Newsletter · · Ashutosh Agarwal

Tariff Refunds, a Chinese Robot IPO, and America's Power Crunch - Trade War, Tariffs & Reshoring - Week of August 24, 2026

Trade, reshoring and industrial-policy newsletter for the week of August 24, 2026. Tariff refunds are landing on income statements while a managed-trade wall goes up deal by deal, a Chinese humanoid-robot IPO popped 460%, and the grid emerged as the binding constraint on both AI and reshoring.

Trade War, Tariffs & Reshoring

Week of August 24, 2026: Tariff Refunds, a Chinese Robot IPO, and America's Power Crunch

The Week in One Breath

Three storylines ran through the podcasts this week, and they rhyme with each other more than they look like it at first.

First, the tariff money is finally moving, in both directions. Companies that overpaid duties before the Supreme Court struck down President Trump's emergency tariffs earlier this year are now getting real checks back, $128 billion of refund claims accepted so far, and more than 40 big companies have booked $9.6 billion of it. At the very same time, a new tariff wall is being built deal by deal: Canada spent the week staring down a 50% tariff, then apparently settled for a 15% "membership fee" to keep selling into the U.S., the same number Europe and Japan already accepted. The old free-trade world isn't coming back; a "managed trade" world is being poured in its place.

Second, the robots had their coming-out party, and it happened in Shanghai, not Silicon Valley. A Chinese humanoid-robot maker, Unitree, went public and popped 460% on day one. The message for U.S. investors: China is treating factory robots as a strategic industry, and the buildout is real, cheap, and fast.

Third, and this is the biggest, deepest story of the week, America is short of electricity, and everyone finally agrees on it. The head of the organization that keeps the U.S. grid reliable put a number on it: we need about 200 gigawatts of new power capacity in the next five to seven years, half of it for data centers. One energy CEO framed the stakes bluntly: a single year of grid delay can cost an AI company $12 billion in lost revenue, and China is building power ten times faster than we are. Power is now the thing that decides who wins the AI race, and, increasingly, whether factories can come home at all.

Let's go through it.

1. Trade & Tariffs: Refunds Flow Out the Front Door While a New Wall Goes Up Out Back

The refund cascade is real, and it's landing on income statements

The single biggest trade story of the week is money physically moving back to companies. After the Supreme Court struck down the President's emergency ("IEEPA") tariffs earlier this year, businesses started filing to get their overpayments back, and the checks are now large enough to move earnings.

As The Rundown laid out on August 20: as of the end of July, more than $128 billion of refund claims have been accepted and are being processed, and more than 40 S&P 500 companies have reported about $9.6 billion in refunds. Some of the checks are enormous: "Apple, for example, got nearly $2.2 billion in refunds. Nike got almost $1 billion. FedEx got around $800 million. And Target got more than $750 million." That last one is the eye-opener: for Target, the tariff refund "accounted for 40% of their total earnings for the quarter." In other words, a big slice of a quarterly profit came not from selling anything, but from getting overpaid taxes back.

Breaking Points (August 18) added the per-share detail that matters to anyone modeling earnings: Apple's refund was worth about $0.11 a share, and GE Healthcare's about $0.18 a share, alongside Nike ($986 million), FedEx ($800 million), Amazon ($640 million), and GM ($500 million).

Here's the part investors should sit with: what a company does with the refund tells you a lot about its pricing power. The Rundown noted that FedEx is passing the money back to customers, it literally built a website tracker so shippers can follow their refund. Walmart is doing the same on the consumer side: on Telltales (August 23) and again on Marketplace (August 20), the story was Walmart taking a $2.9 billion refund and putting it straight into price cuts for shoppers. Companies with strong brands and captive customers (think Apple) keep the money; companies competing on price (FedEx, Walmart) hand it back to stay competitive. Same refund, opposite strategy, and that difference is a live read on who has pricing power right now.

Two caveats worth holding onto. First, the refunds are not evenly spread and not fully paid: on Global Fresh Series (August 19), the figure cited was roughly $100 billion of CBP refunds issued so far, only about 60% of estimated collections, meaning a large chunk is still stuck in the pipeline. Second, this is now a lawyers' feeding frenzy: The Rundown flagged that Costco and other retailers are being hit with class-action lawsuits from shoppers demanding their cut of the refunds. The macro version, from The Dividend Cafe (August 17): somewhere between $25 and $50 billion a month is being returned to the private economy, a real, if messy, stimulus.

The new wall: Canada blinks, and "managed trade" replaces free trade

While the old tariffs get refunded, a new architecture is being built one country at a time, and this week that country was Canada. It was a genuine cliffhanger: a 50% U.S. tariff on roughly $20 billion of Canadian goods was paused on August 19 and set to snap back on August 22 if there was no deal, as Hub Podcasts reported (August 21). That episode spelled out how wide the damage would have been (the 50% would have hit "machinery, electronics, plastics, rubber, furniture, textiles, wood products, chemicals, food products, and clothing," with Ontario and Quebec facing roughly a 5 percentage-point jump in their average effective tariff rate and British Columbia about 7 points) and estimated the tariffs could cost Canada around 90,000 jobs. (For the record, Reuters World News on August 22 reported the 50% did briefly take effect at midnight before the deal news firmed up, the week really did go to the wire.)

The most useful framing of what Canada actually agreed to came from Sean Speer on Hub Podcasts (August 21). Strip away the drama, he argued, and the Trump administration's whole strategy makes a cold kind of sense: the U.S. market is so valuable that trading partners will simply pay a toll to keep selling into it. "In Europe, Japan and elsewhere, that premium seems to be about 15 percent. And if these reports are true, that's about where Canada is settling as well." The reported terms: autos tariffed at roughly 15% (down from 25%), with an exemption for U.S.-made parts; softwood lumber tariffs staying in place; concessions on Canada's protected dairy quotas; and "harmonization" of Canada's digital rules toward U.S. preferences. His verdict is the line to remember: "the days of the vision of continental free trade from north to south is over. We are in an era of managed trade."

Why should a U.S. investor care about a Canadian auto tariff? Because of where the factories go next. Speer's warning, echoing auto-industry contacts, is that 15% is not zero, and in a "low-margin business under so much pressure," a 15% wedge combined with the faster tax write-offs U.S. manufacturers now get under the "One Big Beautiful Bill" could produce a "giant sucking sound" of auto investment moving from Ontario to the U.S. That's the reshoring engine working exactly as designed: the tariff isn't really about the tax revenue, it's about relocating the plant. The auto-tariff standoff was a running thread all week: The Canadian Investor (August 22) pegged the terms at vehicles cut from 25% to 15% and steel and aluminum from 50% to 25%, and Automotive News Daily Drive (August 19) used the standoff as the backdrop for LG Energy Solution's battery-strategy pivot.

Not everyone is celebrating. Cost of Living (August 21), recorded when a collapse still looked possible, walked through what "no deal" would mean for ordinary Canadians, a reminder that "managed trade" has a human cost on the other side of the border.

Odds and ends on the tariff wall

  • Solar and drones: DHUnplugged (August 19) flagged a new 15% tariff on polysilicon solar products and a tariff on foreign-made drones aimed squarely at DJI, the dominant Chinese drone maker, a small but telling example of tariffs being used as targeted industrial policy against specific Chinese champions.
  • Tariffs coming off, too: Professional Ag Marketing (August 21) reported the administration waiving tariffs on 300,000 tons of imported ground beef (a 25% discount over 90 days) explicitly to bring down grocery prices. When beef prices get political, even a tariff-first administration will cut a duty. Direction of travel matters: tariffs are a dial, not a ratchet.

Who actually pays? The "companies are eating it" theory

The most interesting economic debate this week was whether tariffs are showing up in prices at all, and the surprising answer from several podcasts is: less than expected, because companies are choosing to absorb the cost.

On Macro Mondays and the sister Real Vision episode (both August 17), the argument was that corporations are suppressing price increases for two reasons: they'd face bad PR for hiking prices, and they're getting tariff rebates from the Treasury anyway. The evidence offered: consumer-goods producer prices fell roughly 75 basis points last month, and that's the category most exposed to tariffs. In plain terms: the one place you'd expect tariff-driven inflation to show up first is going the other way.

That's not the whole story, though. On Marketplace All-in-One (August 19), Wells Fargo economist Sarah House pointed out that imported computer prices are up 17% year-over-year, versus just 4% during the worst of the post-pandemic supply-chain mess, but noted that's being driven by AI data-center demand as much as tariffs. KPMG's Diane Swonk added the crucial timing point: import prices rise before tariffs even apply and before goods hit shelves, so the consumer impact comes with a lag. Fed officials are watching the same puzzle: San Francisco Fed President Mary Daly, on Bloomberg Surveillance (August 20), said producers are not yet pushing input-cost increases through to consumers. And economist David Rosenberg, on Wealthion (August 21), noted tariff revenue is adding "a couple of points to headline growth" while consumers push back so hard that some businesses (he singled out hotels) are rolling back earlier price increases.

The takeaway for a portfolio: the "tariffs = inflation" trade that everyone braced for hasn't fully arrived, partly because companies are eating the cost to protect market share. That's good for consumers and bad for margins, and it's a margin story that hasn't fully shown up in estimates yet.

China: the soybean tell

The clearest read on U.S.-China temperature this week came, as usual, from the farm podcasts. The setup: an expected Xi Jinping visit to Washington around September 24 is acting as the anchor for Chinese buying decisions. On AG Bull (August 23), the detail was that China's state grain buyer, Sinograin, has been auctioning off its imported soybeans weekly since late July, with clearance rates rising and offering sizes shrinking (from around 500,000 tonnes), a sign it's working down stockpiles, potentially to make room for new U.S. purchases tied to the summit. Hoosier Ag Today (August 17) put it more directly: there's "a very big book of soybean export sales to China as we speak," a sharp contrast to "very little on the books last year." Commodity Week (August 20) framed the target as the widely cited 25 million metric tons. Watch the summit: soybeans are the barometer, and right now the needle is pointing toward a thaw.

2. Reshoring & Critical Minerals: The Policy Is Real, the Geology Is Stubborn

If tariffs are the stick pushing factories home, critical minerals are the raw material those factories can't get without China's permission, and this week's podcasts made clear the U.S. government now understands that and is throwing real tools at it. The catch: policy can't overrule geology.

The clearest map of U.S. minerals policy all week

The single best explainer came from Zainab Usman on The Trade Guys (August 17). She sized the problem precisely: the U.S. is 50–100% import-reliant for 31 of the 60 minerals the U.S. Geological Survey considers critical, and heavily dependent on China for at least 14 of them. Then she laid out the three-part toolkit Washington is now using, worth understanding because it's the framework behind almost every mining headline you'll read this year:

  1. Financial tools: the government is literally buying equity stakes in U.S. mining companies through the DFC, EXIM, and the Department of Energy. (Government taking ownership positions in private miners would have been unthinkable a few years ago.)
  2. Regulatory tools: mainly permitting reform to speed up new mines.
  3. Trade deals: a growing web of "critical mineral trade deals" (she's careful to say they're not legally binding treaties): a U.S.-Japan deal covering seven minerals, plus agreements with Malaysia, Thailand, and the DRC, the beginnings of an "ex-China" trading bloc coordinated by the U.S. Trade Representative.

Her sober bottom line: the financial and permitting tools "have not really addressed the key challenge," because "no matter what you do domestically, you are hampered by geology," some minerals simply aren't in the ground here, and some refining "cannot be done domestically." That's why the trade deals with mineral-rich developing countries matter so much: you can't tariff your way to a mineral you don't have.

China is still holding the processing card

The reason all this urgency exists came through on The Northern Miner Podcast (August 18), in an episode literally titled "Mineral War." Guest Tomasz Nadrowski walked through China's rare-earth export controls, their bite on Japan and Europe, the resulting prices for specialty elements like dysprosium and yttrium, and, a detail worth noting, China's rare-earth collaborations with Iran. The uncomfortable truth investor Chris Temple put on Stocks To Watch (August 19): China controls 60–80% of rare-earth processing, and current U.S. permitting reform "addresses only 2% of the problem." Refining, not mining, is the chokepoint, and it's the part that's hardest and slowest to rebuild.

Copper: the metal everyone agrees is short

Copper was the week's most consistent minerals theme, with a striking amount of agreement across very different speakers:

  • Chris Temple, again on Stocks To Watch: the world "will need as much copper between now and 2050 as has been mined in all of history."
  • The Morning Call (August 18) cited BHP's CEO projecting copper demand rising from 34 to 50 million tons by 2050, needing 10 million tons of new production by 2035.
  • Investor Adrian Day on The David Lin Report (August 20) called it a "structural supply deficit," saying flatly "we simply don't have enough copper to meet the demand expectations," and stressed the picture is "very, very clear" for the next three to five years because mine timelines are so long.
  • On The Great Simplification (August 19), Ed Conway gave the physical reason it's so hard: copper ore grades have collapsed from 40% in 19th-century Montana to about 0.4% today, meaning you have to dig and crush 100 times more rock for the same metal (using enormous amounts of water and diesel) and net-zero plans "fail to account for these material constraints."

When a mining bull, a mining major's CEO, and a de-growth environmental commentator all land on the same conclusion (not enough copper) the direction is credible. Keep the healthy skepticism you always should around small-cap mining pitches (Temple's specific names, like Gunnison Copper "expected to supply over 10% of U.S. copper needs" and Abitibi Metals on The KE Report, August 22, are the promoter's job), but the macro deficit is not seriously contested.

The reshoring you can actually point at

Beyond the macro, a handful of concrete "it's actually happening" data points surfaced this week, mostly in smaller mining and battery podcasts:

  • Nickel: Company Interviews (August 17) covered Canada Nickel's Crawford Project winning federal approval plus C$21 million in funding, a tangible capital event, not a forecast.
  • Phosphate: Rock Stock Channel (August 20) marked First Phosphate's Nasdaq listing (backed by $20 million-plus from the Canadian government), and referenced the wider pattern of DOE support, including funding for Standard Lithium and a $1.2 billion loan for battery-materials firm Sila, plus EXIM Bank financing for Canadian projects.
  • Lithium from oil wells: On MicroCapClub (August 21), LibertyStream described pulling lithium out of West Texas oilfield brines, targeting commercial production in 2027, aiming for 600 tonnes a year for an Alabama industrial customer at about $6,200 a tonne, an example of the genuinely new extraction tech reshoring is trying to seed.
  • Rare-earth refining R&D: Dig Deep (August 20) highlighted a DOE-backed effort (eCLARA with Argonne National Lab and Virginia Tech) building an AI "digital twin" to crack the rare-earth separation problem, attacking exactly the processing chokepoint China dominates.
  • The financing model: Sustainable Business Podcast (August 18) explained why these deals are getting done: government offtake agreements and floor prices (guaranteeing a mine a minimum price so it can get financed) are the tools making otherwise-unbankable projects bankable. Floor prices are the quiet linchpin of the whole reshoring push.

One honest caveat: the mainstream "big-name reshoring announcement" (a new fab, a marquee auto plant) was thin in the podcasts this week, the reshoring evidence skewed toward small-cap mining and battery names. The broadest reference was CBRE's 2026 outlook (August 18) noting semiconductor and manufacturing reshoring clustering in the Midwest, Arizona, and Texas, and the housing demand it creates.

3. Automation & Humanoids: The Robot IPO Heard Around the World

The dominant automation story this week was not American. It was Unitree, a Chinese humanoid-robot maker whose Shanghai IPO became a spectacle, and a strategic statement.

What actually happened

On Squawk on the Street (August 19), the on-the-ground reporting captured how different China's robot moment feels: the retail portion of the IPO was oversubscribed 5,500 times, the founder is seen "in meetings with President Xi Jinping," and the robots are "heavily featured in TV galas" and at a running series of robot conferences. The company even launched a new robot called "Superman" that "can jump two meters high." And the follow-on effect is telling: competitors Lumos and Chery's robotics unit both signaled they're now considering IPOs of their own. This is a government deliberately inflating a robotics investment cycle the way it once did with solar and EVs. A nice detail for the "it's all connected" file: Madison Huang, Jensen Huang's daughter, who runs marketing for NVIDIA's physical-AI platforms, was spotted at the robot conference.

The financials, pulled together from The Rundown (August 21) and Squawk: Unitree surged 460% on debut to a valuation cited between $51 and $66 billion, on roughly $235–250 million of revenue and about $41 million of earnings, having sold ~5,500 humanoid robots last year. The contrast The Rundown drew is the whole point: 5,500 robots shipped "versus Tesla's zero for Optimus."

How the U.S. players stack up

That contrast may not last. On Rich Habits (August 21), the numbers cited were: Tesla targeting 8,000 Optimus units a month by December, Figure AI having manufactured 1,000 of its Figure 3 robots, and Morgan Stanley projecting Chinese humanoids will reach hundreds of thousands of units by 2030. Treat the Tesla target as ambition rather than fact until units actually ship, but the race is clearly being run on both sides of the Pacific.

For the "when does this pencil out" question, Chat GPT Podcast (August 21) offered the industry's rough economics: an average humanoid unit price around $55,000 by 2030, a payback of under a year, and 73% of companies saying they plan to deploy them (42% within one to two years). Those are survey-and-forecast numbers, not booked orders, but sub-one-year payback is the threshold that turns a novelty into a capital-budget line item.

The quieter, more investable automation story

Away from the humanoid hype, the more grounded robotics news this week was industrial and already generating revenue:

  • Autonomous construction: On Bloomberg Tech (August 17), Bedrock Robotics' CEO described using NVIDIA chips to run software that autonomously controls excavators and dump trucks, with paid deployments across three sites moving "multiple millions of cubic yards of earthwork," pitched explicitly as the answer to labor shortages from reshoring, data-center construction, and housing. This is the connective tissue: you can't build the factories and data centers without the dirt-moving labor, so the robots move the dirt.
  • Warehouse robots that don't need robotics engineers: The New Warehouse (August 17) featured Peer Robotics' human-in-the-loop mobile robots aimed at mid-sized operators who can't afford complex integrations.
  • Software-defined factories: The Robot Report Podcast (August 21) covered R2 Labs orchestrating existing factory machines with on-device AI vision, using a sheet-metal customer that detects defects and adjusts painting in real time.
  • The proof it works at scale: Gartner ThinkCast (August 20) cited Unilever running 120 robots that automate 90% of a logistics operation, cutting costs 24%. That's the number that makes a CFO lean in, not a jumping "Superman."

The honest read: humanoids are still mostly a story about pre-orders, forecasts, and national pride. The automation that's actually earning money today is the boring warehouse-and-jobsite kind, and that's where the near-term investment case is cleaner.

4. Power & Megaprojects: The Deepest Story of the Week

Every other theme in this letter eventually runs into the same wall: electricity. AI needs it, reshored factories need it, and America hasn't had to build much of it in decades. This week the podcasts turned that from a vague worry into hard numbers, and the numbers are staggering.

The definitive number: 200 gigawatts, half of it data centers

The most authoritative voice was Jim Massey, CEO of NERC (the North American Electric Reliability Corporation, the body responsible for keeping the grid from failing) on POLITICO Energy (August 17). His headline projection: "around 200 gigawatts of new capacity needed over the next five to seven years," and "about half of that is data center." To make it tangible, he offered his own yardstick: 200 gigawatts is "200 San Francisco's." We need to build the power for 200 San Franciscos, and we need it this decade.

Two things in his interview deserve special attention.

First, this is not only a data-center story, it's a reshoring story too. Massey was emphatic that people over-focus on data centers: the other half of the growth is "industrial facilities, battery plants, chip plants, all of this industry that left the country, you know, in the 90s and 2000s that's coming home." Reshoring and the AI buildout are drawing on the same strained grid at the same time. That's the crux of why this matters for every theme above: bring the factories home and power the AI boom, and you've doubled the demand on infrastructure that already can't keep up.

Second, he gave a genuinely useful tool for telling real demand from hype. With AI companies "shopping around" and utilities planning for load that may never appear, how do you know what's real? Massey's answer: follow the money. Utilities now require "quite sizable" deposits before they'll treat a project as real, and increasingly demand contracts locking a customer in "for at least 10 years" or paying "liquidated damages." When you see a data-center announcement, the question isn't the gigawatt headline, it's whether a deposit and a 10-year contract sit behind it. He also flagged the regulatory milestone to watch: NERC intends to deliver enforceable data-center connection requirements to federal regulators (FERC) by the end of this year.

The "we're losing to China" framing

The most vivid version of the crunch came from Hanan Happy, CEO of ExoWatt (an on-site solar-power startup backed by Sam Altman and Andreessen Horowitz), on Motley Fool Hidden Gems (August 23). His comparison is the one to quote in a meeting: "China adds 540 gigawatts of power to the [grid] per year. So more than 10 times our capacity", giving China, in his words, "10 to 11 times more capability and capacity than we do to build infrastructure and bring that online."

He also made the economic stakes concrete in a way that reframes power from a cost line to a competitive weapon: a single year of grid delay "could cost a hyperscaler $12 billion in missed revenue." And he sized the demand shift viscerally: just a couple of years ago the largest U.S. data center was about 100 megawatts; now the building blocks are 300–700 megawatts, average data centers are about a gigawatt, and some reach 10 gigawatts. "A gigawatt data center essentially is consuming the equivalent of 1 million U.S. households worth of energy. So you're basically saying I am building a city from scratch for a million people", and doing it as fast as possible. His warning about why this is so hard is the theme of the whole section: "It's not like software... You have to move metal. You have to move concrete. You have to move earth."

How big is the money, and can the supply chain even keep up?

The dollars are as large as the gigawatts. On Engineering Influence from ACEC (August 17), the consulting firm FMI forecast over $1 trillion of power-related investment through 2030, roughly half for transmission and distribution (the wires), a third for generation (the power plants), with data-center large loads driving 80% of the forecast increase. But the same analyst flagged the binding constraint: the industry "cannot currently produce and install equipment fast enough." Money isn't the bottleneck; the supply chain is.

Two operators put faces on that bottleneck:

  • The interconnection queue is the real traffic jam. On Supercool (August 19), Critical Loop's CEO argued the constraint isn't generation but the wires ("transmission and distribution") with utilities quoting 2–3 year waits to connect a new facility, and cases as long as 15 years, even though the grid is under 50% utilized. Renewable Rides (August 18) sized the pile-up: a Lawrence Berkeley National Lab study counts 10,300 projects waiting to interconnect, 1,400 gigawatts of generation and 890 gigawatts of storage. That's the queue that turns a "cheap" power project into a multi-year wait.
  • Turbine backlogs are brutal. On Open Circuit (August 19), Arbor Energy's CEO said traditional turbine makers now face 7-plus-year backlogs, which is precisely why so much money is flowing to novel approaches (his own company is building small CO2 turbines with 3D-printed parts to undercut conventional gas plants by the early 2030s).

The workarounds: go off-grid, or go nuclear

If the grid can't connect you for years, you build your own power. That's the "behind-the-meter" push, and it was everywhere this week:

  • On The Infrastructure Investor Podcast (August 17), EQT's Jan Vesely described a 20+ gigawatt land bank and 250+ gigawatts of energy projects, with the firm "happy to go fully behind the meter", pairing on-site gas with batteries so data centers don't add to consumer power bills.
  • Big Digital Energy (August 19) reported Meta's "Hyperion" project needing seven additional gas plants in Louisiana (via Entergy) to feed a single data-center campus.
  • Nuclear is having its moment: Energy Central (August 18) noted data centers could grow from 4% of U.S. electricity today to 12% by 2028, driving Amazon (over $500 million into Pennsylvania nuclear), Google, Meta, and Microsoft to lock up nuclear capacity. And on The Atomic Show (August 17), Radiant Nuclear's president detailed a one-megawatt transportable micro-reactor ("Kaleidos") shipping to Idaho National Lab for testing through March 2027 and to Buckley Space Force Base in 2028, already "cost-competitive in a quarter of U.S. energy markets," with plans to build 50 units a year. The pitch is grid independence: avoid the utility's delivery charges and the interconnection queue entirely.

The healthy counter-arguments

Not everyone thinks the sky is falling, and the dissenters are worth hearing precisely because the consensus is so loud:

  • On Ripple Effect (August 18), Wharton's Sergei Netasin agreed AI is pushing electricity demand up ~2% a year (versus 0.1% historically) but stressed the grid is only 55% utilized (there's slack to use) and, importantly, that AI data centers explain at most 10% of the recent 30–35% jump in electricity prices, with fuel costs and aging infrastructure doing most of the work. That's a useful corrective to blaming every rate hike on AI.
  • On Catalyst with Shayle Kann (August 20), Think Labs' CEO argued there's enough latent grid capacity to connect most data centers today without building dedicated on-site generation, and that piling generators behind the meter actually complicates grid operations rather than helping. He'd rather treat big data centers as "micro utilities" responsible for reinforcing the grid.
  • Most striking, on Gradient Dissent (August 18), PG&E's CEO Patty Poppe made the counter-intuitive case that load growth could actually lower electricity rates: when data centers, EVs, and heat pumps add demand, the utility's fixed costs get spread across more usage, so per-unit bills can fall, reversing decades of stagnant demand. If she's right, the "AI will crush ratepayers" narrative has it backwards.

The bull and bear cases aren't really in conflict on the facts, they're arguing about who pays and how fast. But the direction is not in dispute: the U.S. needs a vast, decade-long buildout of wires, turbines, and reactors, and the companies that supply that buildout, and the metals it consumes, are where the demand is heading. Backlash is building too: Factor This's cleantech roundup (August 21) noted that resistance to data centers is now "cross-partisan," a political risk that could slow permits even where the economics work.

Operators vs. Pundits: Who's Building and Who's Opining

A discipline we keep every week: separating the people actually building things (whose numbers come from their own operations and are worth extra weight) from investors and commentators (whose views are opinions, often sharp, but opinions).

Operators / insiders (running the actual businesses):

  • Jim Massey, NERC CEO (POLITICO Energy): 200 GW needed in 5–7 years; deposits and 10-year contracts as the reality filter; enforceable rules to FERC by year-end.
  • Hanan Happy, ExoWatt CEO (Motley Fool Hidden Gems): China building power 10x faster; $12B cost of a one-year grid delay.
  • Jan Vesely, EQT (Infrastructure Investor): 20+ GW land bank; going fully behind the meter.
  • Radiant Nuclear's president (The Atomic Show): micro-reactor to Idaho National Lab in 2027, Space Force in 2028.
  • Arbor Energy (Open Circuit) and Critical Loop (Supercool): 7-year turbine backlogs; 2–15 year interconnection waits.
  • Patty Poppe, PG&E CEO (Gradient Dissent): load growth could lower rates.
  • Bedrock Robotics (Bloomberg Tech), Peer Robotics (The New Warehouse), R2 Labs (The Robot Report): automation earning revenue on jobsites and in warehouses today.
  • LibertyStream (MicroCapClub), First Phosphate (Rock Stock Channel), Canada Nickel (Company Interviews): real minerals projects hitting funding and listing milestones. (Note: small-cap miners are also promoting their own stock, weight the milestones, discount the price talk.)

Pundits / analysts / economists (opinions and forecasts):

  • Chris Temple (Stocks To Watch, KE Report) and Adrian Day (David Lin Report): the copper-deficit and rare-earth-processing case.
  • Ed Conway (The Great Simplification) and Zainab Usman (The Trade Guys): the materials-constraint and minerals-policy framing.
  • Sean Speer (Hub Podcasts): the "managed trade" read on the Canada deal.
  • Sarah House (Wells Fargo), Diane Swonk (KPMG) (Marketplace All-in-One), Mary Daly (SF Fed) (Bloomberg Surveillance), David Rosenberg (Wealthion): the who-pays-for-tariffs debate.
  • Sergei Netasin (Wharton) (Ripple Effect): the "AI is only ~10% of the price rise" corrective.

When an operator and a pundit disagree, start with the operator's numbers, they have money riding on being right.

The Week Ahead: What to Watch

  • The Xi-Trump summit (~September 24). Every farm podcast this week treated it as the anchor for Chinese soybean buying. Watch weekly U.S. export-sales data and any Sinograin auction news as the tell for whether the thaw is real.
  • Canada deal fine print. The framework leaked this week (15% autos, lumber, dairy, digital rules); the actual signed terms, and Mexico's parallel CUSMA talks, are the next shoe to drop.
  • NERC's data-center rules to FERC by year-end. Massey committed to enforceable connection requirements; the first draft will shape which projects can actually plug in.
  • More tariff-refund disclosures. With 40+ companies already reporting $9.6B, the next earnings season will bring another wave, and the FedEx-vs-Apple "keep it or pass it back" split remains a clean pricing-power signal.
  • Robot-IPO contagion. Unitree's 460% pop has Lumos and Chery's robot unit lining up. Watch whether the enthusiasm jumps to U.S.-listed robotics and automation names.
  • Turbine and interconnection headlines. With 7-year turbine backlogs and a 1,400 GW interconnection queue, any easing (or worsening) is a direct read on how fast the power buildout, and the reshoring it enables, can actually go.