The Boom Lenders Can Recall

Plus: Mosseri takes the stand, China's record recall, Australia bars AI-made songs.

Here's What's on Our Plate Today

  • 🧪 The hose shop in Bremen, Georgia: how the AI build-out delivers what tax cuts promised.

  • 🍪 Mosseri testifies in Meta's addiction trial, China recalls 4.3M cars, Australia bars AI-made songs.

  • 🛠️ Three things worth trying: EIA's price data, Subsidy Tracker, Data Center Map.

Let’s dive in. No floaties needed.

In partnership with

Get certified in digital marketing — for free.

Most marketers learn digital marketing by doing it — trial, error, and hoping it sticks. But when your pipeline depends on it, there's no room for guesswork.

HubSpot Academy's Digital Marketing Certification covers everything that moves the needle: SEO that actually gets you found, email strategies that convert, social media and paid ads that reach the right audience, and AI tools that make your team faster without adding headcount.

3 hours and 24 minutes. 10 lessons. 30 videos. Free.

Join 200,000+ certified professionals who have built sharper strategies — and the skills to prove ROI to leadership — with HubSpot Academy.

Get certified today. No cost. No credit card.

*This is sponsored content

Blu Dot surpasses 2,000% ROAS with self-serve CTV ads

Blu Dot used Roku Ads Manager to drive incredible results for its furniture sales event. Its strategy hinged on custom audiences and retargeting, where intent was strongest.

“Roku has been a top performer,” said Blu Dot’s Claire Folkestad. “We have seen…CPMs lower than any other CTV partner we've worked with.”

*This is sponsored content

The Laboratory

TL;DR

Trickle-down works after all. This version just arrives with an invoice.

  • Why it travels now: a tax cut can sit in a buyback. Building something forces purchases of concrete, steel, and electrician hours, so the money reaches a hose shop in Georgia.

  • Where it stops: in the host towns, jobs arrive, but wages don't move, and housing gets pricier. Much of the equipment is imported anyway.

  • Then the bill lands: the same households pay more for electricity, since the grid charges everyone under the wires for the capacity data centers need.

  • It can be called in: the boom runs on borrowed money, unlike a tax cut, so the chain carrying orders outward carries cancellations back.

  • Nobody's netted it out: the gains have names and numbers. The costs appear as a rate on a power bill, which is why nobody can yet say whether it was worth it.

The AI build-out is doing what tax cuts promised, & billing the same towns for it

Southeastern Hose is a family business in Bremen, Georgia, that makes corrugated metal hoses for steel mills and chemical plants. It is unglamorous work, and for years the order book rose and fell with heavy industry. Then a new kind of customer started calling, because data centers, the warehouse-sized buildings full of computers that run AI systems, need cooling and exhaust lines too. Reuters reported on August 19, 2026, that the firm's revenue has tripled in five years, and that its workforce has grown from 90 people to 150.

60 hires at a company of 90 is not a marginal adjustment on a spreadsheet. It means the shop runs more shifts, and it means finding 60 people willing to work them. Each of those people now takes home a wage that nobody in Bremen was earning five years ago, which is what people mean when they talk about growth reaching a town.

The same thing is happening further up the chain at Generac, the Wisconsin company known for the backup generators that start when a storm cuts the power. The firm is retooling its factories for the far larger machines data centers need, at a cost it puts at $250M by the end of 2027. It says it has $1.6B of orders booked and expects to hire about 1k people.

Money arriving at the top of the economy and ending up in a hose maker in Georgia is what trickle-down was always supposed to look like. The promise was about tax policy, and it ran roughly like this. Cut taxes on wealthy people and profitable companies, and they will invest the savings, benefiting everyone else through jobs and wages.

Decades of evidence say the money mostly did not make that trip. The most careful test of the idea, published in Socio-Economic Review, tracked every major tax cut for the rich across 18 wealthy countries between 1965 and 2015. The cuts made the rich richer, lifting the top 1% share of national income by more than 0.7 percentage points, and did nothing measurable to growth or unemployment.

The reason has less to do with greed than with instructions, or the lack of them. A tax cut hands a company money and says nothing about what to do next, so it can go into a share buyback (a company spending its cash on its own shares) and never leave the building. Deciding to build something is different, because the decision to build is also a decision to buy concrete, steel, transformers, cable, and a few hundred hours of an electrician's time.

It is the difference between a neighbor who wins money and a neighbor who builds an extension. The winnings might be spent on your street, in another town, or not at all. The extension has to pay a bricklayer, since it cannot go up without one. A data center is that extension at an industrial scale, and the bricklayer, in this case, is a hose shop in Bremen.

How far the money actually travels

The harder question is what happens in the towns that host the buildings themselves, which are not the same places as Bremen. Two economists, Dany Bahar and Greg Wright, went looking for that end of the chain and hit a problem straight away. Counties that win data centers were already doing well because the buildings are built where land is cheap, power is plentiful, and the local economy is growing. Comparing those counties with the rest of the country makes a data center look responsible for growth that was already underway.

Their solution, published by Brookings and updated on August 10, 2026, was to find 52 data centers that were announced and then canceled. Those counties had done everything the winners did, right up to the point where the project fell through, which makes them a fair comparison. Measured that way, a county's first big facility lifted employment in data processing by 56% over a decade, which comes to somewhere between 100 and 200 jobs.

For the people who already lived there, the picture is thinner than the announcement suggested. Wages did not rise at all, and home prices rose by 2% to 5%. A family that was in the county before the trucks arrived got no raise, a more expensive house, and a neighbor or two with a new job.

Suppliers like Southeastern Hose sit outside that measurement and do better, because those are jobs at firms selling into the boom rather than jobs at the buildings themselves. A further share of the money never reaches an American payroll at all. Much of the equipment inside these buildings is bought abroad, as a Federal Reserve staff note published on July 17, 2026 sets out.

The same towns get the electricity bill

The other half of the story arrives in the mail, one month at a time. A data center is a customer that buys electricity constantly and never closes for the night, and the grid it plugs into was built for a smaller world. A separate Brookings analysis published on March 13, 2026, found that electricity costs have risen 42% since 2019, while overall consumer prices have risen 29%. Data centers are one of several pressures on that number.

PJM, the grid operator that keeps the lights on for 65M people across 13 states, pays power plants years in advance to guarantee they will be available when demand peaks. The cost of those guarantees rose from $2.2B to $14.7B in a single year, and data centers accounted for nearly two-thirds of the increase. That money does not stay with the grid operator because it is collected from the households and businesses living under those wires.

So the family that got no raise from the new facility is now paying more to keep its own lights on. That is a real transfer, and it runs in the opposite direction from the one described in the old promise. The case for paying it is that AI will eventually make everything else cheaper. That case is not settled. An economist at the International Monetary Fund told Reuters on August 20, 2026, that the productivity gains from AI may not be large enough to hold prices down.

The money paying for it is borrowed

The last difference is where the money comes from, and more of it is borrowed than it looks. The Bank for International Settlements described the arrangement in its Quarterly Review of March 16, 2026, and it works like this. A separate company is set up to own the data center, outside lenders put up the loans, and the technology firm takes a small stake and signs a long lease. The lease is a bill it has to pay for years, though it counts as rent rather than debt, which the BIS calls 'shadow borrowing'.

A tax cut stays in place until a legislature repeals it, while a building boom paid for with borrowed money stops the moment lenders lose their nerve. The chain that carried the orders out to Bremen would carry the cancellations back along the same route, and 60 jobs are the first thing a shop of 150 reconsiders when the order book thins. The operations executive at Southeastern Hose told Reuters he worries a bust would domino through his customers and drag down the steel mills he started out serving.

His workers sit at the far end of a chain that begins with a loan taken out by people they will never meet. That is the part of this story with no equivalent in the old argument, since a tax cut cannot be called in.

Nobody has counted both sides

Settling any of this would mean adding up what the build-out is worth and subtracting what it costs, and that number does not exist. The same Federal Reserve note explains that measuring the impact of AI investment on growth is a genuine problem. The national accounts have no line for it, and much of the equipment is bought from abroad.

Fortune reported in February 2026 on a finding that changes the shape of everything above it. Without AI spending, American corporate investment in equipment would have been negative. Negative investment means the rest of American business was letting its machinery wear out faster than it replaced it. Every other sector was shrinking its equipment stock, and AI spending alone kept the national figure above zero.

What is left is an asymmetry between what can be counted and what cannot. The gains have names and numbers attached, from Generac's $1.6B of booked orders to the 60 people Southeastern Hose has hired. The costs appear on a power bill that lists kilowatt-hours and a rate, but says nothing about which machines drove the price.

Thursday Poll

🔨 Suppliers get the jobs, host towns get the power bill. What's the fair fix?

Login or Subscribe to participate in polls.

Hire smarter with Athyna, save up to 70% on salary costs.

Athyna connects you with top LATAM AI talent, fast!

*This is sponsored content

3 Things Worth Trying

  • EIA Electricity Data Browser: State-by-state power prices going back decades, so you can see whether your own rate moved before the announcement did.

  • Good Jobs First Subsidy Tracker: Searchable database of local tax breaks and incentives, useful for checking what your county actually handed over for those jobs.

  • Data Center Map: Global directory of facilities by region, handy for counting how many sit inside the grid that bills you.

Quick Bits, No Fluff

  • Mosseri takes the stand: The Instagram head denied hiding safety findings from the public as 29 states argue Meta engineered its apps to hook children.

  • China recalls 4.3M cars: Nine automakers, including Tesla and Xiaomi, will fix hard-to-find emergency door releases, the largest recall yet in the world's biggest car market.

  • Australia bars AI-made songs: ARIA will exclude wholly AI-generated tracks from its charts after a Madonna cover with AI vocals reached No. 4.

The Toolkit

  • AssemblyAI: Speech-to-text API that handles transcription, speaker detection, and audio intelligence for production apps.

  • Framer: Pro website builder with AI agents on the canvas, so teams design, write, and ship faster.

  • Tabnine: AI coding assistant that runs privately on your stack, for teams that can't send code out.

Rate This Edition

What did you think of today's email?

Login or Subscribe to participate in polls.