Reformed Millennials - Learn Earn and Invest

I fixed the sink…That’s the whole story, honestly. There was a drip under the kitchen sink, I took a photo of it, asked the thing what I was looking at, and then I fixed the thing. No plumber. I know that confirms a few priors about me, but I bring it up because that little win did more to change how I feel about this technology than any essay I’ve read.You can sit around complaining about AI. You can keep telling everybody that software is dead. Or you can actually participate. Learn how to use the technology. Pay attention to the companies benefiting from it. And when the stock market starts rewarding those companies, don’t be afraid to own them.The tapeRough month. The market stopped trading AI and started trading AI against the cost of money, and both can and are true at once. The 10-year has spent most of October above 5% and closed Tuesday a hair under it. Midterms are next Tuesday. Everybody is nervous and nobody can tell you exactly why, which is usually what nervous looks like.Then Anthropic filed its S-1 on Monday. Press reports have it raising around $100 billion at a valuation near $2 trillion, which would make it the biggest IPO ever.Revenue in 2025 was $4.6 billion. Revenue in the second quarter of 2026 alone was $11.6 billion. Third quarter, by the company’s own math, was about $16.8 billion. Nobody has grown like this before. Revenue per employee is over $6.6 million. Ninety percent of their code is now written by their own model.Two things I pulled out of it. First, this company spends most of its compute on research and training, not on serving you and me. The CFO has said that publicly. The part that serves customers is extremely profitable. The part that builds the next model is where the money goes. Which tells you that for the next 12 to 18 months, demand for chips and power isn’t slowing down, because you only stay in this race by spending.Second, the market is right to be twitchy about the multiple. Growth slowed from roughly 147% quarter over quarter to about 45% in Q3, and that’s a real slowdown, not just big numbers getting bigger. OpenAI shipped a model in September that matched or beat theirs and some of the revenue followed. The harness turned out to be less sticky than people hoped. Meanwhile, token prices keep falling and GPU rental prices keep rising, which is a strange thing to watch at the same time. SemiAnalysis has about 50 gigawatts of new data centre capacity coming online in 2027. When that lands, rental rates come down, and then we find out what intelligence actually costs.My read:Is that the aggregate demand across every model, open and closed, is the number that counts, and its really difficult to account for. Anthropic going public is the first time we get a clean look at any of it. I think that’s bullish for the sector short term and medium term.The other datapoint from the week was Microsoft offering a voluntary exit package to about 60,000 people, roughly 27% of the company. Highest-paid workforce on earth. That’s the efficiency showing up in the numbers, and it’s scary if you work there, and it’s exactly why we dedicated the rest of the show to Cognitive Mercantilism... say that fast 3 times.Cognitive mercantilism, explained.Venkatesh Rao published an essay on Saturday called Cognitive Mercantilism. I sent it to about 20 people. Mel had already read it… because of course she did.I want to lay it out carefully because it’s the best frame I’ve found for why the loudest AI pessimists are the people with the most degrees or the most “intelligence”.Mercantilism is what came before capitalism. From roughly 1500 to 1800, if you asked a king or a general or a landowner how an economy worked, they’d tell you something like this. Wealth is a fixed pile. It’s land, gold, silver, and the labour of the poor people who work that land. The point of policy is to grab more of the pile: more territory, a trade surplus, more bullion in the vault. My gain is your loss. Rao’s twist, is that mercantilism was never the philosophy of merchants. It belonged to the aristocrats and the military and the people collecting rents. The actual traders and the people starting businesses were the ones pushing against it the entire time. Capitalism from that push.Under the surface, mercantilism needed four beliefs to hold:* Wealth is scarce and finite.* Ordinary people can’t be trusted with freedom. They’re unruly and need managing by their superiors.* Supply is everything. Control the productive stuff, charge high prices. Demand from the poor doesn’t count because the poor have nothing to spend.* Most people are only good for farming. With 80% of the population in the fields, that looked like the natural order of things.What broke it was demand. Rao leans on the historian Joyce Appleby here. Capitalism wasn’t inevitable. It showed up when people discovered that wealth could be created and not only captured, and that ordinary people could also be customers of production. When a farmer could suddenly want tea and a clock and a window with glass in it, he worked more to get them. Days worked per year went from around 250 to 300. Nobody forced these ppl to work more... Desire did the persuading and the effort followed that desire. Markets got pulled into existence by the wanting.Now swap the word “wealth” for “intelligence” and you have our modern ai conundrum.Cognitive mercantilism treats intelligence as a fixed resource that lives in credentialed brains, to be hoarded, kept expensive, and handed out by approved custodians. And the same four beliefs apply here. * Intelligence is scarce. * Regular people can’t be trusted with powerful tools. * Supply of that resource is everything, so we MUST “control the frontier” because its too dangerous. * And most people are only good for whatever they’re doing now.And who holds this cognitive worldview today? It sure as heck isn’t the plumbers. It’s the people who spent their lives going to school, hoarding intellegence so they could sell it at a higher price. These folks tend to have their entire identity and income wrapped up in owning intelligence and selling it. Doctors, lawyers, consultants, advisors, academics, the media, and yes, the fronteir labs themselves. So it follows that the technocrats building this think it’s the end of the world. They’re the aristocrats in this story. A technology that makes intelligence cheap and abundant hits them exactly where mercantilism hit the landowners in 1780. They’re losing control of their scarce thing.An electrician who started using it to bid jobs and write invoices and run his own estimating does not see it that way. He sees a tool that let him do something he couldn’t do last year. Those two people are not going to agree about AI, and it’s not because one of them is ignorant.Mel put it in economics terms. Think of a production possibility frontier, the curve of everything an economy can make with the resources it has. Most policy is arguing about where on the curve to sit. This moves the curve. That’s a different kind of thing.She also made a point I’ve been chewing on since our chat... Opposition to AI is turning into a luxury belief, the same way being anti-energy did. If you already have enough wealth and security that you don’t need more, you may struggle to understand why anyone would want a tool that helps them get more. And every society needs an elite of some kind. The thing capitalism does well, and the thing this technology does well, is lower the barrier to becoming one.A few caveats we should point out.The Engels’ Pause was real. Between 1800 and 1840 output per worker in Britain rose something like 46% while real wages rose about 12%. The gap went to the people who owned the machines. The handloom weavers didn’t get retrained, they got wiped out. Wages only caught up when the technology spread into enough industries to create new work and when the institutions changed: unions got legalized, the vote got extended, the Corn Laws got repealed. If people … This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit reformedmillennials.substack.com

Delta raised fares about 20% from a year ago and nobody stopped flying… Premium revenue was up 18% last quarter. Ed Bastian’s explanation was blunt: the top 40% of American households are “nearly $40 trillion wealthier than they were just a few years ago.”However, that same economy eating airfare increases produced 372 large US company bankruptcy filings in the first half of 2026. That’s the most since 2010, and one more than the same stretch last year. Two years running.We have Zombie companies and we have breakthrough technology businesses. What we need to start asking ourselves as investors: Does this mortgage, this rental, this stock still work if money costs 8%?The tapeWe recorded Tuesday morning. The S&P 500 closed that day at 7,805, a fraction under the record it set on October 6th. The Nasdaq sat at 27,204. Small caps kept sliding.Bonds are the story. The US 10-year touched 5.36% last week, the highest since 2002, and closed Tuesday around 5.2%. The Fed’s September minutes said another hike “would likely be appropriate by year end.” Brent is above $105, with a hurricane shutting in most of the Gulf of Mexico’s output.Earnings season opened with JPMorgan missing for the first time in 17 quarters. Goldman’s revenue was up 23%. Korea’s market fell 5.5% on Tuesday after Chinese memory maker CXMT announced a denser DRAM design, and SK Hynix dropped 10% in a day.Under an index near its highs, by one count 134 S&P 500 stocks are down 40% or more from their peaks. The average stock is not having the year the index is having.We grew up on cheap moneyMel and I came into adulthood as rates were falling. If you bought in Canada in 2016, 2017 or 2018 you borrowed at 3% or 4%, and every renewal looked cheaper than the last. Then COVID took rates to zero. A whole generation, mine included, learned that 2% to 4% was normal.What we need to come to terms with is that period was the exception, not the rule. The better mental model now is 2003 to 2014, not 2016 to 2024. Central banks aren’t trying to juice growth anymore. They’re trying to hold down prices in an economy that, in the US at least, is running extremely hot.Somebody who financed a rental at 2.75% five years ago is renewing at something closer to 4.75% or 5.25%. That cost goes to the tenant. The tenant has less left at the end of the month. Multiply that across the country and you have the next ten years of Canadian politics: rent, groceries, and whether anyone under 40 can buy anything.Mel made the uncomfortable point on the episode. Prices only cool when demand falls, and demand falls when people can’t afford things. In her words, what we’re trying to accomplish “doesn’t always come in the packaging of what we want.”There’s exactly one painless fix, and it’s abundance. More homes, more power, more of whatever is scarce. Every other fix is a decision about who gets hurt...Into the end of the year, The Bank of Canada will probably sit still. September jobs came in at minus 68,300, unemployment hit 6.5%, and the loonie touched an 18-month low. The odds of a hike this month fell to about one in four. That’s the good news, and it isn’t very good. Ottawa’s ban on foreign buyers of homes also expires January 1st, and the government has to decide what to do with it.The great extensionIn the zero-rate years, a lot of companies borrowed cheap and pushed their maturities as far out as they could. They didn’t fix anything. They bought time. Those maturities stack up from 2028 and peak around 2031.You can see where it’s cracking first. Fitch has private credit defaults at a record 6.1%. In office, 71% of the distressed CMBS balance is tied to a refinancing that failed or is about to, per CRED iQ. The buildings pay their bills and still can’t roll the loan. Telus cut its dividend 55% in July to pay down debt.Meanwhile high-yield spreads sit around 3.1%, close to the tightest they’ve ever been. Credit markets are relaxed. Bankruptcy court is busy anyway.A client moved over to us recently from an advisor holding 120 individual stocks. No one person can follow 120 businesses. Going name by name, I’d guess 30% to 40% of them were dividend payers that only work because money cost 2% or 3%. When those companies refinance, nobody is coming to rescue them. They’ll be sold for parts.So I’m running everything we own through six tests:* EBIT covers interest less than 1.5 times, or less than once on a three-year basis.* Net debt above 4x EBITDA (6x for REITs and utilities).* More than 30% of debt floating, or more than 25% maturing by the end of 2028.* Negative free cash flow after dividends, two years running.* Marginal cost of debt above return on invested capital.* Share count growing more than 5% a year, or convertibles as the main funding source.The fifth test catches the business that only makes sense when money is cheap.There are always exceptions… Debt on its own isn’t the problem. Carvana was the poster child for a broken balance sheet in 2022. Today it’s at 1.0 times net debt to EBITDA, and in August it refinanced its 9% notes into a cheaper term loan. Debt plus improving unit economics works. Debt plus a model that needs 3% money doesn’t.AI at 8%…Parts of the AI buildout run the same borrow-to-build playbook at a much higher cost of capital. The best counterargument I read came from Gavin Baker at Atreides. Anthropic’s S-1 shows roughly $413 billion of non-cancellable commitments to its four biggest suppliers. That headline number is designed to scare people instead of inform them… Spread over the life of the contracts, it works out to about $31 billion a year of cloud spending, which a company at Anthropic’s revenue can pay for itself, plus around $161 billion at Broadcom that Broadcom and private credit will fund. Most of those contracts were also signed below today’s price for compute.His bigger point I agree with completely. Expensive money is good for this buildout. We’re short power, short chips and now short capital, and that’s what keeps it from turning into a bubble. Run this cycle with zero rates, TSMC tripling its capex and turbines on every shelf, and you get an overbuild. An overbuild is the worst outcome for investors and for everyone else. Slower and longer is better.Who’s spendingSo where does the money that isn’t stuck refinancing go?I’ve been working backwards from one assumption. If robotics is solved within a decade, manufactured goods get a lot cheaper. If that’s right, the smart way to spend discretionary dollars between now and then is on what stays scarce. My list keeps landing on travel, the Four Seasons, golf courses and real life. For Canadians the banks are the uncomfortable part. Financials are about a third of the TSX. Stripe says more than 86% of the Forbes AI 50 use it. The fastest-growing companies in the world aren’t building on big-bank infrastructure, and our banks need to decide what they’re going to offer those companies. Some will figure it out. Some won’t, and the market will price the difference.Three provinces over three weeksQuebec voted October 5th and elected a Parti Québécois minority on about 28% of the vote, the lowest share ever for a governing party in that province. Paul St-Pierre Plamondon has promised a sovereignty referendum in his first term, though not before Trump leaves office.Alberta votes Monday. There are ten questions. Question ten, which sits first on the ballot, asks whether to start the legal process toward a binding vote on separation. Questions one to five deal with immigration, social programs and voter ID. Six to nine are about Ottawa: who appoints superior court judges, abolishing the Senate, opting out of federal programs with funding, and provincial law winning in provincial areas.Mel’s advice is worth repeating. Vote for what you actually want. Don’t vote strategically. If you want to send Ottawa a message without starting a separation process, questions six through nine are where that message goes.The surprise of the episode was how differently our … This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit reformedmillennials.substack.com

Alberta filed its application for a West Coast oil pipeline on July 2nd 2026... It was designated a project of national interest this morning in Fort Mac.On Tuesday the Prime Minister stood in Vancouver and welcomed LNG Canada’s Phase 2 at Kitimat, a $33 billion final investment decision that doubles the facility and ranks as the second-largest single private investment in Canadian history. So inside one week this country said yes to a million barrels a day of oil heading south-west and a second helping of natural gas heading north-west.Meanwhile the most downloaded app in America is a robot from Meta that comes with its own computer.I called our Q3 client letter A Bridge to the Future, and after this week I think that should be changed to The Pacific Link to the future… One bridge is being built out of software and it runs through Menlo Park. The other is being built out of steel and gas turbines, and it runs from Bruderheim Alberta to Vancouvers Delta Port. MARKET Update:The Fed went on September 16th, a 12-0 vote taking the funds rate to 3.75% to 4%. First hike since 2023. The 10-year hit 5% on the 14th and was through 5.1% by the 23rd, the highest in nineteen years.That has done something strange to the market. The Nasdaq touched 27,000 this morning and sits roughly half a percent off all-time highs, but about 70% of its components are below their 50-day moving average. The equal-weight S&P, where Sherwin-Williams counts as much as Apple, is in a 17.5% drawdown.When money costs more, the companies that can raise it at the lowest spread get a tailwind and everyone else gets re-rated. The index looks fine because eight companies are fine. Underneath it, the broader economy is being repriced.I think we have a rough three or four weeks into the midterms and then a good finish to end the year… Uncertainty parks capital on the sidelines, and midterms manufacture uncertainty. Earnings have run well ahead of expectations this year while multiples contracted, which means the market got cheaper in 2026 even as it went up. The seven to nine months after a midterm is historically the best stretch of the four-year cycle.A bridge made of softwareUntil this quarter, an AI was something you talked to. Muse is something that does things for you.The trick is simpler than it sounds. Meta gives every user their own computer, a real one running in Meta’s data centre, and the agent logs into it and works on your behalf with your calendar, your email, and your habits.An agent is just an AI with access to a computer. Before this year the only people who had that were engineers willing to wire it up themselves. Meta handed one to everybody on September 8th, and Microsoft did roughly the same thing for office workers on the 25th with a persistent Copilot agent called Autopilot that gets its own identity, memory and workspace inside your company’s Microsoft 365.An Anecdote from my life:I was sitting in the sauna with Adrianna and typed “book me a hair appointment tomorrow at Parlor Barbra.” It read my email, already knew who cuts my hair, checked my calendar and booked 3:40. One prompt. A few days earlier I asked it to sort out airport parking for a Seattle trip and it had already pulled my WestJet booking, the ninety-minute buffer and my plate number. Two prompts, fifteen seconds, replacing fifteen minutes of form filling.Then my son’s soccer registration failed on the league app. Muse flagged the email, I re-registered him, the schedule went into the calendar, and it sent the whole thing to Adrianna with a reminder to pack his socks and shoes. That last part matters more than the haircut. Mel made the point on the episode that the mental load of running a household has always been carried disproportionately by women, and the thing I noticed is that it didn’t get redistributed to me. It got lifted off both of us.Ben Thompson’s version is better than mine. He asked Muse to organize the recipes he’d saved on Instagram, and five minutes later he had a small custom app that categorized them. Nobody built that app. Nobody will ever sell it. It exists because one guy wanted it and the marginal cost of wanting is now about five minutes. His conclusion is that apps stop being destinations and become “mere suppliers,” and that the scarce thing moves from discovery to inspiration, which makes whoever owns the agent “the ultimate gatekeeper of not just user demand, but desire.”Google and Meta built the two best advertising businesses in history by solving discovery. Getting things done is about to be abundant. What stays scarce is the desire part of getting things done.Why Meta wins this even though its model isn’t the best. This is Clayton Christensen, straight out of the textbook. Once a product is good enough, the competition stops being about performance and starts being about convenience, and then distribution. Meta bet that a good-enough model inside a great product beats a great model inside a chat window. The labs with better models still do not have a consumer surface with this kind of lock-in, and there is almost nobody on earth with more distribution than Meta. It also finally explains the glasses, which were never a product so much as a delivery mechanism...The adoption numbers are worth being precise about, because I got them slightly wrong on the recording. Muse crossed five million downloads in twenty-two days per Sensor Tower. ChatGPT took fifty-six days to the same mark, Grok took 103, Claude took 492. It has held the top of the US App Store free iPhone chart for twelve straight days. And Meta has been spending its own house ad inventory on it, up to half of its daily impressions in the two weeks ending September 27th, which it has essentially never done for anything but Threads. Companies don’t put their own megaphone behind something until the retention data says it’s working.The tell that it’s real is that Amazon said no. On September 20th Amazon blocked Muse from shopping its site. Amazon’s advertising business did more than $68 billion last year, roughly double the profit of its retail operation, and an agent walking straight to checkout doesn’t look at a single sponsored placement. The next day Shopify opened checkout to Muse through Shop Pay. Walmart and Target have been backing an open agent-checkout standard with Google and Shopify.The incumbent with the most to protect pulls up the drawbridge. The challengers hold the door. That’s the same pattern as the second-place carriers signing with the iPhone, and it tells you more about where this is going than any demo.The distribution story is enormous.The bridge made of steelPacific Link is 1,250 kilometres from Bruderheim to the coast near Delta, almost entirely along the Trans Mountain corridor, one million barrels a day, $35.2 to $43.7 billion. It is the first project listed under Schedule 1 of the Building Canada Act.Ownership lands at 45% Ottawa through Trans Mountain, 45% Alberta through the Alberta Petroleum Marketing Commission, and 10% Pembina through construction with an option on up to another 10% at commercial operation. Indigenous communities are being offered a minimum 10%, financed through the Indigenous Loan Guarantee Programs.This is not a construction approval. Yet. The Major Projects Office and the Canada Energy Regulator have until September 1, 2027 to finalize conditions covering ownership, benefits, Indigenous rights, environmental protection and local hiring.The federal government is putting its own name and its own money on the timeline, which is a different thing from a press release.Speed is the resource. Alberta is not out-competing Virginia or Texas on access to capital. Capital is the one thing the world has plenty of. What Alberta can compete on is the willingness to say yes quickly, and ninety-one days from application to designation is not a normal number on this continent. The metric to guard from here is months from application to shovel. If that number holds, everything else… This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit reformedmillennials.substack.com

Welcome back everyone. Summer’s over, whether we agreed to it or not, and you can feel everyone with impact returning to their desks. It shows up in the inbox, on the timeline, and most of all in the bond market.Market Update📈📉Warsh gave his Jackson Hole keynote and dropped the pretence: “we have work to do.” Core PCE is running 3.3 percent against a 2 percent target, headline 3.7 on oil, and the market has genuinely priced a September rate hike into play. That’s a stunning shift from March. You can see it in the ten-year, in gold, and especially in small caps, which carry the highest financing costs and get hurt first when the tail of the curve rises.IMPORTANT MARKET STRUCTURE CHANGE: While the Fed talks tough, the Treasury has been quietly running a yield-cap operation, buying down the long end of the curve, because mortgage rates ripping higher into a midterm cycle is a political problem. The Fed is actively tightening while the Treasury tries to loosen, two arms of the same government pulling against each other, and the Fed chair doing the tightening is the one Trump picked himself. And the Treasury is losing. Yields sit above where they were when the operation started.Why? Go back to the Ben Thompson question we covered two episodes ago: are we going to have enough money? For eighteen months the bottlenecks were GPUs, optics, power and the market has worked diligently toward solving each one. The new bottleneck is capital. For the first time in roughly twenty years, the hyperscalers are issuing serious amounts of debt, and per Cembalest’s new Eye on the Market piece, their 2026 issuance is about $320 billion, which works out to roughly 70 percent of all new long-duration Treasury borrowing this year. When Washington goes to raise money, it now competes with Microsoft and Google for the same buyers, and frankly, on the financial metrics, some of those companies are maybe more creditworthy than the government. That competition is what’s pushing the long end up, and the long end is what sets your mortgage rates.The AI infrastructure layer, the biggest contributor to S&P earnings growth for two years, has ground sideways to down, with many names off nearly 40 percent from the June peaks. Beating earnings and guiding up wasn’t enough:Earnings season is over and essentially all of the bottleneck tech companies beat and guided higher. It didn’t matter. When the cost of capital rises, your multiple compresses, and beat-and-raise stops being enough. The market wants perfection.Is the spending justified? The revenue says yes. Google Cloud’s backlog jumped $52 billion in a single quarter to over half a trillion dollars. AWS grew 37 percent, Azure 43, Google Cloud 82. Earnings revisions are at the high end of anything we’ve seen in thirty years. Two asterisks belong on all of it, though. First, capex and R&D from these companies went from $460 billion in 2024 to nearly $900 billion over the last twelve months, so some of the market’s earnings strength is the spending itself. Second, an unusually large share of S&P earnings right now comes from unrealized gains on stakes in Anthropic, OpenAI, and SpaceX, positions marked up from raises done at $200 to $500 billion valuations. Anthropic is expected to go public in about eight weeks at something like $2 trillion. Those are real gains on paper, and paper is doing some of the market’s lifting.My honest read: none of this is fatal. The proverbial trains are full and running on time, and demand for the railway has never been more justified. Expect a choppy six to eight weeks while the bond market finds and appropriate yield level, and remember that these rate limiters are healthy. Bubbles without governors are the ones that end in 2008. Bubbles with them tend to leave behind the railways and the fiber, the things that define the next century. I think AI ends up delivering something like the railway plus the internet combined. But from here, you have to know exactly what you own. It won’t be as easy as it was twelve months ago.Canada note: the Bank of Canada is expected to hold at 2.25 for a seventh straight meeting. Our headline CPI of 2.8 is gasoline-driven and our core is bang on target at 2 percent, so there’s no case to hike. The loonie sits at 1.39, banks don’t see a Canadian hike until 2027, and anywhere between 1.30 and 1.40 the economy is happy. Meanwhile Alberta’s Q1 fiscal update landed with oil averaging roughly $90 against a $60.50 budget assumption. As Albertans, we are on the right side of nearly every force squeezing everyone else.Why this isn’t the dot-com rerunThe question every client asks: is this a bubble yet? Here’s the framework that finally made it click for me.The internet had one dimension of exponential growth: user count. Everyone paid a flat fifty or a hundred bucks a month, and using it ten times more didn’t cost ten times as much. One curve, and once penetration hit about 50 percent, it saturated. That was March 2000.AI has two dimensions which is different than the Internet/fiber era.Penetration crossed 50 percent a few months ago, right on schedule. But usage per person is just leaving the station. Median enterprise AI spend is about $12 per employee per month today. The plausible destination is around 10 percent of a white-collar salary, call it $1,200 a month. And it’s already visible: a16z said this week that their portfolio founders are running at multiples of the median. The internet stopped at fifty bucks a month. This doesn’t stop at twelve.And the unit economics of the infrastructure have changed underneath us. A gigawatt costs $40 to $60 billion to build, roughly 70 percent of it flowing to Nvidia. But look at SpaceX’s filings renting compute to the labs and the software companies: in some cases a gigawatt is generating on the order of $100 billion in revenue. Put out $40 to make $100 in a year and tell me that’s malinvestment. The usual caveat stands, all exponential growth is an illusion before the ceiling arrives. But this ceiling has an extra dimension holding it up.The backlash, and the window of opportunityThe next leg of risk isn’t chips or money. It’s politics. The blanket hatred of data centers on social media is real and spreading: New York has its moratorium, Virginia is wobbling, and now Texas, the build-anything state, just imposed a temporary pause. Mel’s framing deserves to be quoted at length. This is an arms race, the first our generation has seen since the space race, and the question isn’t whether the world is going there, it’s who arrives first and captures the upside. Skepticism about water and power is legitimate, and we have institutions and regulators built to answer exactly those questions. But ask who benefits when North America blocks its own growth. It’s China, period, end of story. And a society that can be talked into fighting itself over obvious opportunities is a society that can be divided and conquered without anyone setting foot on a battlefield.Her challenge to listeners was the best moment of the episode: whenever a TikTok take hardens your politics, ask yourself whether TikTok would permit that same discourse about the Chinese government inside China. You know the answer. So why is it happening here?For Alberta, every moratorium elsewhere makes our window wider and more valuable than it was six months ago. Bring-your-own-generation, anchor tenants paying their own freight, ratepayer bills going down. We have the model. The only question is whether we defend it before someone imports the fear.CUSMA/USMCAThe escalation, quickly: on August 25 Canada suspended negotiations and announced a dollar-for-dollar response to US 50 percent tariffs on $27.6 billion of Canadian exports, effective September 8, with groceries largely carved out, alongside a $7.5 billion support package. Trevor Tombe’s estimate in The Hub is worth reading carefully: the aggregate impact is small, about 2.5 percent on the average export rate, but the aggregate hides… This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit reformedmillennials.substack.com

Welcome back everyone.Market Update📈📉Start with one number that has nothing to do with AI. The 30-year Treasury hit 5.31% this week. That’s the most expensive long-term money since 2007. I was in high school the last time capital cost this much. Canada’s long bond is at levels last seen in 2010, and it’s the same story in Germany, France, and Italy. Everywhere you look in the developed world, the cost of borrowing for thirty years is breaking out, and it’s happening at the exact moment the biggest companies on earth need to borrow more than any group of companies has ever borrowed.You might assume these companies are so rich it doesn’t matter. That’s no longer true. On top of the capex you can see on their income statements, nine large tech companies are carrying roughly $3 trillion in commitments that don’t show up on their balance sheets, most of it tied to AI and much of it longer-dated than people assume. Notably, on the calls this quarter, the CEOs themselves stopped comparing this moment to the dot-com era. They compared it to the railroads.Ben Thompson was on Invest Like the Best this week and said a thing I haven’t been able to stop thinking about since. Everyone argues about whether we’ll have enough compute or enough electricity. He thinks the harder question is simpler: are we going to have enough money? For three years the market has treated capital as infinite and chips, power, and memory as the bottlenecks. Each of those has been a bottleneck. But watch the walk down the capital ladder. First the buildout ran on big tech’s cash flows. Then it moved to debt and off-balance-sheet structures. Now Nvidia is partnering with Apollo, Brookfield, Blackstone, and KKR on a $500 billion vehicle to reach into pension funds and insurance floats, the deepest and most conservative pools of capital on the planet. To get that money, the private equity shops have to underwrite data centers like real assets: how long do they run, what cash do they throw off, what’s the honest depreciation schedule. And Nvidia effectively had to underwrite its own CUDA moat to give them the confidence to do it.Which brings us back to the railroads… In the 1870s the world literally ran out of money mid-buildout. Nearly every railroad went bankrupt, many more than once. BUT, the trains kept running. The railroads opened the West and powered the economy for a century and a half. The technology worked and the investors got wiped, at the same time. The great returns on the rails didn’t really show up until Buffett started buying them a hundred years later. The returns on this buildout need to show up a lot sooner than that, or we’re in for a world of hurt.So is there real revenue? Yes, and the numbers are hard to process. Anthropic is running at roughly $65 billion in annualized revenue and is expected to go public within about ten weeks. They believe they exit this year at $100 billion or better, and there are credible investors who think $400 to $500 billion by the end of 2027 is on the table. If the data centers are the railroads, Anthropic and OpenAI are the biggest trains on the track, and the freight is real.One more distinction: a data center is really two businesses. * Training, done on the newest chips, is a loss leader. You can’t sell a training run. * Inference, serving your prompts, is believed to carry margins in the 90s for Anthropic. The training bill is why the leading labs are barely profitable despite those inference margins. And that explains the strategy underneath the open-source push. The hyperscalers don’t want to spend trillions building infrastructure only to serve two customers who keep all the margin. So Nvidia, Meta, Google, and Amazon are backing open-weight models, Kimi K3 and Muse Spark among them, that now run three to six months behind the frontier. Their goal is to compete the frontier labs’ margins down and pull that profit back toward the infrastructure layer. That’s a healthier market, and it’s why the picks-and-shovels names have rallied so hard these past few weeks.Canada, America’s farm team:Mel’s segment this week deserves your full attention. Her opening argument was blunt: “Canada does not have the leverage it thinks it does, and the longer we spend beating our chest, the longer we delay fixing our own house.” Trevor Tombe’s piece in The Hub made the rounds as a good-news jobs story, complete with a victory lap tweet from the Prime Minister. However, if you read the actual article the employment gain was within the survey’s margin of error, and Tombe’s real conclusion was that Canada is returning to normal growth “along a permanently lower path.” Our peer countries grew about 1.7% while we contracted. Mexico, heavily exposed to the same US trade risk, outgrew us. So did the EU. So did post-Brexit Britain. With commodity prices this friendly, that’s not bad luck; that’s just horrific policy choices coming to bear.Canada has become the farm team for the American economy. There are now 517 US-based companies with Canadian founders that have together raised $414 billion. Fifty-six of them are worth more than a billion dollars each. That’s roughly the size of our entire federal budget, built by people we educated and then lost. And it’s not an accident. The Americans built a conveyor belt for our talent: the Thiel Fellowship and 1517 Fund for high schoolers, Neo Scholars and Emergent Ventures and YC’s apply-and-defer for students, the Founders Fund residency and SPC for new grads, then Y Combinator and the a16z speedrun. Each rung lifts a kid from Waterloo or the U of A up and out, funds them, houses them, and celebrates them. We spend twenty-five years raising a great entrepreneur, and then they pay taxes in Delaware, California or Texas.Part of this is culture… We have no better country on earth in which to be average, and I mean that as a genuine compliment. But we treat the top one percent as the enemy, and they are listening. Mel put it well: “nobody’s asking you to bring out the violins for millionaires, but a good start would be to get off their necks.”The Fraser Institute’s new study says the average Canadian family now spends 42% of its income on taxes, more than housing, food, and clothing combined, and the tax bill has grown faster since 1961 than any of them. If we want the infrastructure, the services, and the rising standard of living, we need the people who pay for it to want to stay. Elbows up is a privilege. It means you’re already post-economic. The people still trying to build don’t have that luxury.The new status symbolsWe closed on a lighter thread that I think is a real investing signal. The perfectly curated life is dead as a status marker. When everyone’s kitchen is Instagram-ready and anyone can buy the bag, its no longer signal... That has become a base case. We should watch what’s replacing it: refinished shoes instead of new ones, thrifted furniture, and increasingly, not posting at all. Mel’s list of the new status goods was better than mine: muscle you visibly worked for, clear eyes and real rest, the ability to sit still with a book for an hour, a family, and the quiet flex of not needing a job. (agree to disagree on this one)My wife and I have been watching Cape Fear on Apple, the Scorsese production with Javier Bardem and Amy Adams. Best show I’ve seen in a year. But the moment that stuck with me was Adrianna saying she loved how Amy Adams looks because she looks real. Six months ago we watched a Nicole Kidman movie and her comment was that nothing in her face moves. When the tastemakers start prizing real over done, that eventually shows up in earnings, the same way new entrants showed up in Lululemon’s and Nike’s numbers. Hollywood is still how America exports its culture, and culture is where consumer trends are born. Keep an eye on the aesthetics-procedure complex. The volumes may quietly tell the story first.New episodes every two weeks. Keep learning, keep earning, and keep investing.Podcast & YouTub… This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit reformedmillennials.substack.com