The Ledger

The supermarket can leave. Its ban on a rival can stay.

2026-09-13T16:40:53.000Z

A supermarket can leave a property and still help keep another grocer from taking its place. The shelves can disappear while a restriction attached to the land survives. For shoppers waiting for a cheaper alternative, the obstacle may be a document they never see.

That mechanism is part of the grocery competition fight described in the Competition Bureau’s September 3 progress report. The report also records progress: major chains have taken steps to give up property controls. The useful question is which barriers are being removed—and what that actually allows a rival to do.

Two ways to keep a competitor off the map

A restrictive covenant can attach a restriction to land, limiting what a later owner may do with it. In grocery retail, that can stop the site being used by a competing food business even after ownership changes.

An exclusivity clause works through a lease. A supermarket tenant can restrict the landlord’s ability to rent other space to a competitor, or limit which products another tenant sells.

Those are different tools, with different scopes. The Bureau’s enforcement guidance says land restrictions can be long-lasting and affect future owners. It also recognizes that a limited exclusivity clause can sometimes support investment: an anchor tenant might need protection to justify opening in a new shopping centre. Not every restriction meets the legal test for enforcement.

The argument is over how much protection is justified, across how much land, for which products and for how long.

A real barrier, followed by a carefully worded win

In January 2025, the Bureau said Empire had agreed to remove a property control in Crowsnest Pass, Alberta. According to that announcement, the restriction had been imposed in 2017 and protected Empire’s IGA from competition. At the time, the Bureau described it as the community’s only grocery store.

The announcement said a rival could move forward with plans for a second store. It did not establish that the second store had opened or that prices had fallen. This article has not independently verified either outcome.

That distinction matters. Removing a legal barrier gives a rival a chance to enter. A business still needs premises, financing, staff, suppliers and enough customers to make the store viable.

Empire’s current promises change the story

Empire’s published commitments say it will not enforce restrictive covenants on properties, including those previously sold, or register new ones. It also says it will stop enforcing specified categories of exclusivity, including restrictions on specialty food retailers such as butchers and bakeries, and will limit the geography, products and duration covered by future grocery lease exclusivity.

That is meaningful movement. It would be inaccurate to describe the company’s current position as an unchanged commitment to enforce every old restriction. It would also be inaccurate to say every form of exclusivity has disappeared: the company’s commitments distinguish between categories.

Empire argues that property controls can support investment and development, particularly in underserved communities. Its position deserves to be read alongside the Bureau’s concern that restrictions can prevent competitors from entering.

The Bureau expanded its investigation in June 2026, obtaining court orders for documents, information and testimony. That announcement explicitly reported no conclusion of wrongdoing. The September follow-up says the Bureau continues investigating property controls and monitoring changes by grocers.

For shoppers, a store closing is visible. A clause that limits its next use is much harder to see. For a prospective rival, permission to compete on paper and a workable site are both essential. Removing restrictions is progress; a new store with competitive prices is the outcome people can actually shop at.

The next grocery price war might begin with a change to a land title.

AI giants promise to pay their own power bills. Show Canadians the receipts.

2026-09-13T16:28:02.000Z

A new data centre needs more than chips. It needs power—and someone to pay for the grid around it. Ottawa says that someone should not be Canadian households or existing businesses. Amazon, Google, Microsoft and other tech companies have signed on. The promise is worth having. But a signatory list cannot tell you who pays for a specific substation—or who carries the cost if a project uses less electricity than expected.

A national promise meets a local power bill

On September 3, Ottawa released five Responsible Data Centre Development Principles. Amazon Web Services, Anthropic, Bell, Google, Meta, Microsoft and OpenAI are among the signatories. One principle says project-driven electricity costs should not shift to households and existing businesses.

The detailed framework says developers should pay the costs of connecting and serving their projects, including directly attributable generation, transmission, substations and grid upgrades, in proportion to their scale and impact. It also calls for independently verifiable local-impact information and tracking of material commitments, while allowing confidential commercial, security and privacy information to remain protected.

The crucial jurisdictional sentence sits in Ottawa’s own release: data-centre decisions are fundamentally local, and the national framework complements provincial, territorial, municipal and Indigenous processes. The principles set an expectation. Local approvals decide how that expectation meets an actual grid and an actual rate structure.

Buying electricity is only the first receipt

Three questions are easy to blur together. First: who pays for the electricity a facility consumes? Second: who pays for the new wires, substations, generation or storage needed to serve it? Third: what happens to those costs if a project arrives late, runs below forecast or is cancelled? The third is an analytical question for contracts and regulators, not an allegation that a company has already stranded costs.

Consider a hypothetical substation built only because one large facility is approved. Paying the monthly electricity bill answers the first question. Paying for that substation answers part of the second. A public record explaining who carries the remaining cost if the facility never reaches its forecast demand would answer the third.

The scale makes these questions worth asking. In its 2026 reference scenario, Ontario’s Independent Electricity System Operator expects data centres—not all of them necessarily AI facilities—to represent 8.6% of provincial electricity demand in 2050, about 60% more than its previous forecast. That is a long-range scenario, not current demand or a measured household bill increase. The IESO also models substantial uncertainty. It says most capacity needs can be met with existing and recently procured resources until the mid-2030s, although energy needs emerge earlier.

Where the promise gets tested

British Columbia offers a concrete example of the decisions below the national headline. A January 30 provincial release says prospective AI and data-centre projects must compete for clean-electricity access. B.C. planned allocations of up to 400 megawatts over the first two years, with projects assessed on price and economic, community, data-sovereignty and environmental benefits. Well-advanced projects with specified agreements or deposits continue through the existing interconnection process; traditional industries are outside this competition.

That selection process does not itself show whether other electricity customers will bear any project costs. It shows where project-specific conditions can be examined. Google says it supports paying for all power it uses and infrastructure needs directly driven by its growth. Meta says its Sturgeon County, Alberta project will fund required generation and grid infrastructure with no local ratepayer impact. Those are attributed company commitments. The next step is evidence that the approval, contract and later performance match the promise.

There is a fair upside. More domestic compute can support Canadian services, research and businesses; supplier-funded generation or grid additions can strengthen capacity; host communities may gain jobs, procurement and tax revenue. The national principles give communities a common standard for demanding those benefits.

The public test should be simple: for every major project, show the expected load, the connection and serving costs, who pays each major upgrade, who bears cancellation or underuse risk, and how regulators concluded existing customers are protected. The companies have signed the promise. Now put the cost breakdown where Canadians can inspect it.

Can I put money back into my TFSA in the same year I withdraw it?

2026-09-13T13:06:00.000Z

Yes, if you are a resident of Canada, but only to the extent that you still have otherwise unused TFSA contribution room. Taking money out does not give you that room back immediately. The amount withdrawn is added to your available room on January 1 of the next calendar year. If your room is already zero, a September 2026 withdrawal does not let you replace that money in 2026. A same-year deposit would create an excess contribution. The CRA taxes an excess at 1% per month for as long as it remains in the account. Its guidance is direct: check that you already have enough room before replacing a withdrawal.

Two cases on one calendar

Assume each person is an adult resident of Canada, withdraws CAD 5,000 in September 2026, and makes no other relevant transaction. The withdrawal itself has no effect on 2026 room. The final column is a counterfactual showing what would happen if the person deposited the full CAD 5,000 that year; it is not a recommended transaction.

Starting point before withdrawal	Room available for the rest of 2026	Counterfactual: full CAD 5,000 deposited in 2026
Otherwise unused room: CAD 0	CAD 0	CAD 5,000 excess
Otherwise unused room: CAD 2,000	CAD 2,000	CAD 3,000 excess

The first case is the cleanest example. With no room left, withdrawing CAD 5,000 in September leaves the person's 2026 room at zero. Depositing that CAD 5,000 back during 2026 would create a CAD 5,000 excess.

The second case shows why the answer can sometimes be “partly.” The person may contribute up to CAD 2,000 during 2026 because that room already existed. Depositing the full CAD 5,000 would exceed the available room by CAD 3,000.

If the person leaves the withdrawn money outside every TFSA for the rest of 2026, the CAD 5,000 withdrawal is added to the contribution-room calculation on January 1, 2027. It is added alongside the new year's annual limit and any unused room carried forward. That does not create room during 2026, and this example makes no assumption about the 2027 annual limit. Calculate the person's actual available room before making any new contribution.

If you have already contributed too much, the CRA says to withdraw the excess as soon as possible and file a TFSA Return.

The annual limit is only one part of your room

The annual TFSA dollar limit for 2026 is CAD 7,000, but that is not necessarily what you can contribute now. The CRA calculates available room from the current year's limit, unused room carried forward, withdrawals made in the previous year, and contributions already made this year. The limit applies across all of your TFSAs, rather than separately to each account. The CRA sets out the formula in its guide to calculating TFSA contribution room.

Your CRA account is not a live ledger. TFSA issuers must report the previous calendar year's transactions by the end of February of the following year, and the CRA says account information is updated once a year in the spring. A recent contribution can reduce your room immediately without appearing online yet. Use your own complete records from every institution and compare them with the transactions the CRA has processed.

Moving a TFSA is a different transaction

If your goal is to move money from one TFSA or institution to another, ask the receiving financial institution to arrange a direct transfer. The CRA says a direct transfer does not affect contribution room, although an institution may charge a transfer fee. Its TFSA transfer guidance explains the distinction.

Withdrawing the money yourself and depositing it into another TFSA is not a direct transfer. The new deposit counts as a contribution and needs available room. The withdrawal is added back only in the next calendar year.

Frequently asked questions

Can I replace part of a withdrawal in the same year?

Yes, if you already have enough unused contribution room. In the CAD 2,000 case above, up to CAD 2,000 can be contributed in 2026; the withdrawal does not expand that amount during 2026.

Does opening another TFSA give me more room?

No. Your available contribution room is shared across all of your TFSAs. Contributions to each account count against the same total.

Is the room shown in my CRA account enough to make the decision?

No. It may not include recent transactions. Reconcile the CRA information with your own records from every TFSA before contributing.

Does lower inflation mean my grocery bill should fall?

2026-09-13T13:06:00.000Z

Lower headline inflation does not tell you whether groceries became cheaper. Headline CPI combines food with shelter, transportation and other categories, so food can move differently. Even if food inflation itself slows, a positive food inflation rate means the measured food basket still became more expensive over the comparison period. To tell whether grocery prices fell, you need the change for the relevant food items or food basket, using the same quantities over the same period.

The confusion comes from treating inflation like a price tag. The price level is what the basket costs; inflation is the rate at which that cost changes. The Bank of Canada explains that lower positive inflation means the price level rises more slowly. A slowing positive rate is disinflation. A falling price level requires a negative rate of change.

One fixed-basket calculation

Imagine a weekly list with the same quantities and quality each time: the same size carton of milk, loaf of bread, dozen eggs and bag of rice, plus fixed weights of chicken and produce. Suppose this basket costs CAD 100 at the starting point. The rates below are hypothetical and use consecutive one-year comparison periods; they are not current Canadian grocery data.

Point in time	Calculation	Basket price
Starting point	—	CAD 100.00
After hypothetical 5% inflation	CAD 100 × 1.05	CAD 105.00
After hypothetical 2% inflation	CAD 105 × 1.02	CAD 107.10

The inflation rate slowed from 5% to 2%, but the basket became CAD 2.10 dearer in the second year. It is 7.1% above its starting price: (107.10 ÷ 100 − 1) × 100 = 7.1%.

For comparison, a hypothetical 2% price decline after the first year would produce CAD 105 × 0.98 = CAD 102.90. The basket would become cheaper than in the prior year, yet remain above its original CAD 100 price.

Why your receipt can differ from headline CPI

Statistics Canada describes CPI as the change in the cost of a fixed basket of goods and services. The national basket represents average Canadian household spending and covers eight broad components. Those components are weighted by their share of consumer spending.

Headline CPI therefore summarizes many prices; it does not say that food moved by the headline percentage. Food prices can rise faster than headline CPI, rise more slowly, stay flat or fall while the overall index rises.

Your household also buys a different mix from the representative national basket. The stores, brands, quantities and products you choose affect your bill. Statistics Canada accounts for quantity and quality when comparing equivalent products: a smaller package at the same sticker price represents an effective increase, even though the amount printed on the shelf label did not change.

A practical personal check is to compare a short, consistent grocery list over time. Use the same product sizes and a clearly defined period. That will not replace CPI, but it will show whether your own repeat purchases became dearer or cheaper.

FAQ

Can monthly grocery prices fall while year-over-year grocery prices still rise?

Yes. The comparison periods have different starting points. In a hypothetical example, a basket costs CAD 100 a year ago, CAD 106 last month and CAD 105.50 this month. It fell about 0.5% from last month but remains 5.5% higher than a year ago.

Can some grocery items get cheaper while food inflation remains positive?

Yes. An index combines many products. Individual items can fall while increases elsewhere leave the total food basket above its earlier price.

Is lower inflation the same as a lower cost of living?

No. Positive inflation still raises the measured price level. Statistics Canada also distinguishes CPI from a full cost-of-living index: CPI prices a fixed basket, while a cost-of-living measure asks what is needed to maintain a given standard of living.

Stocks sit at records while the Fed debates a hike — the strangest bull market in years

2026-08-17T09:00:00.000Z

The S&P 500 ended Thursday at 7,798.99, a record close, up 0.65% on the day, with the Nasdaq up 0.81%, in a session Reuters reported was driven by memory-chip makers — Sandisk up 13.7%, Micron up 4.2% — and the big platforms. The trigger was July's producer-price report, which came in flat on the month and slowed to 4.7% year-on-year from June's 5.5%, a day after consumer prices rose just 0.1% in July, leaving headline CPI at 3.4% and core CPI at 2.5% — its lowest since March 2021.

That data did something unusual to the market's Fed arithmetic. Futures pricing tracked by CME's FedWatch had put the odds of a September rate hike above 70% earlier this summer; after the week's inflation and jobs data those odds collapsed to roughly 30%, per coverage in American Banker and Reuters. The rally, in other words, is not about cheap money coming — markets still price a better-than-90% chance of a higher policy rate by year-end — it is about tightening arriving more slowly than feared.

Underneath the macro relief sits an earnings season that has broken records of its own. With over 90% of the S&P 500 reported, LSEG data cited by Reuters puts profit growth at 32.7% year-on-year excluding mark-to-market gains at Alphabet and Amazon; Bessemer Trust's tally has aggregate growth near 50% with 87% of companies beating estimates and ten of eleven sectors growing. Notably, the market has climbed 22% over twelve months while its forward price-to-earnings multiple fell from 22.4 to 20.2 — the gains have been paid for by profits, not by stretching valuations.

The Ledger's read: a record set against hike risk is sturdier than one set against cut hopes, because it doesn't depend on rescue. The vulnerability is specific instead — index-level results now lean heavily on AI-linked earnings continuing to compound, and on oil, which has eased but answers to the unresolved situation around Iran and the Strait of Hormuz, staying quiet. Watch the week's FOMC minutes and Walmart's results for the first tests.

A Fed at war with itself heads for Jackson Hole

2026-08-17T08:30:00.000Z

Kevin Warsh took over the Federal Reserve in May with the policy rate at 3.50–3.75% and inflation above target for a fifth year, aggravated early in 2026 by the U.S.–Israeli war with Iran and its effect on energy prices. His first meetings have exposed a committee genuinely divided: the July decision to hold was carried over three dissents in favour of an immediate hike, including Cleveland Fed President Beth Hammack, and Reuters reports governors Christopher Waller and Lisa Cook have both said they would support increases unless inflation cools soon.

"The question is, how quickly do we need to deliver on that 2% objective," Hammack said last week, in remarks reported by Reuters — the hawks' case in a sentence. The doves' case is that the shocks doing the damage are the kind that fade: Richmond's Thomas Barkin noted that tariffs, oil and the AI investment boom account for much of the acceleration and "should pass." July's data leant their way — core CPI at 2.5%, producer prices flat on the month — and market odds of a September hike fell from a coin-flip to roughly a third.

Warsh himself has refused to spell out a reaction function, an omission American Banker reports has kept policy uncertainty unusually high, while President Trump publicly blames the chair's "hostile" colleagues for blocking the cuts he wants. Into that vacuum come two events: Wednesday's minutes of the July meeting, which will show how close the committee actually came to moving, and the Jackson Hole symposium on August 27–29, a chair's traditional stage for declaring a framework.

The Ledger's read: the substantive fight is not about September — it is about whether five years above target has started to move the public's expectations, which the University of Michigan survey puts at 4.3% for the year ahead. If the committee concludes it has, the glide-path argument dies and the hike case writes itself, whatever this month's prints say. Read the minutes for the word "expectations," not for the vote.

The American consumer just blinked

2026-08-17T08:00:00.000Z

July's retail sales report, tracked by the Census Bureau at $763.6 billion, showed spending down 0.6% from June — and still down 0.2% after stripping out autos and petrol. The same morning, the University of Michigan's sentiment survey fell 7.6% on the month and 12.4% on the year, with the damage concentrated in expectations for business conditions rather than households' own finances. Only 8% of respondents expect their incomes to outpace inflation over the coming year, and the survey's year-ahead inflation expectation ticked up to 4.3%.

The softness has company. Initial jobless claims rose to 209,000 in the week to August 7, ending a run of sub-200,000 readings, though the four-week average of 199,000 still describes a historically tight labour market and unemployment sits at 4.1%. Reuters reports wages have declined in real terms over the past six months — which is the quiet arithmetic behind the sentiment numbers: prices at 3.4% against tepid pay growth is a pay cut, whatever the payroll count says.

Markets, in the current configuration, treated all this as good news, because a cooling consumer argues against the rate hike part of the Fed was pushing for. "It's a case of be careful what you wish for," Northlight Asset Management's Chris Zaccarelli told American Banker — consumer spending traces to nearly 70% of GDP, and a slowdown big enough to keep the Fed on hold is also big enough, if it runs, to reach corporate profits.

The Ledger's read: one soft month after a World Cup-flattered quarter is a wobble, not a turn — Q2 spending was boosted by the tournament and tax relief, so some July payback was mechanical. The number that matters more is the expectations one: consumers acting on 4.3% expected inflation demand raises, resist prices and pull forward purchases, and that behaviour is self-fulfilling. Walmart's results this week are the single best read on whether the blink becomes a flinch.

River’s $1.1bn financing and Lovable’s $13.3bn valuation: what August’s AI funding numbers show

2026-08-17T07:30:00.000Z

Correction and revision note — September 13, 2026: The original standfirst said Lovable’s valuation had tripled. Lovable’s announcements show that it rose from $6.6 billion on December 18, 2025, to $13.3 billion on August 12, 2026—about 2.02 times, or approximately double. Wispr’s credited report postdates this article’s original timestamp; its announcement appears here as part of this revision. This revision uses sources now publicly available and does not represent a human factual review of the original article.

Three AI companies announced large financings in August 2026. The numbers landed close together, but they describe different transactions and should not be treated as interchangeable measures of company performance.

River AI announced on August 11 that it had raised $1.1 billion across what it called its Series Seed and Series A. The company said General Catalyst and AMP PBC led the financing, with strategic investment from Nvidia and AMD Ventures and participation from Y Combinator and Temasek. River had introduced itself publicly on June 10 and shared an API for training and adapting open models. Those public records do not establish the company’s founding date.

Lovable announced a $400 million Series C on August 12 at a $13.3 billion valuation. Its previous announcement, dated December 18, 2025, said it had raised a $330 million Series B at a $6.6 billion valuation. Dividing 13.3 by 6.6 gives about 2.02, so the later valuation was approximately double the earlier one. The $400 million is the amount Lovable said it raised in the Series C; $13.3 billion is the valuation attached to that round.

Wispr’s own August 17 announcement said the voice-software company had raised a $280 million Series B at a $2 billion valuation, led by Menlo Ventures. Wispr said the round brought its total capital raised to $361 million.

The distinction between these figures matters. An amount raised is the capital committed in a financing; a valuation is the value assigned to the company in that transaction. Neither number alone establishes revenue, profitability, how the money will be spent or what investors will earn. The supported conclusion is narrower: River, Lovable and Wispr each announced substantial financing in a six-day span, while the disclosed amounts and valuations measured different parts of those deals.

The AI trade passed its earnings exam — the bill arrives later

2026-08-17T07:00:00.000Z

The AI complex spent the week proving it can still clear the bar it keeps raising. Super Micro Computer reported quarterly earnings 84% above analyst estimates and jumped 19%, AP reported, with guidance above expectations too. The next session, Sandisk surged 13.7% and Micron 4.2% after strong results from Lenovo read across to the memory market, helping push the S&P 500 to its record. AI computing supplier CoreWeave also featured in the advance.

The demand side of those results is a spending number without modern precedent: Reuters-cited estimates put Big Tech's AI investment above $700 billion this year, up from roughly $400 billion in 2025. Earnings season suggested the market has, for now, stopped punishing the spenders — Reuters reports investors have come to read aggressive capital expenditure as a response to visible demand rather than hubris, helped by strong forecasts from Microsoft and Amazon and by the financial strength of the companies writing the cheques.

The strength is also broader than the megacaps. Bessemer Trust's season tally shows ten of eleven S&P sectors growing earnings, with the breadth extending well beyond the AI leaders — an important detail, because it means the market's profit base is less concentrated than its market capitalisation.

The Ledger's read: the exam that matters has moved. The question is no longer whether AI-linked companies can grow into their guidance this quarter — they keep doing so — but whether $700 billion a year of capacity finds paying use before the hardware inside it ages. That is a utilisation question, not an earnings question, and it will be answered in compute prices and depreciation schedules over quarters. Our deep dive this week takes that question apart properly.

Why 'bad news is good news' is back — and how to tell when it stops working

2026-08-17T06:30:00.000Z

Last week produced a market pattern that reads as nonsense to anyone outside finance: shops sold less, hiring data disappointed, and equities rose. The mechanism is rate expectations. With part of the Federal Reserve arguing for a hike, every piece of soft economic data lowers the odds of tighter policy — so bad news for workers arrives as good news for asset prices. Northlight's Chris Zaccarelli described the week to American Banker as "three positive reports in a row," two of which were positive only in that inverted sense.

The regime has a well-defined failure mode, and it is worth stating before it happens rather than after. Bad news supports equities only while it stays small enough not to threaten earnings. The moment soft demand data starts cutting revenue forecasts, the same reports flip sign: they stop promising a gentler Fed and start promising weaker profits. With consumer spending traceable to nearly 70% of GDP, the distance between "cooling enough to hold rates" and "cooling into the earnings line" is not large.

Three indicators mark the boundary. Watch real yields rather than the policy rate — the ten-year Treasury near 4.69% against easing inflation means the bond market is doing some tightening on its own, and Reuters notes it is real yields that pressure the AI-heavy long-duration end of the equity market. Watch guidance rather than results: this season's rare negative outlooks are the tell to count. And watch the shops directly — which is why a Walmart print, this week, carries more regime information than any Fed speech.

The Ledger's read: enjoy the inversion while it lasts, but treat it as a phase, not a law. The trades that work in "bad news is good news" — long duration, long the index — are precisely the ones that break when the regime flips, and regimes flip mid-quarter, not at scheduled meetings.

What 4.3% expected inflation means for your cash

2026-08-17T06:00:00.000Z

Buried in last week's University of Michigan survey is the number that matters most to household finance: consumers expect 4.3% inflation over the year ahead, up from 4.2%, and only 8% of respondents expect their incomes to keep pace. Set that against the rate environment — a Fed policy rate of 3.50–3.75%, and measured inflation at 3.4% — and the household arithmetic is uncomfortable: cash in a typical account earning below the policy rate loses purchasing power at roughly the gap between its yield and whichever inflation number you believe.

Which number you believe is the crux. Measured CPI has been easing — core is at 2.5%, its best since 2021 — while expectations have been rising. Both cannot describe the coming year. If the official trajectory holds, cash and short-duration savings are close to whole in real terms for the first time in years. If households' expectations prove right — and five years above target is why they distrust the official path, a concern several Fed policymakers share, per Reuters — then unhedged cash keeps leaking value at 2026's pace.

The practical responses are unglamorous. Money parked long-term at a rate below the policy rate is an unforced error in either scenario — the spread between lazy deposit rates and best available short-term yields is pure cost. Fixed long-term borrowing gets cheaper in real terms if expectations are right, so the panic refinance has a case against it. And wage negotiation is where expectations become reality: the 92% who don't expect to keep pace mostly won't ask.

The Ledger's read: the expectations number is a better guide to household behaviour than to actual inflation — surveys overshoot in both directions. But personal finance is played against your own costs, not the CPI basket, and the only universally correct move in a 3–4% world is refusing to hold cash at 1%.

Microsoft’s capex forecast lost $15bn. Read the accounting note.

2026-08-17T05:30:00.000Z

Corrected September 13, 2026: the earlier version described an aggregate capital-spending estimate as AI-only spending and confused nominal and real Treasury yields. We removed the unsupported aggregate and historical comparisons, corrected the yield figures and rebuilt the analysis around identified sources. The original publication date is retained.

Microsoft’s calendar-2026 capital-spending forecast went from roughly US$190 billion in its April earnings call to about US$175 billion in its fiscal fourth-quarter call. The US$15 billion difference invites an obvious conclusion: less building, less money at risk. Microsoft’s explanation does not support that shortcut.

The company said in its fiscal fourth-quarter call that it was extending the estimated useful lives of data centres and office buildings from 15 to 25 years, effective in fiscal 2027. That changes how more future data-centre leases will be classified. Finance leases count in its reported capital expenditures; operating leases do not. Outside this effect, Microsoft said its calendar-2026 investment expectations were unchanged. These were building-life estimates, not a claim that GPUs would last 25 years.

The number and the obligation

The useful question is what changed in the underlying commitment. A smaller reported capex figure can come from cancelling capacity, paying less for equipment, changing delivery dates or changing which accounting category captures a lease. Those possibilities tell different stories about future costs and demand.

A hypothetical comparison makes the distinction plain. Two businesses might use similar buildings while one buys and the other rents. Comparing their upfront capital-spending figures alone would miss the rent the second business still has to pay. The businesses are not automatically equally risky; the contracts, duration and exit rights matter. The point is that one headline number cannot answer all those questions.

That is also why adding up company forecasts demands care. Readers need the same period, a defined group of companies and consistent treatment of leases. An aggregate for capital expenditure should not silently become an estimate for AI alone. Microsoft described its fourth-quarter hardware investment as serving both AI and non-AI infrastructure. The shared buildings and equipment make a clean label harder, not less necessary.

Strong earnings do not settle the investment case

There is evidence behind the market’s optimism. In its August 14 investment update, Bessemer Trust reported positive earnings growth in ten of eleven S&P 500 sectors. It said the index had returned 22% over the preceding twelve months while its forward price-to-earnings multiple fell from 22.4 to 20.2. Those are Bessemer’s reported figures for that period, not today’s market readings.

That supports a narrower argument: rising earnings helped support share returns. It does not establish the profitability of every new data centre. An existing business can earn substantial cash while a new project still depends on uncertain future utilisation, prices and costs. The investment case needs both sides of that ledger.

Financing comparisons need the same precision. On August 14, the U.S. Treasury’s ten-year nominal par yield was 4.68%; its ten-year real par yield was 2.41%. The real measure comes from inflation-protected securities. They are different benchmarks, not interchangeable descriptions of the same rate.

What would change the argument

For investors assessing this build-out, three comparisons are more useful than cheering or fearing a single spending total: capacity actually used against capacity installed; cash receipts against construction and contractual payments; and an asset’s expected earning life against its depreciation assumptions.

These are questions to investigate, not evidence that Microsoft has hidden a loss or that the AI boom must fail. A demand shortfall would matter differently from an accounting reclassification. So would a contract that leaves a customer paying for unused capacity. The documents have to establish which situation applies.

Our analysis is that the spending race is also a contest over measurement: what counts as investment, when the cost appears and whose obligation survives if expectations change. For the electricity side of that question, read our analysis of AI companies’ power-cost promises. For another distinction a large headline can blur, read our comparison of AI financing and company valuation.

The next time a forecast drops by billions, start with the footnote. It may change what the number means more than what gets built.

The Ledger

The supermarket can leave. Its ban on a rival can stay.

Two ways to keep a competitor off the map

A real barrier, followed by a carefully worded win

Empire’s current promises change the story

Related reading

AI giants promise to pay their own power bills. Show Canadians the receipts.

A national promise meets a local power bill

Buying electricity is only the first receipt

Where the promise gets tested

Related reading

Can I put money back into my TFSA in the same year I withdraw it?

Two cases on one calendar

The annual limit is only one part of your room

Moving a TFSA is a different transaction

Frequently asked questions

Can I replace part of a withdrawal in the same year?

Does opening another TFSA give me more room?

Is the room shown in my CRA account enough to make the decision?

Related reading

Does lower inflation mean my grocery bill should fall?

One fixed-basket calculation

Why your receipt can differ from headline CPI

FAQ

Can monthly grocery prices fall while year-over-year grocery prices still rise?

Can some grocery items get cheaper while food inflation remains positive?

Is lower inflation the same as a lower cost of living?

Related reading

Stocks sit at records while the Fed debates a hike — the strangest bull market in years

A Fed at war with itself heads for Jackson Hole

The American consumer just blinked

River’s $1.1bn financing and Lovable’s $13.3bn valuation: what August’s AI funding numbers show

The AI trade passed its earnings exam — the bill arrives later

Why 'bad news is good news' is back — and how to tell when it stops working

What 4.3% expected inflation means for your cash

Microsoft’s capex forecast lost $15bn. Read the accounting note.

The number and the obligation

Strong earnings do not settle the investment case

What would change the argument