Newsletter · · Ashutosh Agarwal

US Canada Trade War Moves From Tariffs to Procurement Lockouts - Trade War, Tariffs & Reshoring - Week of September 14, 2026

Trade War, Tariffs & Reshoring for the week of September 14, 2026. Podcast synthesis of the escalating US-Canada trade war shifting from tariffs to federal procurement lockouts, the US-China standoff over rare earths and soybeans before the September 24 summit, reshoring reality checks in pharma, chips and solar, the critical-minerals scramble, humanoid robots meeting their first real financials, and the power crunch underneath it all, covering podcasts from September 7 to 13, 2026.

Trade War, Tariffs & Reshoring

Week of September 14, 2026: US Canada Trade War Moves From Tariffs to Procurement Lockouts

A weekly read on tariffs, reshoring, factory robots and the infrastructure being built to hold it all up, drawn entirely from the past week of podcasts (September 7–13, 2026).

The week in one read

Two stories dominated the week, and they rhyme.

The first: the US-Canada trade war stopped being a tit-for-tat tariff fight and turned into something colder and more surgical. Canada finally fired its retaliation (15% to 50% duties on roughly C$27.6 billion of American goods), and Washington answered not with a bigger tariff number but by reaching into the machinery of the U.S. government: locking Canadian firms out of federal and state purchasing worth an alleged $50 billion a year. The lesson operators kept repeating: the pain is moving from the tariff line into procurement, visas, banking rules, places that are harder to see and harder to fight.

The second: everyone is now bumping into the same wall. Whether you want to re-industrialize with copper mines, humanoid robots, chip fabs or AI data centers, you eventually run out of two things: critical raw materials that China controls, and electricity the U.S. grid can't deliver fast enough. Mining legend Robert Friedland put it most vividly, and a parade of energy operators put numbers on it: conductor wire is twice as expensive as it was, gas turbines are sold out for years, and America has built 300–400 miles of long-haul power lines in five years when it needs 5,000.

Below: the Canada war and who's actually paying; the US-China chess game over rare earths and soybeans; the reality checks on reshoring pharma and chips; the critical-minerals scramble; the humanoid-robot hype meeting its first real financials; and the power crunch that sits underneath all of it.

1. The Canada trade war goes from sledgehammer to sniper rifle

After months of threats, the fight became real this week. Canada's retaliatory tariffs took effect September 8: 15%, 25% and 50% duties on about C$27.6 billion of U.S. imports, matching the 50% the U.S. put on roughly $20 billion of Canadian goods last month under an old law (Section 338 of the 1930 Tariff Act). Prime Minister Mark Carney pledged to hit back "dollar for dollar, rate for rate," with tariffs on U.S. steel, aluminum and iron doubled to 50% and targeted duties on furniture, clothing, appliances and dairy (Prof G Markets, "Canadian Economist: Trump’s Tariffs Are A Gift To Mark Carney," Sept 9; Bloomberg Daybreak: US Edition, "Canada Imposes Fresh Tariffs," Sept 8).

Washington's answer within hours is the part professionals should pay attention to. Instead of raising the tariff number, the administration barred Canadian alcohol, dairy and motorcycles starting September 29, added duties on cheese, boats, golf carts and mattresses, and (the big one) directed the General Services Administration (the government's purchasing arm) to start shutting Canadian suppliers out of federal and state procurement (RETHINK RETAIL, "Tariffs Hit Canada as AI Reshapes Product Discovery," Sept 11; The Rundown, "US-Canada Trade War Gets Uglier," Sept 9).

The sharpest read came from Sean Speer at Hub Podcasts, "Should Canada keep fighting Trump’s tariffs?" (Sept 9). He called the new sectoral tariffs "pretty marginal" (only about $1 billion a year of goods) but flagged the procurement move as "a bigger deal." His framing is the quote of the week for anyone with Canadian exposure:

"Don't underestimate that the U.S. government itself is sophisticated… if the administration reaches into the government and says, produce 5, 10, 15 ideas that we can sign through executive order… the system will do that."

The value at stake in that procurement system (everything from highway signs to trash cans for state parks) is an alleged $50 billion a year of Canadian contracts, versus the $1 billion in the headline tariffs. Speer warned the next moves could target visas, taxation, or even a review of Canadian banks' U.S. licenses, which he said could trigger something like a run on those banks' U.S. operations and pressure their share prices and capital ratios back home. His plain-English label for the strategy: "Chinese water torture," small, destabilizing hits every few days that keep uncertainty high and freeze investment decisions across Canada.

Why it matters: A tariff you can model. A government that can arbitrarily exclude you from contracts, slow your visas, or study your banking license is a much harder risk to price. The market hasn't fully absorbed that the toolkit just widened.

The politics cut against a quick deal

Both sides think they hold the stronger hand, which is exactly why this drags. Canadian economist Mike Moffatt (Missing Middle Initiative) said the tariffs, if they persist a year, shave only about 0.3–0.4 percentage points off Canadian GDP, and Carney has caught a lucky break: oil (Western Canadian Select) has jumped from about $50 to $80 a barrel on Middle East tensions, cushioning the blow. Politically it's a gift. Carney's Liberals lead 47% to 31% (Nanos), a 16-point margin, and roughly 90–95% of non-conservative Canadians back the counter-tariffs. As Moffatt put it, "You can't get Canadians to agree on 90 to 95 percent on anything. They agree on that" (Prof G Markets, Sept 9).

The trade-policy veterans at CSIS were blunt about the American misread. On The Trade Guys, "The Beef Imports, G20, and U.S.-Canada Trade Tensions" (Sept 9), Bill Reinsch noted the U.S. bet is "we're big and they're small… eventually they'll fold because the pain will get too great," and called it "an unproved thesis." His analogy: when a fight feels existential, people resist harder, not less. With Canadian opinion 73% behind Carney's rejection of the last deal, folding looks unlikely soon.

Crucially, the Trade Guys stripped away the noise (the French-language flap was a "red herring") to identify what the fight is actually about: China. The real U.S. sticking points (limits on Canada signing trade deals with non-market economies, and heavy-truck rules) both trace back to keeping Chinese components from slipping into North America duty-free. As Scott Miller put it, Chinese circumvention "is Trump administration's number one issue for North America," and they're working it "positively with Mexico" while worried Canada is drifting the other way. That's the crux: this is a China-access fight wearing a Canada costume, and it won't resolve easily.

The first hard corporate numbers: BRP

The most useful data point for investors came from an operator's income statement, not a pundit's opinion. On The Canadian Investor, "Trump Targets Canadian Companies" (Sept 10), the hosts walked through BRP's (Ski-Doo/Sea-Doo maker) latest quarter: revenue beat and grew double digits, but gross margins fell nearly 10 percentage points, with roughly 80% of that drop (about 7.5 points) attributable directly to tariffs. BRP now expects $200 million in tariff hits this year and $225 million in fiscal 2028. It even had to hand a distressed critical-parts supplier $75 million in emergency support ("an exceptional circumstance… never happened before").

The kicker: BRP has already been moving production and re-sourcing parts to dodge tariffs, and the bill still keeps climbing. Management said it won't relocate production for real until it gets "a predictable environment," which, as the hosts noted, may simply never come. Their conclusion, worth internalizing: this is not a temporary storm to wait out but "a structural shift… from globalization to deglobalization," and a change of administration in Washington probably won't reverse it (Biden kept most China tariffs; Democrats may soften tone, not substance).

Separately, Trump's threat to block Bombardier drew a useful fact-check: Bombardier draws about 56% of sales from the U.S. but keeps only ~6% of its property, plant and equipment there (versus ~89% in Canada), likely the basis for the "build more here" pressure. And on the bank threat, foreign banks can operate in Canada but face real restrictions (Bank Act Section 546 bars them from taking deposits under $150,000), which is what protects the big Canadian banks' retail franchise.

2. Who's actually paying: refunds owed, costs eaten, and a beef workaround

Away from the geopolitics, the week's small-business stories showed where tariffs land: on importers who can't raise prices, and on companies still waiting for refunds the government owes them.

On Marketplace, "What’s the new normal for global oil?" (Sept 10), fig-bar importer Peter Firth of FFF Associates said he has to absorb tariff costs because his customers, the bakers who make fig bars, refuse price increases that would dent sales. And Death by Audio's COO said the company is owed roughly $100,000 in tariff refunds it can't collect, so it can't hire.

That refund backlog is a real, quantifiable overhang. On Marketplace All-in-One, "As prices rise, so too does our credit reliance" (Sept 7), a Denver activewear retailer (AKTIV) reported paying about $27,000 in tariffs and getting only $3,000 back, around 11%. The bigger number: the Tax Foundation estimates $166 billion in tariff refunds are owed to businesses, with 40% still unreimbursed, and refunds have exceeded new tariff revenue in some recent months. This is working capital trapped across thousands of small importers, a quiet drag on hiring and inventory heading into the holidays.

On the political-relief side, the administration is trying to bring grocery prices down by importing more beef. On The Trade Guys (Sept 9), Scott Miller explained the mechanics: an extra 300,000 metric tons of foreign beef at a lower tariff rate (about 2% of annual U.S. consumption), mostly lean trim from Argentina's pasture-raised cattle, which pairs naturally with America's fatty trim to make 80/20 ground beef. He called it "a pretty smart policy" on the merits, but one that's easy to attack, which is why ranchers pushed back and an executive order for rancher relief followed.

3. The US-China chessboard: rare earths as the real leverage

The through-line in every China conversation this week: chips are America's leverage, critical minerals are China's, and a Xi-Trump meeting on September 24 is the moment both sides are circling.

The must-listen was mining financier Robert Friedland on Odd Lots, "Robert Friedland on the World’s Monumental Shortage of Copper" (Sept 12). His core argument on the copper tariff decision Washington is weighing: tariffs would genuinely help U.S. mining, but at a cost. "If copper is six bucks and you put on a 30% tariff… you'd have $7.80 copper. I could afford to hire a mine worker at the United Steelworkers of America and mine in the United States." The other side: domestic inflation, higher rates, and a midterm political problem. His frame is the whole debate in one line: "Do you want to re-industrialize America or do you not?"

The economics of why the U.S. lost the industry are stark in his telling: Congo copper grades of 3–6% and software engineers or equipment operators for "$800 or $1,000 a month," versus U.S. steelworkers wanting "over $100 an hour to go underground," plus environmental and permitting constraints. He noted America built about eight ships last year while China built roughly a thousand, and that mining companies in aggregate are worth "less than 1% of the S&P 500."

But his real warning was about critical metals as a weapon. Without gallium, "there is no NVIDIA"; you need scandium for solid-oxide fuel cells and rhenium in the core of a natural-gas turbine, and China controls the supply. He said U.S. manufacturers now have to submit their plans to a Chinese military commission to get certain metals, which Beijing can refuse "for any reason or no reason." His picture of the September 24 summit: it can go two ways, like "a game of rugby where I break your collarbone and you break mine but we don't kill each other," or it can spin out of control. He even revealed hyperscalers are now approaching miners directly ("just saying my god we got a problem") because they can't find the electricity or raw materials to build what they've promised. And on equipment: the wait for a giant mill ring gear has stretched from 4.5 years to "six or seven or eight," with force-majeure clauses now standard.

Sinologist Ryan Hass, on Sinica Podcast, "Strategic Stalemate: Ryan Hass on How Beijing Sees America and Trump" (Sept 11), reinforced the leverage read: in the 2025 escalation, when U.S. tariffs on China hit 145%, Beijing "pushed back and compelled President Trump to retreat," largely through its rare-earths chokehold, the same mutual-vulnerability dynamic (China's minerals vs. America's chips and jet engines) that governs the board now.

Soybeans: China buys ahead of the summit

Agriculture gave a real-time read on how the summit is being pre-negotiated. China is buying U.S. soybeans "pretty aggressively" ahead of September 24: it has purchased roughly 9 million metric tons toward a stated 25-million-ton commitment (about 36%), with fresh flash sales through the week, and soybeans hit a 32-month high (Grain Markets and Other Stuff, "Soybeans Hit 32-Month High," Sept 11; Professional Ag Marketing Podcast, "WASDE recap and market reaction," Sept 11). Analysts think those purchases could add 6–7 million tons to USDA's export projections, a concrete, tradable sign that Beijing wants the summit to go well.

One dissent worth logging, from the pundit side: economist Matt Klein on Top Traders Unplugged, "What If AI Actually Pushes Interest Rates Higher?" (Sept 9) argued U.S. tariff energy has been misdirected at Canada rather than China, whose trade surplus has ballooned from $100–200 billion a year pre-pandemic to over $1 trillion now, and that Europe may be next to act, potentially with 35–40% tariffs reflecting an estimated 35% undervaluation of the yuan.

4. Reshoring reality checks: pharma and chips

The reshoring cheerleading met some sober operator voices this week.

On generic drugs, Off Script, "What Will It Take to Reshore Generic Drug Manufacturing?" (Sept 8) featured consultant Tom Strohl (Oliver Wight Americas) on the 2028 tariff cliff designed to push generic production home. His verdict: tariffs are "a good start" but nowhere near enough. The India-U.S. labor-cost gap is "huge," and what matters is total landed cost: logistics, distribution, and the working capital tied up in inventory for perishable, expiration-dated drugs with notoriously volatile demand (miss a shipment and distributors "will penalize suppliers harshly"). His timeline reality check: two years is not enough to move the volume back, and reshoring the active pharmaceutical ingredients (APIs) is "actually more complicated than moving generic drug manufacturing" because of the plants and emissions permitting required. Translation for investors: the 2028 deadline will get things "in motion," but expect a long, messy transition and don't underwrite a clean supply-chain flip.

On semiconductors, the tone was more triumphant but thinner on this week's specifics. Keith Krach, on America’s Roundtable (Sept 13), credited the onshoring of TSMC as the catalyst for "$600 billion of investment in the U.S. semiconductor industry," pulling other chipmakers along.

The most concrete tariff-and-reshoring case study, though, was solar. On Clean Power Hour, "Section 232 Is a Price Floor, Not a Tariff" (Sept 8), the hosts argued the Section 232 tariffs on solar wafers and cells are pushing module prices up about 2 cents per watt every month while creating "chaos rather than spurring domestic manufacturing." The math on why: First Solar dominates U.S. module capacity at roughly 25% (about 14.5 GW of thin-film), while traditional crystalline-silicon cell makers account for only about 6 GW, not enough domestic supply to absorb the policy without prices simply rising. It's a clean example of the reshoring paradox: protect the industry faster than you can build it, and you mostly just raise costs. (For the policy mechanics, trade-compliance pros pointed to Sections 232 and 301 being used to rewrite country-of-origin rules, Simply Trade, "Be the Willow and Bend with the Winds of Trade Change," Sept 11.)

5. The critical-minerals scramble is now a Western industrial policy

If the theme of the week is "reshore the supply chain," the minerals episodes showed governments actually writing checks, and a wave of operators pitching that the West is finally paying up.

Tungsten (UK). The standout operator story: Tungsten West is restarting the Hemerdon mine in Devon, England, backed by a £71 million UK government funding package (£36M equity, £25M debt, a £10M standby facility) on top of a £40M equity raise in February. The CEO said they went from "fully funded to producing" in about a week. Why the government cares: China controls ~80% of primary tungsten concentrate but has swung from exporter to net importer as its own industrial and defense demand rises. At full run-rate (3,300 tons of tungsten oxide), Hemerdon could supply "20% or more of non-Chinese sourced production," and the UK government secured the right to buy half the output for energy and defense uses (Dig Deep, "Tungsten West, Critical Minerals, and the UK Mining Comeback," Sept 10).

Rare earths. CEO George Bennett made the "undervalued" pitch for Rainbow Rare Earths on Company Interviews (Sept 10). The hook: Rainbow recovers rare earths from phosphogypsum (fertilizer-industry waste already above ground) at 4,400 ppm, "6 to 10 times higher grade" than clay projects, with no mining risk. Its two projects (Phalaborwa in South Africa, Uberaba in Brazil) carry post-tax IRRs of 40–45%, EBITDA margins over 70%, combined EBITDA of about $300 million a year by 2030, and capex of only ~$280–330 million. His valuation argument: Rainbow trades at a ~$250 million market cap, while Cerro Verde was acquired for $2.8 billion for a similar ~$400 million EBITDA profile. He also flagged the policy shift that resets the whole sector's economics: U.S./Australian government floor prices of $110/kg for NdPr (the magnet metals), after which even Chinese NdPr prices rose to $120–140/kg. His framing of the stakes: rare earths are a small market by dollars but underpin "$6.3 trillion of industry."

Copper, at home. Domestic copper is stirring too: veteran miner George Ogilvie is drilling the Leviathan deposit in Idaho, targeting the ~20-billion-pound scale that attracts major producers (Mining Stock Daily, "The Mining CEO Behind Two Major Exits Starts Again," Sept 11).

Lithium, unconventionally. Two signals that lithium supply is being reimagined. Alkali Labs, funded by the Defense Department's BioMade institute, is using engineered microbes to pull lithium from Permian Basin produced water, potentially ~50,000 tons of lithium carbonate a year, roughly current U.S. demand (Chemical Week, "How can biomanufacturing help secure lithium supply?" Sept 8). And the International Lithium Association now counts 88 members (up from 30 four years ago) representing 85% of global production, with ExxonMobil, Chevron and Equinor joining, Big Oil moving into lithium via non-traditional extraction (The Global Lithium Podcast, Episode 243, Sept 8).

Why it matters: The West is no longer just talking about supply security, it's underwriting it with floor prices, government equity stakes and offtake rights. That changes the risk profile of Western junior miners, but it's a slow build, and most of these projects are years from meaningful volume.

6. Humanoid robots: the hype meets its first real financials

This was the week the humanoid-robot story got a reality check, from an actual balance sheet.

For the first time, we can see inside one of these companies. Ahead of its SPAC merger with Churchill Capital (valuing it at $2.5 billion), Agility Robotics filed financials. On The Robot Report Podcast, "How Monumental uses robotics to build walls" (Sept 11), the hosts laid out the numbers: $1.8 million in 2025 revenue, a $140 million operating loss, and $100 million of cash burned, a $2.5 billion valuation that's "roughly 1,400 times what Agility Robotics made in 2025." Its Digit robot costs about $400,000 to buy outright or $535,000 over five years as a service, "a long way from the $20,000 mark that has been promised" (that promise is Elon Musk's, and the hosts used it to show how unrealistic today's economics are). Agility has deployed Digit at nine customer sites for 65,000 operating hours; the tote-moving task it does works out to roughly $100,000 a year, and the question hangs there: "Can a person do that better?"

More reality checks: China's Unitree IPO'd less than a month ago and its stock has already tanked 53% from its high, with Chinese regulators moving to change how these firms list. And Figure is partnering with Nscale to deploy up to 100,000 Nvidia GPUs to train its models, which one host used to argue we may be heading toward an "AI winter" where compute becomes scarce and expensive as every company piles in.

The scale of China's lead is the sobering backdrop, laid out beautifully on Big Take Asia, "How India Is Training the Robots of the Future" (Sept 8). Citigroup projects the human-like-robot market could reach $7 trillion by 2050. China already accounts for more than 97% of global humanoid shipments and installed nearly 300,000 industrial robots in 2025, about eight times the U.S. The episode's most striking detail is about the fuel these robots run on: data. Because there's no internet trove of "first-person" footage of human hands doing physical work, companies are strapping iPhones to the foreheads of workers in India (plastic-sorters, factory hands) to record every micro-gesture. India provides raw footage for about $7 an hour ($30 when cleaned and annotated), versus Tesla paying $48 an hour for motion-capture data in Palo Alto. The buyers, hidden behind strict NDAs, are described as Google, Tesla, Nvidia and Meta. The poignant line: workers are "recording themselves into their own job extinction." Unitree's founder still predicts robots could do 80% of household tasks "within a decade."

The counter-thesis, and the more investable one this week, came from operators building task-specific robots, not humanoids. On Forging The Future, "Andy Lonsberry of Path Robotics on Why Physical AI Is Ready for the Factory Floor" (Sept 10), the Path Robotics CEO framed the labor problem driving reshoring: a shortage of about 320,000 welders over four years. His pitch is that adaptable (not just precise) robots open "80% of the market versus precision's 20%," and (notably for the automation-vs-jobs debate) he said no workers have been fired due to his robots' deployment. The economic reframe of the week came from a Tesla veteran on The Dynamo Show (Sept 9): the AI model is only 10% of the problem (data, sensing and speed are the other 90%), a "$300,000 robot replacing one worker" isn't viable yet, and the whole bet is that AI-enabled automation eventually makes electricity, not labor, the primary cost of manufacturing.

Meanwhile China's champions kept shipping: XPeng commissioned a dedicated humanoid factory in Guangzhou (its "Iron" robot has 76 degrees of freedom and three in-house chips delivering 2,250 TOPS), targeting mass production by the end of 2026, and its robotics unit raised $900 million at a $6.3 billion valuation in August (The Road to Autonomy, Sept 11; In Good Company with Nicolai Tangen, "XPeng Co-President," Sept 10).

Why it matters: The Agility filing gives the whole sector its first hard benchmark, and it's ugly. Expect the market to start separating "physical AI" companies with revenue and a task that pencils out from those selling a $20,000 humanoid dream that costs $400,000 today. And note the recurring punchline: every robotics thesis eventually routes back to electricity.

7. The power crunch: the wall everything hits

This was again the deepest, most-covered theme of the week, because it's the constraint underneath reshoring, chips, robots and AI all at once. The story broke into four parts.

The demand shock is real and sudden. The IEA expects data centers to drive roughly half of all U.S. electricity demand growth through 2030 (Brew Markets, "Apple Joins the Fold & Casey’s Consumer Check," Sept 9). The Futurists put scale on it: data centers could be 7–12% of U.S. electricity by 2035; a hyperscale campus draws 200–500 MW (as much as 415,000 homes), and a frontier AI campus 1 GW (830,000 homes) (The Futurists, "Power is China’s AI Advantage," Sept 8). The clearest single illustration: Texas has received about 474 GW of proposed new load, five times the state's record peak demand, prompting Governor Abbott to order a slowdown audit of 250–300 projects (The Neuron: AI Explained, "OpenAI Astra, Local AI, and the Hardware Race," Sept 9). And the interconnection queue math is brutal: about 230 GW of data-center interconnection applications over five years against only 93 GW of expected power-plant buildout (Interchange Recharged, "Speed to power," Sept 8).

Everything to build power now costs 2–3x more. The single most useful operator briefing was Shayle Kann on Catalyst, "Do data centers actually increase electricity prices? No…and yes" (Sept 10). His supply-chain inventory: "Conductor is like twice as expensive. Transformers are two plus times more expensive. Switchgear is two times more expensive. Gas power plants are two to three times more expensive." This is a genuine demand shock rippling through every component. The subtle-but-important point on rates: locally, a big new data center often lowers bills (EPRI found data centers have cut customer bills about 6% on average), because the utility makes the data center pay more than its fair share. But globally, rates are rising for almost everyone, because two-thirds of all utility spending is just maintenance and hardening of aging infrastructure, and all of that is now more expensive too.

The political backlash, "who pays," is now the story. Public opinion has turned hard against data centers (a recent Gallup poll cited water use first, then energy use, then electricity prices as the top objections), and, tellingly, people don't trust the utilities, developers or politicians telling them bills won't rise (Catalyst, Sept 10). Utilities plan $1.4 trillion of capex by 2030 (up 21% from $1.1 trillion), and one analyst argued it's the capex, not the demand, that drives rate hikes in a monopoly-utility model (Open Circuit, "Electricity bills are on the ballot," Sept 8). The industry's answer is ratepayer-protection tariffs: AEP Ohio requiring upfront deposits, an NV Energy-Microsoft protection deal in Nevada (Electric Perspectives, "EEI 2026: AI and Advanced Nuclear," Sept 8). In Canada, the same fight is live: about 20 Alberta data-center projects at $1B+ each, including a Meta facility of C$10–14 billion paired with a $5 billion gas plant (The Canadian Investor, "Data Center Revolt," Sept 7).

How they're trying to get power fast: gas now, nuclear later, transmission never. With the grid gridlocked, hyperscalers are going off-grid. Meta is deliberately installing inefficient, fast gas turbines to hit a 12-month timeline instead of the 24 months a grid-connected build takes (Growth Stories With IBD, "Voters Revolt Against Big Tech," Sept 11). The turbine bottleneck is so acute that GE is spending $11.75 billion to acquire CPP for specialized turbine components, and SpaceX is building its own casting operation to cut 18 months off deployment (Brew Markets, Sept 9). Nuclear is the mid-term hope: Holtec is seeking a $900 million IPO to restart Michigan's 800 MW Palisades plant, and SMR developers (NuScale's potential 6 GW TVA project, Bloom Energy's 2.8 GW Oracle deal, Oklo) are racing, but SMR power costs $80–130/MWh initially (versus renewables at $51–70) and won't scale for years (Brew Markets, Sept 9; The 7investing Podcast, "3 Nuclear Stocks Racing to Power the AI Data Center Boom," Sept 10).

The deepest structural problem is transmission. On The Futurists (Sept 8), the hosts spelled out the embarrassment: it takes 7–15 years to build long-haul transmission, and in the last five years the U.S. has built only 300–400 miles of it against the ~5,000 miles it needs, "not even 10% of what we know that we need." China, by contrast, will deploy 600 GW of solar in 2025–26 alone (the U.S. built 283 GW in 70 years) and builds long-haul lines about four times faster. Their proposed fix mirrors this week's Canada story in reverse: frame the grid as national security (as Eisenhower did to build the interstate highways) to break the state-by-state permitting logjam.

Zooming out, the capital scale is historic. On Capital Decanted, "The Prisoner’s Dilemma: Investing in the AI Build-Out" (Sept 8): hyperscalers spent $400 billion of capex in 2024, about 2% of GDP, or 8% when adjusted for five-year GPU depreciation, higher than the railroad buildout's 6% of GDP in 1872. There were 113 data-center transactions worth $69 billion in 2025 (including BlackRock/MGX's $40 billion Aligned deal), and utility capex grew 22% year-over-year to $212 billion. And a caution on concentration: Oracle's AI backlog hit $664 billion, but nearly half comes from OpenAI alone (The Rundown, "Oracle Rises As AI Backlog Hits $664B," Sept 11).

There's also a reshoring angle to power itself: an executive order declared a manufacturing emergency for grid components, with 60–80% of power-system parts imported and waits of five years for large gas generators and 2–3 years for transformers (Energy News Beat, "National Emergency Securing U.S. Bulk Power Sources," Sept 9).

Why it matters: Power is now the binding constraint on every growth story in this newsletter. The near-term winners are anything that shortens the path to electrons: turbines, transformers, on-site generation, battery storage. The near-term risk is the ratepayer backlash turning into permitting freezes (New York and Texas already paused projects) and the political "who pays" fight.

8. Autos: the tariff whipsaw and mixed policy signals

Autos remained a live tariff front, with the added twist of Washington sending conflicting messages to the same company. On InsideEVs Plugged-In Podcast, "Ford’s Fight With The Trump Administration" (Sept 11), the Department of Transportation criticized Ford's CATL battery licensing and Chinese partnerships (a Geely tie-up in Spain, building the Lincoln Nautilus in China until 2030) as troubling alignment with Chinese state-backed firms, while the White House simultaneously praised Ford's $3 billion Michigan battery plant and its 1,700 jobs. The hosts noted GM and Stellantis face no similar scrutiny despite comparable China ties, "confusing and inconsistently applied policy" that makes planning hard.

Meanwhile the reshoring capital keeps flowing (Ford investing about $1 billion in a Kentucky paint shop, part of $4 billion in Kentucky) even as Europe's auto base contracts hard: Volkswagen's German restructuring will cost an estimated €16 billion ($18.6 billion) with about 100,000 layoffs, and European suppliers (per CLEPA) have announced 104,000 cuts in 2024–25, potentially rising to 350,000 by 2030 on Chinese competition (Autoline Daily - Video, "AD #4375," Sept 11). And on the China-EV front, Canada is opening a small door: a 50,000-vehicle quota (under 3% of its market) that Tesla is expected to fill most of, while BYD, Chery and Geely line up to enter (The EVs for Everyone Podcast, Ep. 283, Sept 11).

Operators vs. pundits: whose voices to weight

A note on sourcing, because it matters this week. The most actionable signal came from people with skin in the game, not commentators:

  • Operators and insiders (weight these): BRP's actual tariff hit ($200M this year, ~7.5 points of gross margin); Robert Friedland on real equipment lead times and the copper math; Tungsten West's and Rainbow Rare Earths' CEOs on government-backed Western supply; Agility Robotics' filed financials ($1.8M revenue, $140M loss); Path Robotics' welder-shortage economics; Shayle Kann and utility executives on component costs and ratepayer tariffs. These are numbers from inside the machine.

  • Pundits and strategists (useful context, discount accordingly): the many "tariffs are futile / tariffs are essential" macro takes (Peter Schiff, Dividend Cafe), the Canada political read, and the big-picture AI-capex debates. Directionally helpful, but they're opinions, not order books.

Where the two disagreed most: pundits kept debating whether tariffs "work," while operators had already moved on to managing a world where tariffs, deglobalization and supply insecurity are permanent facts: re-sourcing parts, taking government equity, and racing for power.

The week ahead: what to watch

  • September 24: the Xi-Trump summit. The single biggest catalyst on the calendar. Watch for any deal linking China's rare-earth/critical-metal exports to U.S. chip and jet-engine access, the two levers every China episode circled. China's aggressive soybean buying is the tell that both sides want it to succeed.

  • The U.S. copper tariff decision. The White House has reportedly not yet decided on refined-copper tariffs; a 30% rate would, per Friedland, push copper toward $7.80/lb and reshape domestic mining economics, and copper prices are already jumpy on the speculation.

  • Canada's next move. Does Carney escalate again (possibly energy export taxes) or take the "off-ramp" some advisers favor after the relatively modest U.S. sectoral response? Watch the GSA procurement exclusions for how far the non-tariff toolkit goes.

  • U.S. midterms and mail-in ballots. Early mail-in voting begins soon; with ~30% of votes cast before Election Day and diesel at record highs feeding inflation, the political clock is a real constraint on how hard the administration can push tariffs into the fall.

  • Humanoid-robot disclosures. With Agility's numbers now public, watch for the next filings (Figure, others) and whether Unitree's post-IPO slide (-53%) spreads. The sector's first real benchmark is set, and it's low.

  • Power permitting and ratepayer fights. After Texas's slowdown audit and New York's pause, watch for more states inserting ratepayer-protection tariffs or freezing data-center approvals, the "who pays" backlash is now the binding political risk on the AI buildout.