Newsletter · · Ashutosh Agarwal

Lutnick Says Build Chips in America or Pay Tariffs as Canada Retaliation Lands - Trade War, Tariffs & Reshoring - Week of September 7, 2026

Trade War, Tariffs & Reshoring for the week of September 7, 2026: Commerce Secretary Howard Lutnick spelled out the chip-tariff doctrine as build here or pay, Canada's dollar-for-dollar retaliation was set to hit September 8th after the first tariff-linked bankruptcy, and a new count put 189 gigawatts of data-center gas plants on the drawing board.

Trade War, Tariffs & Reshoring

Week of September 7, 2026: Lutnick Says Build Chips in America or Pay Tariffs as Canada Retaliation Lands

A weekly read on tariffs, reshoring, factory robots, and the infrastructure being poured to make it all run, built entirely from this past week's podcasts.

The week in one read

The conversation came roaring back this week, and it was noisy in every direction at once.

The single most important thing anyone said all week came from Commerce Secretary Howard Lutnick, sitting for a fireside chat on CNBC's Squawk Pod. Asked whether the administration is about to slap sweeping tariffs on imported computer chips, he didn't dodge. The tariffs are real, he said, but they are a lever, not the goal. "If you build in America, we will give you tariff relief. And if you don't build in America, then you will pay tariffs." That one sentence is the whole reshoring strategy in plain language: tariffs as a cudgel to force factories onto U.S. soil. It's the same playbook, he said, the administration already ran on medicines, and he claims it's working, pointing to "$1.2 trillion of commitment to build semiconductors in America."

At the same time, the U.S. and Canada trade war stopped being a threat and became a live event. Fifty-percent U.S. tariffs on roughly $20 billion of Canadian goods took effect over the weekend of August 21–22, and Canada's dollar-for-dollar retaliation, 15% to 50% on about $28 billion of American goods, is set to hit September 8th, the day after this issue lands. The first corporate casualty already filed for bankruptcy.

And underneath both stories runs the same current: an enormous, slightly frantic build-out of power plants, factories, and robots to feed the AI boom. One report this week counted nearly 200 gigawatts of new natural gas plants on the drawing board, an amount equal to roughly 40% of the entire existing U.S. gas fleet. Whether even half of that gets built is the trillion-dollar question.

Here's what was actually said, and why it matters.

1. The Canada tariff war goes live

For months this was a slow-motion negotiation. This week it broke.

The mechanics, laid out cleanly on Canada's Hub Podcasts: talks collapsed late on Friday, August 21st, and over the weekend the U.S. imposed 50% tariffs on about $20 billion of Canadian goods that had previously crossed the border duty-free under the North American trade pact (USMCA, known in Canada as CUSMA). The legal hook is unusual, Section 338 of the 1930 Tariff Act, a nearly century-old provision the administration reached for after courts blocked earlier tariff moves. Prime Minister Mark Carney has promised matching retaliation starting September 8th.

Three of Canada's big banks put numbers on the damage, and they don't fully agree, which is itself the story:

  • Scotiabank (economist Derek Holt) calls it a contained, sector-by-sector shock. It pegs the effective tariff rate at 6.5% across all Canadian goods exports and 8.6% on goods bound for the U.S., still, remarkably, the lowest rate the U.S. has imposed on any trading partner. Holt sees no meaningful hit to national unemployment and reads Ottawa's two-week delay before retaliating as a deliberate off-ramp.
  • National Bank (Stéphane Marion and Mathieu Arsenault) is darker: 100,000 to 130,000 Canadian jobs at risk if the tariffs hold and governments don't cushion the blow, pushing unemployment up about half a point from 6.4%. The average tariff rate on Canadian goods roughly doubles to about 6%, and the pain concentrates in apparel, textiles, electrical equipment, and furniture. Quebec and Ontario are most exposed (~11% and ~9%).
  • TD argues Canada can spend its way through it, counting more than $1 trillion in announced federal and provincial projects over the next decade, and it has already set aside about $500 million in provisions against trade risk.

One number worth holding onto, from RBC chief executive Dave McKay: more than 80% of U.S.-bound Canadian exports still cross the border duty-free. This is a war fought at the margins, but the margins are where jobs live.

Who blew it up? Lutnick's version, on Squawk Pod, is that a deal was essentially done and the Canadians torpedoed it at the last minute for domestic political reasons, showing up "Friday at four o'clock" with a new demand to exempt medium and heavy trucks (think big commercial rigs, not pickups) that "would have cost America billions and billions of dollars." He tied the whole fight to Canada's electoral calendar, a sovereignty vote in Quebec and an independence push in Alberta, and predicted Ottawa turns cooperative again after October 12th, once those elections pass.

The Alberta Premier, speaking on Squawk on the Street, took the operator's view from the other side of the border: autos and energy supply chains are so tightly stitched together that both countries lose, and threats like cutting off electricity exports to the U.S. would be self-defeating. The Premier's plea was simply for the president and prime minister to get in a room directly before the September 8th escalation.

The skeptics were louder than usual. Economist Phil Magness, on The David Lin Report, argued the administration fundamentally misreads what a trade deficit is, treating it as money "lost" rather than a mutual exchange, and noted the roughly $62 billion gap Trump cites with Canada is mostly energy the U.S. chooses to buy. His bottom line: more than a year of tariffs has not actually shrunk the overall U.S. trade deficit. And investor Kevin O'Leary, on the ITM Trading Podcast, called the 50% rate flatly "unsustainable," warned it could tip the economy toward recession and triple some grocery costs, and predicted Carney comes back to Washington within 60–90 days to cut a deal, with uranium and potash likely carved out, because the U.S. simply can't replace them (America burns about 50 million pounds of uranium a year and mines barely 1 million).

The first casualty

You don't have to wait for the macro models. A furniture maker just failed. On Furniture Industry News, the story of the week was Southern Motion Inc., a Mississippi recliner manufacturer, filing for Chapter 11 bankruptcy on August 31st and asking for $5 million to keep operating. CFO Mary Beth Mayer named the causes directly: "a slow retail furniture market and increased manufacturing costs associated with tariffs," plus fuel and shipping costs tied to global tensions. The Canadian 50% duty that went live August 22nd is stacking on top of container rates that jumped 25–30% in mid-August. (The parent company, Hong Kong listed Man Wah Holdings, also points to a landlord dispute, so this isn't tariffs alone, but tariffs are front and center in the company's own telling.) Bassett and FlexSteel have both flagged higher energy costs too. This is what the "sectoral shock" the banks describe looks like on the ground: one plant, one payroll, at a time.

The practical fallout for anyone who imports

Two trade-desk podcasts turned this into a to-do list. On Logistics Matters with DC Velocity, Mark Burstein of Inspectorio walked through why autos are the hardest case, parts cross the U.S. and Canada border multiple times before a car is finished, and warned that companies now have to prove the country of origin and trade-pact eligibility of every component, or risk penalties including a 40% tariff for "transshipment" (routing goods through a third country to dodge duties). And on Selling on Giants, the message to e-commerce brands was blunt: Canada's retaliation hits "hundreds of product categories" from September 8th, steel, aluminum, furniture, electronics, cosmetics, and you can no longer assume that because you shipped it from a U.S. warehouse it counts as American-made.

2. Build here, or pay: the reshoring doctrine gets explicit

The Canada fight is the noise. The semiconductor plan is the signal.

On Squawk Pod, Lutnick spelled out a strategy that ties directly to the reshoring theme this newsletter follows. The design: tariffs on imported chips, but with relief for any company that builds in the U.S., the exact structure already used on pharmaceuticals. He framed it as a done deal in the industry's mind ("all of the companies know they're coming"), and rattled off the scoreboard he's keeping:

  • $1.2 trillion in committed U.S. semiconductor investment.
  • U.S. chip production going "from less than 2% when we walked in" toward 40%, and, if Intel's turnaround works, potentially 50%.
  • TSMC building a "$265 billion" complex in Arizona; Micron committing "$250 billion" of memory factories, over half a trillion dollars from two companies alone.
  • Taiwan, as part of its trade deal, agreeing to move 40% of its chip production to America, with another $20–30 billion round of investment supposedly to be announced within the week.

Whether those figures prove out is a fair question. These are the administration's own numbers, and cumulative "commitments" are not the same as poured concrete. But the strategic logic is now stated out loud: the tariff is the stick, the factory is the point. Lutnick also repeated his revenue math, tariff income "crossing $300 billion a year, $400 billion a year," as part of a deficit-reduction argument. It's worth noting that even sympathetic accounting struggles to make that add up: the Hub Podcasts segment cited an estimate that U.S. interest payments alone will clear $1 trillion this year, more than double what tariffs bring in, against a cumulative deficit projected near $24 trillion over the next decade.

The counterweight came from the currency world. On VoxTalks Economics, the argument was that average U.S. tariffs now sit around 12% and that this alone has nudged U.S. interest rates up by about 0.4 percentage points while quietly eroding the dollar's special status as the world's safe-haven currency, because tariffs shrink America's effective footprint in global trade. If tariffs were to double toward 26%, the guest argued, the euro starts to look like a genuine alternative reserve currency. That's the long-tail risk hiding behind the reshoring win: you can pull factories home and weaken the dollar at the same time.

3. China, soybeans, and the summit that sets the tone

The U.S. and China relationship stayed in "managed truce" mode, but with a date on the calendar. Morgan Stanley strategists on Thoughts on the Market flagged a U.S. and China summit set for September 24th as one of the fall's key policy events. Their base case is modest: both sides trying to hold the "managed stability" reached in the spring, with small, incremental deliverables, some agricultural purchases, some Boeing orders, a bit of AI dialogue, but with real room for "tactical escalation" if either side wants it.

The most tangible piece of that is soybeans, and here the news was genuinely encouraging for U.S. farmers. On Grain Markets and Other Stuff, the read was bullish: China has already bought about 7 million bushels in a recent flash sale and is on track to hit its roughly 25-million-metric-ton purchase goal, with buying "accelerating at a pretty good clip." The timing is still fuzzy, the initial 12-million-ton tranche is spread from late fall into February, but after a long stretch where soybeans were a political football, actual purchases are showing up. Think of soybeans as the barometer for whether the September 24th summit is producing anything real.

4. Autos: the apocalypse that wasn't (yet)

One of the more useful reality-checks of the week came from Shift: A podcast about mobility, asking why the tariff-driven "autos apocalypse" analysts predicted never actually arrived. KPMG's automotive lead argued the industry simply adapted, absorbing costs through sticker-price credits, refunds, and reshuffled supplier relationships, and now treats tariffs as one more source of ongoing business volatility, alongside the EV slowdown and Chinese-brand competition, rather than a knockout blow. It's a helpful corrective to doom narratives: big, capital-heavy industries are better at swallowing shocks than the headlines suggest. The caveat is that Canada's escalation adds a fresh, integrated-supply-chain twist that the earlier adaptation didn't have to price in.

5. The reshoring supply chain: minerals and chips get real projects

Reshoring chips and cars only works if the raw materials and the fabs exist. This week brought concrete, on-the-ground project news, mostly from operators, not commentators.

Critical minerals. On The KE Report, Stillwater Critical Minerals CEO Mike Rowley reported that the company's Montana deposit (Stillwater West) more than doubled in its updated resource estimate to 267 million tons, holding an estimated 4.8 billion pounds of nickel, copper, and cobalt plus 7.4 million ounces of platinum-group metals, gold, and rhodium across eight metals. The pitch is squarely a reshoring pitch: a domestic U.S. source for exactly the minerals that electric motors, batteries, and defense hardware need, and that China currently dominates the processing of.

Lithium. On The Global Lithium Podcast, Standard Lithium's team described Arkansas's Smackover formation as potentially able to support 200,000–400,000 tons of annual lithium carbonate production across multiple projects over time. The reality check embedded in the same conversation: the first phase (Southwest Arkansas) is still under construction, the final investment decision is targeted only for this fall, and offtake agreements and financing (partly via government export-credit agencies) are still being nailed down. Reshoring the battery supply chain is real, but it moves at the speed of construction and permitting, not press releases.

Chips. On The Economics Show, Dan Kim tracked the CHIPS Act build-out through actual facilities, TSMC's Arizona fabs, Intel's next-generation 18A process, Samsung's Texas plants, the physical counterpart to Lutnick's "40% of production in America" ambition. And on Bloomberg Tech, the marquee corporate move was Nvidia's $3.5 billion investment in MediaTek, a reminder that even as Washington pushes production onshore, the chip supply chain remains deeply, expensively interlinked with Asia.

6. Robots on the factory floor: the humanoid versus specialist fight

If reshoring is going to work despite higher U.S. labor costs, robots are the assumed answer. This week the operators building those robots picked a very public fight with the hype.

The bullish, on-the-ground data point came from The Road to Autonomy: Chinese humanoid maker UBTech shipped 921 humanoid units in the first half of 2026, largely into automotive and factory assembly work, with BYD and Xiaomi now entering the field. Real robots, doing real factory jobs, in real quantities.

But the most interesting voices were the founders throwing cold water on the humanoid dream:

  • Shamoon Siddiqui of Human Friendly Robotics, on The ConTechCrew, flatly rejected general-purpose humanoids (Tesla's Optimus, Unitree, Figure) as inefficient for actual work, and instead builds single-task machines, one robot for laying floor tile, another for pulling electrical wire.
  • Rajat Bhageria of Chef Robotics, on Smart Friends, made the same argument from the food-production world (his robots have made 130 million meals): business customers want task-specialized machines that are superhuman at one thing, not a clumsy humanoid that's mediocre at everything.

The pundits agreed the timeline is long. On ARK's FYI – For Your Innovation, the take was that commercialized humanoid robots are 10-plus years away, a far harder problem than self-driving cars because factories and homes are so unpredictable and there's so little training data. Figure AI is even paying people to wear headsets to gather that data.

Meanwhile the money keeps coming regardless of the timeline. This Week in Startups documented "physical AI" funding heating up, citing Nvidia CEO Jensen Huang's line that every industrial company will eventually become a robotics company. And in a nice preview of that future, TechLinked reported Meta is already testing data-center maintenance robots, from Watney Robotics, Kinova, and ABB, to plug cables and reset servers. Not humanoids marching through a factory, specialized machines quietly doing specific jobs. That's the near-term reality.

The takeaway for the reshoring thesis: don't underwrite a U.S. manufacturing renaissance on humanoid robots arriving on schedule. The people actually deploying automation are betting on narrow, task-specific machines, and those are here now.

7. The power crunch: the biggest build-out nobody can quite measure

Everything above, the fabs, the reshored factories, the robots, the AI behind them, runs on electricity. This was the week's deepest and best-reported theme, and the numbers are staggering and slippery in equal measure.

The gas build-out. The flagship story came from Shift Key with Robinson Meyer. A new Global Energy Monitor report counted 189 gigawatts of new natural gas plants, announced, permitted, or under construction, specifically tied to data centers, nearly double the ~97 gigawatts the same group counted just six months earlier. To grasp the scale: the entire existing U.S. natural gas fleet is about 512 gigawatts, so this is roughly a 40% expansion, a pace not seen since the early 2000s. The honest caveat, stressed repeatedly: much of this may never get built. The same "gigawatt" of demand gets requested in multiple places, and huge headline projects (like the ~10 GW OpenAI and Department of Energy site in Ohio, financed partly by Japan and backstopped by Nvidia, or the ~17 GW Fermi America project in Texas associated with Rick Perry) are being built in small phases, often just a few hundred megawatts first. The demand is real; the specific project list is "sand slipping through your fingers."

Why gas and not solar? Three reasons stacked up, per the same conversation: the grid is so backed up that developers go "behind the meter" (building their own power on-site rather than waiting years for a grid connection); the administration has cut clean-energy tax credits and slowed wind and solar permitting; and gas is energy-dense enough to fit on a data-center's own land, which solar isn't. One sobering data point: polling cited on the show found 75% of Americans would oppose a data center being built near them.

Musk enters the turbine business. A recurring thread, on TechCrunch Industry News, was Elon Musk's move to have SpaceX manufacture gas-turbine blades in-house, claiming it can pull turbine production forward by up to 18 months. Why does that matter? Because the turbine makers are sold out: GE Vernova is booked "through 2030." (On How to Trade Stocks and Options with OVTLYR, the backlog figures cited were about 116 GW for GE Vernova and 69 GW for Siemens Energy, the turbine is now the bottleneck the whole build-out waits on.) The TechCrunch segment also flagged the cost: gas turbines in Virginia's "Data Center Alley" are estimated to cause 3.4–6.5 premature deaths per facility per year and $53–99 million in annual health damage. The power has to come from somewhere, and it isn't free of consequences.

Who pays? This became the week's real fight, and it connects the AI build-out straight back to ordinary households. On The Energy Show, researcher Mark McNeese noted a single AI data center now draws 600+ megawatts continuously, a small city's worth of power, all day, and that Texas has requests in its pipeline totaling five times current statewide demand, requests even the developers don't believe will all materialize. The risk: utilities build gas plants for demand that never shows up, and ordinary ratepayers get stuck with the bill for trillion-dollar companies' infrastructure. That fear was echoed on Volts, where the point was that every U.S. electricity market is failing to fairly assign these costs: one hyperscale campus in the PJM grid (the mid-Atlantic system) reportedly hit 30% of local peak demand and triggered charges that spilled across 13 states. And on Motley Fool Hidden Gems Investing, the concrete household number: electricity bills up about 5% over the past year, with a further ~6% impact expected.

The scale everyone agrees on. Even the disagreements share a common floor. UBS's Ulrike Hoffmann-Burchardi, on Alpha Exchange, put the shortfall at roughly 100 gigawatts by 2030 to cover AI plus general electrification, and made a striking point about how insensitive the AI build-out is to cost: hyperscalers are already paying triple the going rate for electricians and double for electricity, so it would take more than a half-percentage-point jump in power costs to slow them down. Moody's Talks sized the construction spending at roughly $130 billion a year by 2030, up from $60–70 billion today. And operator Zach Dell, on David Senra's podcast, framed the whole thing simply: his company (Base Power) exists because electricity demand growth is jumping from about 2% a year to something like 10%, and the grid can't keep up.

Bull versus bear. The financial-market frame for all this was a clean two-sided debate. On The a16z Show, investor Gavin Baker argued AI demand is genuinely outrunning the supply of compute, the bull case for keeping the build-out going full-throttle. On RiskReversal, Paul Kedrosky argued the opposite, that the entire AI investment cycle now hinges on a single company, Nvidia, making it a "single point of failure" if demand disappoints. You don't have to pick a side to see the reshoring relevance: the factories, the power plants, and the robots this newsletter tracks are all, ultimately, bets that this demand is real and durable.

Operators versus pundits: who to weight this week

Separate the people building things from the people talking about them. Both are useful; they carry different information.

Operators worth their weight this week (they have skin in the game and specifics):

  • Sec. Howard Lutnick, not an operator in the usual sense, but the policymaker whose "build here or pay" rule will actually set the terms for every reshoring decision. Weight heavily, discount the self-scoring numbers.
  • Mike Rowley (Stillwater) and Standard Lithium's team, real deposits, real permits, real timelines, including the honest admission of how far off production is.
  • Shamoon Siddiqui (Human Friendly Robotics) and Rajat Bhageria (Chef Robotics), founders who deploy automation for a living and are betting against the humanoid hype. When builders contradict the narrative, listen.
  • Zach Dell (Base Power) and the Alberta Premier, grounded, first-hand views of the power crunch and the auto and energy supply chain.
  • CFO Mary Beth Mayer (Southern Motion), the rare voice naming tariffs as a direct cause of an actual bankruptcy.

Pundits, useful for framing, but discount the confidence:

  • Kevin O'Leary and Phil Magness, sharp, directional views on the Canada fight, but predictions (60–90 days to a deal; tariffs "don't work") are opinions, not data.
  • Gavin Baker versus Paul Kedrosky, the bull and bear AI debate is genuinely informative precisely because they cancel out; hold both.
  • Bank economists (Scotiabank, National Bank, TD), closer to operators in rigor, but note they disagree by a factor of two on the job-loss numbers. The spread is the signal: nobody knows yet.

The week ahead: what to watch

  1. September 8th, Canada's retaliation goes live. The single most concrete catalyst. Watch whether Ottawa actually pulls the trigger on the full 15–50% counter-tariffs or blinks; Scotiabank read the delay as a deliberate off-ramp.
  2. The semiconductor tariff order. Lutnick said companies "all know they're coming." Watch for the formal rollout and the promised Taiwan investment announcement (another $20–30 billion) he teased for the coming week.
  3. September 24th, the U.S. and China summit. The tone-setter for the fall. Soybean purchase pace is the early tell on whether it's producing anything real.
  4. Canada's provincial politics. Lutnick explicitly tied U.S. and Canada de-escalation to elections clearing after October 12th (Quebec's sovereignty vote is October 5th). If he's right, the trade war eases; if he's wrong, it hardens.
  5. Turbine and power bottlenecks. GE Vernova is sold out to 2030 and Musk is trying to build his own turbines. Watch whether the gas build-out's headline gigawatts start turning into poured concrete, and whether the "who pays" backlash reaches your electricity bill.