Matt Scherer

openmarketsinstitute.orgReport - Brace for Impact: Protecting People and the Economy when the AI Bubble Burst — Open Markets InstituteOur report in partnership with Americans for Financial Reform outlines vulnerabilities in the financial system and wider economy that could intensify an AI crash into the next financial crisis, and pr...

What would happen if you took a stock market bubble bigger than the dot-com bubble and combined it with a pile of risky debt bigger than the one that caused the 2008 financial crisis? You'd get something that looks a lot like the AI bubble we have today. www.openmarketsinstitute.org/publications...

Anthropic's IPO filing yesterday revealed at least* $550bn in off-balance sheet debt, for a company with $16B in annual revenue. Paying its bills will require Anthropic to sustain implausible revenue growth given the pressures that are squeezing their--and the whole AI ecosystem's--margins. A 🧵

podcasts.apple.comBrace for Impact: Protecting people and the economy when the AI bubble burstsTechnology Podcast · Series · This report from Open Markets Institute Fellow Matthew Scherer and Americans for Financial Reform Education Fund Senior Policy Analyst Maya Jenkins, titled “Brace for Imp...

Or Apple Podcasts here: podcasts.apple.com/us/podcast/b...

53m

Ok, one more thing. I made a little thread showing the total amount of Anthropic's future spending commitments--which, again, is basically debt by another name. bsky.app/profile/matt...

Matt Scherer@matthewus.bsky.social

I wanted to create a little reference guide showing all of Anthropic's off-balance-sheet obligations (basically, debt by another name). Figured I would share it. Yes, I know, it's a screenshot of an Excel spreadsheet, but at least you know I didn't use AI. This doesn't include everything, even...

wsj.comInside the 11th-Hour Scramble to Save Anthropic’s Debt DealPlus, Microsoft and Skydance turn to voluntary exits to shrink their workforces, and Trump’s tariff exposes America’s reliance on Canadian aluminum.

It doesn't include this $63 billion debt deal signed as they were filing their S-1.

No way to know for sure, but my guess is that the total amount of Anthropic's total future commitments at the very least approaches $1 trillion and may well exceed it.

wsj.comExclusive | Broadcom CEO Intervened to Seal $63 Billion Anthropic Debt DealThe deal, announced on Monday, is the biggest ever to raise money for AI infrastructure.

There is also an undisclosed amount of future lease commitments. Throughout the S-1 the aggregate amounts are left blank. That's not necessarily nefarious (they probably were still negotiating a lot of this and didn't want to lose bargaining leverage), but it means there's stuff left out. And...

Matt Scherer reposted

John Kostyack

@kostyack.bsky.social

Replying to @kostyack.bsky.social

Wall Street scrambled over the weekend to put together an $83B debt package for Anthropic before its S-1 was released. Note the sudden increase in interest rates charged for the $40B in bonds headed into the public markets, and the whopping $23B sent into the opaque private credit markets.

Matt Scherer reposted

John Kostyack

@kostyack.bsky.social

Debt finance is really hard to understand for those not well-versed in financial markets, so here's a helpful explainer. A key takeaway: investors in debt are seeing big problems with Big Tech's thesis on AI and charging more to cover its risk-taking.

Matt Scherer@matthewus.bsky.social

openmarketsinstitute.orgReport - Brace for Impact: Protecting People and the Economy When the AI Bubble Bursts — Open Markets InstituteOur report in partnership with Americans for Financial Reform outlines vulnerabilities in the financial system and wider economy that could intensify an AI crash into the next financial crisis, and pr...

Read our full report here: www.openmarketsinstitute.org/publications...

open.spotify.comBrace for Impact: Protecting people and the economy when the AI bubble burstsPodcast · Open Markets Institute · This report from Open Markets Institute Fellow Matthew Scherer and Americans for Financial Reform Senior Policy Analyst Maya Jenkins, titled “Brace for Impact: Prote...

Or listen to it on Spotify here: open.spotify.com/show/3cG0PKT...

These brutal economics are not likely to improve. Even if demand picks up significantly, competition will continue to be intense. Those forces will apply unrelenting downward pressure on prices even as high costs and resource constraints squeeze companies from the opposite direction.

That is a deadly mix for Anthropic and other AI companies that have taken on massive fixed future spending commitments. With costs continuing to accumulate far faster than non-tech-company spending on genAI, a financial reckoning for Anthropic--and the whole AI industry--looks increasingly likely.

Relatedly, there are external factors that are also delaying data center construction and driving up the cost of training and running AI models. Community opposition is a major factor, but so are logistical challenges of getting grid capacity up and connected, supply chain constraints, etc.

Every delay limits the computational capacity Anthropic is counting on to generate the revenue it will desperately need. And on the aggregate every extra dollar of unexpected expense gets added to the AI ecosystem's already mind-boggling multi-trillion dollar tab.

Demand and commoditization make it hard for model developers to raise prices. But the third force--the extraordinarily high costs of training and running AI models mean that they can't afford to keep prices low either.

bloomberg.comWhy AI Data Centers Cost Billions to BuildA breakdown of the costs to construct a single data center in Texas reveals why developers face tens of billions in expenses.

A recent Bloomberg story broke down how and why it is so expensive just to BUILD a data center--$31 billion to build a 1 GW facility. And that's not counting the also-very-high-cost of electricity and water to power and cool everything once the data center is online. The costs never end.

Turning to the second source of margin pressure: Anthropic is just one of many well-capitalized companies (along with OpenAI, Google, Meta, Microsoft, SpaceX, etc) that all offer models that are largely indistinguishable in their quality and capabilities, a process economists call commoditization.

ft.comAnthropic’s best AI model struggles to attract users as cheaper tools thriveAI lab’s Fable 5 has met with sluggish demand from corporate clients

Then there's competition Chinese AI companies. And the fact that people can do 95% of whatever useful things LLMs can do with small models that they can run locally on their phone or laptop. It's hardly surprising that few companies seem interested in paying for "frontier" models.

apollo.comNo Signs of AI Payoff Outside the Tech Sector | The Daily SparkDespite massive investments and widespread hype around artificial intelligence, the technology has delivered zero financial payoff to corporate America outside of tech. Subscribe for daily updates.

That's consistent with studies repeatedly showing that most non-tech companies are not seeing any ROI on AI investments. My report with @m-m-j.bsky.social (link below) cites several. Here's another (from Apollo, no less) released after our report went to press. Now I know what you're thinking...

hardresetmedia.comThe AI Productivity IllusionProductivity Is About (Much) More than Finishing Tasks Quickly

"But people do stuff in hours with AI that used to take weeks! How can you say it doesn't increase productivity?!?" Answer: true productivity is about creating economic value, not simply getting tasks done quickly. AI is good at the latter but not the former (as I wrote for @hardresetmedia.com ).

cepr.netGDP Growth in Second Quarter Slows to 1.5%, No Evidence of AI Productivity BoomSecond quarter GDP grew 1.5 percent as AI investment supported growth, but productivity remained weak and inflation stayed elevated.

The guiding assumption of the AI boom is that generative AI will massively increase productivity, particularly by automating tasks typically performed by white-collar workers. But AI had no discernible impact on productivity in 2025 or the first half of 2026.

Side note: The Stanford Digital Economy Lab loves to use this chart to argue AI is taking entry-level coder jobs. It actually shows the opposite--job losses started in summer '22 (blue arrow), before ChatGPT was released, and stabilized in spring '25 as AI coding tools arrived (red arrow).

We originally whipped up this figure to illustrate the forces squeezing the AI industry as a whole (which, because of its circular financing and mutually dependent relationships, can fairly be considered as a single unified whole). But they just as easily could be about Anthropic in particular.

First, it's simply not clear that there is trillions of dollars of demand for AI services. The pie of revenue may not be big enough to feed everyone. We keep getting told that AI is a "general-purpose technology" that will trigger a "Fourth Industrial Revolution," etc. Yeah, about that...

* $550bn is just the non-cancelable commitments to Google, Amazon, Microsoft, and Broadcom. Anthropic has 100s of billions more in conditional or cancelable commitments, much, if not most, of which will likely end up on their balance sheet. And that's just the stuff they chose to disclose. It's bad.

I wrote last week about how much debt there is in the AI ecosystem and how much of that is off AI companies' balance sheets. The upshot of that is that they will need to increase revenues at an unreal pace to pay their fast-mounting bills. That looks increasingly unlikely. bsky.app/profile/matt...

Matt Scherer@matthewus.bsky.social

Just 9 tech giants have racked up $3.7 trillion dollars in debt, most of it hidden off their balance sheets. The hidden debt alone ($3.1 trillion) is significantly bigger than the amount of subprime mortgage debt outstanding at its pre-2008 peak.

Matt Scherer reposted

Carl T. Bergstrom

@carlbergstrom.com

18h

"AI can draft an essay for you, especially if you give it a high quality prompt." Really? This is what MIT thinks high schoolers should learn? MIT's "Day of AI" dayofai.org provides materials for K-12 teachers and claims to have reached over 5,000,000 students to date. 1/2

16h

I don’t know which is more WSJ, but this was from earlier this month

kate conger@kateconger.com

wsj is wsj'ing at unprecedented levels

18h

One more time for those in the back.

Elizabeth Warren@warren.senate.gov

Before the 2008 crash, Wall Street hid massive financial risks using accounting tricks. Now Big Tech appears to be following the same playbook. Americans and our entire economy are at risk because Trump’s regulators are choosing to look the other way.

Matt Scherer reposted

Elizabeth Warren

@warren.senate.gov

ourfinancialsecurity.org/resources/br...

Replying to @warren.senate.gov

ourfinancialsecurity.orgBrace for Impact: Protecting People and the Economy When the AI Bubble Bursts - Americans for Financial ReformAn economic crisis from the AI bubble is not inevitable. While the ingredients for a serious crisis are in place, inaction regarding the growing risks would be a dangerous mistake, and policymakers mu...

Matt Scherer reposted

Futurism

@futurism.com

19h

Flávio Bolsonaro was the overwhelming favorite, right up until the moment he unceremoniously lost.

trib.al"Prediction" Markets Were Comically Wrong About the Brazilian ElectionThe prediction market Polymarket had Flávio Bolsonaro as an overwhelming favorite to win over Lula in the Brazil presidential election.

Matt Scherer reposted

Adam Becker

@adambecker.bsky.social

Replying to @adambecker.bsky.social

This, right here, is so deeply confusing to me. How on earth is this legal? bsky.app/profile/matt...

Matt Scherer@matthewus.bsky.social

From my recent report with @m-m-j.bsky.social www.openmarketsinstitute.org/publications...

Matt Scherer reposted

Adam Becker

@adambecker.bsky.social

hardresetmedia.comBlock the IPOWe Can’t Trust Anthropic’s Financial Disclosures. That Means It Shouldn’t Be Allowed to Go Public.

Replying to @adambecker.bsky.social

And those financial statements are dubious too, as @matthewus.bsky.social points out here: www.hardresetmedia.com/p/block-the-...

Matt Scherer reposted

Adam Becker

@adambecker.bsky.social

trib.alNew Analysis Finds Signs the AI Bubble Is About to BlowA new report by OpenMarkets finds that the AI bubble is "unique in recent history in its size, scope, and risk."

Don't believe the hype. Anthropic's IPO filings show it's not in good shape. You just have to dig a little deeper.

Matt Scherer@matthewus.bsky.social

Except, (1) the only reason they are cash-flow positive is that they are getting discounts on virtually all inputs from their Big Tech investors; (2) their revenues stopped growing in Q3; and (3) they have taken on $800B+ in future contractual commitments (50x their now-stagnant operating revenue).

Matt Scherer reposted

Futurism

@futurism.com

22h

"The odds of a deep and severe AI crash are rapidly increasing."

Matt Scherer reposted

Carl T. Bergstrom

@carlbergstrom.com

Your monthly reminder that if you're using ChatGPT to do your science, you're failing your colleagues, community, and all of us. Here, ChatGPT makes up data and presents it as the real thing — then apologizes when confronted.

When the Wall Street Journal accidentally produces the Democrats’ best-ever campaign ad.

nytimes.comPut Your A.I. Agents to Work with These 5 Killer TasksWhen you start using Meta’s Muse and OpenAI’s dots for these tasks — like applying for jobs and booking hard-to-get restaurant tables — you may never go back.

Jfc the AI boosters who pass for journalists at the Gray Lady are getting desperate. Here, Brian Chen talks about 5 "killer tasks" for AI, all of which show more that he needs to examine his priorities rather than that he's actually discovered useful things to do with AI.

And for #5, again--how is this a killer app? You save yourself like 30 seconds vs simply setting a reminder in your phone to check prices once a week? This is worth $100/month?

qz.comBain: AI needs $6 trillion in annual revenue to justify data centersExisting AI services could account for $1.8 trillion of that target, leaving $4.2 trillion in new revenue still to be created

And again--this is supposed to be a list of "killer tasks" for a technology that needs to earn $6 trillion a year to be worth what companies are spending on its infrastructure. qz.com/bain-ai-reve...

For #3, Brian just shows that he's a crappy friend. Parents fight over childcare like wolves fight over fresh meat. If he'd just spent 10 minutes helping her set up Facebook and Care profiles, she would've gotten 50 responses in a day.

For #4 -- see #2, brah. If you care about your kids' nutrition, maybe don't outsource your meal planning to hallucination machines.

Like right off the top...how many people can afford a table at an overbooked restaurant? And for those 1% of the 1%, they can't just pick up the damn phone, call the restaurant, and ask to be put on a waiting list?