%. Against the $1,821 that actually arrives every two weeks, it’s 27.5%. The second number is the honest one. It tells you how much of the money you can actually touch is going to future-you, and how tight the rest of the month will feel.

The same logic cuts the other way. Every raise, bonus and side hustle is smaller than its headline once tax, CPP and EI take their share. Plan on gross numbers and you will reach your goals more slowly than your spreadsheet promises.

Four habits worth stealing

Negotiate in take-home dollars. A $5,000 raise from $55,000 to $60,000 in Ontario adds $3,479 a year to your take-home pay, about $134 a paycheque. In Quebec, the same raise adds $3,137. Still worth fighting for. Just know what you’re fighting for.

Compare provinces on what you keep.

At $60,000, you keep $47,340 in Ontario, $47,691 in Alberta and $45,329 in Quebec. Taxes are only one
line in the moving-cost spreadsheet, but they belong there.

Before you scale back, check what you’d actually lose. Thinking about going part-time or easing into retirement? In Ontario, dropping from $85,000 to $60,000 cuts gross pay by $25,000 but take-home pay by only $16,339, from $63,679 to $47,340, because the dollars you give up were the most heavily taxed ones. That holds for someone with no other income; pensions, CPP or RRIF withdrawals change the math, so run your own numbers.

Let your RRSP do some of the heavy lifting. Contributions made through payroll reduce the tax taken off each cheque, so saving costs your take-home pay less than the amount that goes into the account. How much less depends on your marginal tax rate, which is a number worth looking up before you pick a contribution.

Know your real number, and the rest of the plan gets a lot more honest. The aggregated data behind this piece, refreshed monthly, is public at canpayinsights.ca/data.

Travis Zhang is the founder of CanPay Insights (canpayinsights.ca), a
free Canadian take-home pay calculator checked against the CRA’s
payroll deduction tables.

Is the 4% annual Withdrawal Rule still safe? What 125 years of market history from 25 countries say

The 4% rule was built on the luckiest jar on the shelf.

Image AI-generated by author with ChatGPT

By Hootan Nakhost

Special to Financial Independence Hub

The 4% rule is an American rule

The 4% rule is one of the best-known rules in personal finance. It originates from a 1994 study by financial planner William Bengen. He calculated the maximum safe withdrawal rate for every retirement start year since 1926. He found that withdrawing 4% or less of the starting portfolio in the first year, then raising that amount with inflation each year, lasted 30 years in every case he tested.

The 4% rule is useful because it is simple: you just need to remember one number. Take how much you spend each year, divide by 0.04 (or multiply by 25), and that is the money you need to retire.

One country is a small sample

The 4% number is based on a single country’s history that happens to be one of the best-performing markets in the world. Since 1926, the U.S. hasn’t had a war fought on its own soil, and since 1945 it has been the world’s major power. Retirees in other rich countries were not so lucky. Japan’s Nikkei index took 34 years to get back to its 1989 peak. In the UK, stocks fell about 73% in 1973–74, and inflation reached about 24% in 1975. German savers were wiped out twice: by the 1923 hyperinflation and again by the 1948 currency reform. None of this shows up in U.S. history.

What I tested, and what I found

My simulator uses the Dimson-Marsh-Staunton dataset, which covers stocks, bonds and inflation back to 1900 for most countries; I used the 25 of its countries that count as developed markets, each only from the year it became one (Japan from 1930, Spain from 1959), following Anarkulova et al. (2023) where the data allow.

For this post I replayed the same 30-year retirement 10,000 times, stitched together from blocks of real history from 1900 to 2024: once with U.S. history only, and once with all 25. In the 25-country case, each block of history comes from a country drawn at random. This setup avoids the survivorship bias you get when you always pick the U.S. as the home country. The retiree holds either 100% stocks or a balanced 60/40 portfolio, and spends a fixed, inflation-adjusted amount. To keep it simple, there are no taxes, fees or government benefits.

Bengen asked for a rate that never failed in U.S. history. With thousands of simulated retirements, some will always fail, so I report the rate that lasted 30 years in 9 out of 10 of them.

Once each retiree’s home market can be any of the 25 countries, the rate that lasts 30 years 90% of the time falls by about a third.

What gets you back to 4%

The weak 25-country result assumed each retiree held only their home market. With 30% of the stocks at home and 70% in the rest of the world, the same 60/40 portfolio’s rate rises from 2.8% to 3.6%. With 100% stocks, it reaches about 4.1%.

In this test, bonds lowered the rate. The simulation always uses the retiree’s own country’s bonds, the kind that inflation and war destroyed, so the finding is narrow: home-country bonds didn’t add safety. Global and real-return bonds weren’t tested.

One caveat: an all-stock portfolio can lose half its value in a crash, and the simulation assumes you never sell.

Do you even need 4%?

Benefits in Canada like CPP and OAS change the question. They’re indexed to inflation and paid for life, so your savings only have to cover the gap. Consider a 65-year-old couple with C$1 million in home-country stocks and maximum CPP and full OAS, about $54,500 a year at 2026 rates. Under the 25 countries’ history, their savings can add about $27,500 a year, 2.75% of the $1 million. Together that’s about $82,000 a year of spending, two-thirds of it from benefits. Maximum CPP is uncommon, and these figures are before tax and any OAS clawback. The number that matters is how much of the spending the portfolio has to cover.

What I take from this

On U.S. history alone, about 4.2% held up. Across 25 countries, a home-only 60/40 portfolio held up at about 2.8% and a balanced global one at about 3.6%. Only an all-stock portfolio spread around the world held up at 4%, and only for a retiree who could sit through a 50% drop without selling. Continue Reading…

How to find the Money Leaks you don’t know about

Image created with Grok Imagine by Expense Canary

By Karla Castro

Special to the Financial Independence Hub

Most of us know roughly what we spend on rent, groceries and gas. The harder money to track is the money we never notice leaving. I call these money leaks. A streaming service you signed up for once and forgot. A free trial that quietly turned into a paid plan. A monthly account fee your bank charges every month. A charge from a merchant you don’t recognize.

None of these is large on its own. That is exactly why they last. A small charge that repeats every month can go unnoticed for a long time.

Here is how to find these leaks yourself, with or without an app.

Start with your statements

Pull up the last three months of statements for every chequing account, savings account and credit card you have. Every one. Look for three things:

First, recurring charges. Anything that shows up on the same day each month, or each year, is a subscription. Ask yourself whether you still use it.

Second, bank fees. Look for monthly account fees, overdraft fees, non-sufficient funds fees, foreign transaction fees and ATM fees. These often appear as short codes, so read each line.

Third, charges you don’t recognize. Some are merchants billing under a different name. Some are not. Either way, you want to know.

Check your app store subscriptions

Many subscriptions are billed through your phone, not directly by the company. On an iPhone, open Settings, tap your name, then Subscriptions. On Android, open the Google Play Store, tap your profile icon, then Payments and subscriptions. You may find apps you stopped using long ago still renewing.

Search your email

Search your inbox for words like “receipt,” “renewal,” “trial” and “subscription.” Receipts and renewal notices often arrive by email even when the charge is buried on a statement. When you start a free trial, put the end date in your calendar the same day.

Why leaks hide across banks

If you have one bank account and one credit card, a manual review is manageable. Many Canadians and Americans have more than that. A chequing account at one bank, a credit card from another, a line of credit somewhere else.

Each bank only shows you its own piece. A subscription on your travel card won’t appear in your main banking app. A fee on an old savings account you rarely log into can run for a long time. Leaks hide in the gaps between accounts.

What to do when you find one

Finding a leak only helps if you act on it. Here are the usual options:

Cancel it. If you no longer use a subscription, cancel it through the company’s website or your app store settings. Then check your next statement to confirm the charge stopped.

Ask for a refund. If you were charged for a renewal you didn’t want, contact the company and ask. Some will refund a recent charge, especially if you haven’t used the service since.

Ask your bank about fees. If you were charged an overdraft or account fee, call and ask whether it can be reversed. If fees keep coming, compare accounts. Some online banks and credit unions offer accounts with no monthly fee.

Report what you don’t recognize. If a charge is truly unfamiliar, contact your bank or card issuer right away.

Using a tool (full disclosure)

Disclosure: I built Expense Canary, so I am biased. You can do everything above by hand. I built an app because doing it across several banks every month is tedious.

Expense Canary links your accounts and cards from different banks into one view. There is no budget to set up. It flags monthly account fees, overdraft and other bank fees, recurring charges including free trials that turned paid, and unfamiliar or larger than usual charges. You can also turn on optional notifications when new charges post.

It connects read only through Plaid in the US and Canada. It never sees or stores your bank login and can never move or transfer money. It also won’t cancel anything for you. It shows you what it found, and you decide what to do. It provides information, not financial advice.

It is available on iOS and Android, with a 7-day free trial, then $4.99 USD a month (local pricing may vary). You can learn more at https://expensecanary.com.

Whatever method you choose, the habit matters more than the tool. Set a reminder to review every account once a month. The leaks you find are money you get to keep.

Karla Castro is the founder of Expense Canary, a read-only app for iOS and Android that helps people in the US and Canada spot forgotten subscriptions, hidden bank fees, and unusual charges across their linked banks. She is based in Vancouver. https://expensecanary.com