iiinova — where your digital self keeps evolving

Ed Zitron @edzitron.com · 29/09/2026

Per Reuters, Anthropic has $252 billion in non-cancelable compute obligations across Microsoft, Google and Amazon, and $161.2 billion in Broadcom TPU lease obligations that are "largely non-cancelable". Truly insane. $413 billion in non-cancelable contracts. www.reuters.com/business/ant...

The company said it plans to spend at least $111.1 billion with Alphabet's (GOOGL.O), opens new tab Google, $110 billion with Amazon (AMZN.O), opens new tab and $31.4 billion with Microsoft (MSFT.O), opens new tab under long-term infrastructure service obligations over the next seven to 10 years "regardless of ⁠usage."
Anthropic is separately carrying about $161.2 billion of Broadcom (AVGO.O), opens new tab-related equipment lease obligations that are largely non-cancelable, according to the prospectus. Anthropic filed for an IPO confidentially with the Securities and Exchange Commission in June but its paperwork has not been publicly disclosed.

Anthropic did not immediately respond to a request for comment about the filing.
The company said it committed to pay Google between April 2026 and July 2033 and Amazon between May 2026 and April 2036.
"If our actual spend falls short, we must pay Google the difference," the company said, adding that similar terms apply to its Amazon agreement.
It added its commitment with Microsoft of $31.4 billion between November 2026 and May 2033 "is non-cancelable except in the event of Microsoft's uncured material breach."

2743076

Ed Zitron @edzitron.com · 29/09/2026

On Friday, I'll publish a premium that explores the wider economic impact of generative AI to date and what happens when the bubble pops. To read it, sign up for a paid subscription. Prices start at $7-a-month. edzitronswheresyouredatghostio.outpost.pub/public/promo...

2806

Ed Zitron @edzitron.com · 29/09/2026

Anthropic's numbers are as bad as my worst expectations, and it's impossible to see how it manages to make good on its spending commitments. OpenAI and Anthropic are toxic companies with rotten economics, and should not be allowed to go public. www.wheresyoured.at/dead-money/

In any case, these numbers are as bad as I’ve always thought they’d be, if not a little worse. I don’t see how this company becomes one that can afford its $518 billion in compute commitments, nor do I see how it magically works its way out of the economic equivalent of septic tank.

This company will, if allowed to go public, likely lean on the very same junk-grade/high-yield debt that AI data centers and neoclouds like CoreWeave currently need, and it will do so at volumes of somewhere between $50 billion and $100 billion a year for a company with few assets, endless losses and a CEO with the grace of a drunk elephant.

Anthropic is not the future of technology, nor is it the next Google, nor is it the next Microsoft, nor is it, to quote Reuters, capable of “[transforming] the global economy more profoundly than industrialization, electricity and the internet.”

It is impossible to rationally argue that the economics of OpenAI and Anthropic make any real sense. To claim that this is “just like Uber” or “just like Amazon Web Services” or “just like the Dot Com Bubble” is to bury one’s head in the sand or, on some level, want to know less about the world. This is serious, dangerous, and should not be seen as “business as usual.”

We must treat OpenAI and Anthropic as what they are: economic disasters waiting to happen.

To do anything less is to directly invite danger to the door of every investor that’s allowed to believe that they’re funding the next industrial revolution.

319734

Ed Zitron @edzitron.com · 29/09/2026

Anthropic's leaked IPO prospectus showed that it spent $2.75 to make a dollar in 2025, making it a worse business than OpenAI. It lost $8bn on $4.6bn in revenue. Boosters will say 2025 doesn't count - or, put another way, to ignore your lying eyes. www.wheresyoured.at/dead-money/

This, shockingly, means that Anthropic was a worse business than OpenAI in 2025, when it spent $34 billion to make $13.07 billion (per my own exclusive reporting of its audited financials), or $2.60 to make $1.

While things could change in 2026, it’s important to note how many people said that Anthropic was “a better business” that would “be profitable faster than OpenAI,” which is, until we are able to see both of their audited 2026 financials, somewhere between a myth and an outright lie.

So many people told me that Anthropic was more-profitable! So many people assured me that this company had worked it all out, when in fact Dario Amodei’s horrid son was just as obese as Altman’s, a rotten, unprofitable carcass.

Boosters are already boiling their copium kegs, angrily oinking that 2026 “will be better” and that “Anthropic has been more profitable.” At this point I have less than zero interest in anything that hasn’t gone through an auditor, because it’s very clear that, through either misinforming investors or the media, Anthropic has intentionally obfuscated the full horrors of its economics.

17010

Ed Zitron @edzitron.com · 29/09/2026

The AI bubble bursts when the money runs out. That could be when AI startups and model labs can't raise like they used to, or AI data center developers balk at 11% interest on their debt. Or, perhaps, the growth just slows and the party ends. www.wheresyoured.at/dead-money/

Every single day — even on the weekends — someone asks me either how or when all of this breaks, and my answer is simple: when the money runs out.

Eventually, AI data center debt is going to become untenable for those raising it, because 11%+ rates on already-meager margins makes the maths a little impossible. Once this happens, there will be a fundamental reevaluation of the value of all AI data center debt, which may lead to a sell-off of the underlying bonds and associated debt, which will make any investor deeply entrenched in the GPU credit business extremely nervous and, in some cases, unable to exit their positions in anything short of an embarrassing fashion.

This isn’t likely to happen due to moral or ethical reasons, but as a result of creditors realizing that they’ve got way too much risk tied up in projects that regularly make the news for not getting built. At some point these projects become too risky for even the most mold-poisoned private credit fund or brainless Japanese bank to stomach, and the timeline will accelerate based on either Treasury rates or further data center developments facing cashflow or construction problems.

On the venture capital side, it’s unclear how much dry powder actually remains, how much of it could be deployed into AI startups, and whether it’ll be a case of ‘running out of money’ so much as a moment where everybody gets spooked about AI and stops investing entirely. This would be accelerated by any cashflow issues across any major AI startups, any downrounds (IE: raising at a lower valuation), or failed acquisitions, such as when Anthropic walked away from buying Decart for $6 billion earlier in September.

And really, the biggest sign is the most obvious one — the deceleration of Anthropic and OpenAI. If they aren’t going to pay those $1.3 trillion in compute bills, the jig is up for AI data center demand.

The signs are already there that something is up.

Per Irrational Analysis, Anthropic’s record-breaking “$65 billion in annualized revenue run rate” from July 2026 may have been calculated in the single-most-deceptive way I’ve ever heard a startup do so:

Last month, there was a whole kerfuffel in AI/semis/finance circles on Anthropic July ARR. Two numbers were going around. I don’t remember the numbers and frankly it does not matter. You will see.

One ARR number was the traditional “trailing 28 days * 13” number. Personally I hate this venture-capital clown metric but whatever a lot of people use this.

The traditional ARR number was bad and implied deceleration in growth. So the people massively long Anthropic came up with a new ARR number that was July 31st * 365 days.
That’s right folks. If Irrational Analysis is right, Anthropic’s revenue on July 31, 2026 was $178 million, and because the other calculation — 28 days times 13 — created a lower number, the company chose to go with something that should, at a minimum, have investors hiring lawyers and demanding real, tangible answers about how run rate is calculated. Every single reporter with any Anthropic source that can speak to run rates should be screaming at them for clarity, because this is some sub-Enron bullshit.

Even if you don’t trust that analysis, another from TickerTrends surfaced by Callum Williams of The Economist shows Anthropic’s annualized run rate plateauing since, it seems, the beginning of June, and as Williams said, if this is even broadly correct, it’s really, really bad.

Williams also another TickerTrends chart showing OpenAI’s revenue growth had continued to climb…but was showing the initial signs of a slowdown.

Neither of these companies can afford to slow down, in part because of their massive compute obligations, and in part because their massive valuations are based on them being able to pull in, at least in Anthropic’s case, between $190 billion and $200 billion in annual revenue within the next two years.

18811

Ed Zitron @edzitron.com · 29/09/2026

And for what? I estimate that since 2023, around $800bn in venture investments have gone to AI companies, with around $226bn going to Anthropic and OpenAI. Most AI companies have overinflated valuations that make an IPO or M&A impossible. www.wheresyoured.at/dead-money/

I estimate that since 2023, there’s been around $800 billion in global venture capital investment in AI companies, with at least $266 billion of that going to Anthropic and OpenAI.

Of those investments, I expect at least $300 billion of that equity to be dead money, because, for the most part, AI companies are wrappers or layers built on top of Anthropic and OpenAI’s models, holding very little IP of their own and being burdened with ever-growing opex that mostly flows to the two AI labs that constantly want to compete with their customers. These businesses are fundamentally built on reselling tokens from the large AI labs at a loss, which is why companies like Harvey and Perplexity have to raise hundreds of millions of dollars every few months.

These startups’ continued existence is entirely a function of venture capital, as all of them are deeply unprofitable. This means that before the bubble bursts, these companies will continue to sap the venture capital world of billions more dollars, all with little chance of an acquisition and a near-zero chance of an IPO considering their ugly economics. These economics are also load-bearing for OpenAI and Anthropic, representing around 80% of their revenues, meaning that once they die, the AI labs’ underlying revenues begin to decay.

This also means that these AI startups are, in general, not actually renting AI GPUs, choosing instead to rent them by proxy by using Anthropic and OpenAI’s models. Though some of them talk a big game about building or training their own models, doing so is enormously expensive with little chance of a payoff, especially given the massive advantage in compute, capital and talent held by the labs.

There really is no clean “out” for any AI startup not named Anthropic or OpenAI. Cognition, valued at $48 billion in its latest funding round, is allegedly worth nearly as much as Ford ($59 billion market cap), yet generates a mere $1 billion in ‘annualized run rate,’ which could mean anything, all while losing $800 million. Ford, by comparison, had $187.2 billion in revenue in 2025, with a net loss of $8 billion attributable in part to a massive writedown of its electric vehicle portfolio ($12.5 billion in Q4 2025 alone).

In 2025, Ford sold around 2.2 million vehicles. Cognition, by comparison, makes yet another AI coding agent.

What, exactly, does Cognition do from here? Who buys Cognition? Does it go public? How? It loses tons of money and has a commoditized product!

1613

Ed Zitron @edzitron.com · 29/09/2026

It doesn't help that Oracle's largest SVP debt deal has now entered distressed territory — which will, no doubt, set the tone for other similar debt deals, raising the cost of borrowing at a time when costs across the board are skyrocketing. www.wheresyoured.at/dead-money/

So, let’s talk about the $18 billion in debt behind Oracle’s New Mexico-based Project Jupiter data center, starting with ZeroHedge’s diagram of the structure:

Oracle borrowed $18 billion from a syndicate of financial institutions including BNP Paribas, Goldman Sachs, and two Japanese banks — MUFG and SMBC — that have been in effectively every major AI data center deal, including multiple CoreWeave debt facilities, every Stargate/OpenAI/Oracle data center, and even SoftBank’s bridge loan that it used to fund OpenAI’s 2025 funding round. Additionally, funds related to Blue Owl (who is also invested in multiple different Stargate and CoreWeave facilities) kicked in $3 billion in equity to make sure the debt actually got raised.

This kind of labyrinthine structure is how basically every off-balance-sheet and SPV-based data center debt deal is capitalized — a few billion dollars of equity investment, usually from one of a few private credit funds (EG: Blue Owl, Blackstone, BlackRock) that then raise debt from many of the same investors, something I covered at length in my Enshittifinancial Crisis piece from the end of last year. I also went into detail about the SPV structures a few months ago here.

The reason I bring all of this up is that this kind of SPV is the template for data center debt, and the associated investors are a large chunk of the capital funding it, which means that their ability to continue feeding the beast of AI data center debt is what’s holding up this industry.

And now one of their largest data center debt deals, as mentioned, has entered “distressed” status, which means that any further SPVs they’re involved in will price based on the current state of Project Jupiter, which will be priced both based on the project’s health and the current state of Oracle, which is being dragged down by the questionable health of its many, many data center debt deals, all of which are contingent on OpenAI’s ability to pay it $300 billion over five years.

This means that the price of any debt associated with AI data centers is now skyrocketing, at a time when the price of the goods that debt is buying are skyrocketing, at a time when the underlying construction needed to pay back that debt is taking forever.

1584

Ed Zitron @edzitron.com · 29/09/2026

What makes matters worse is that the more money spent, the more expensive AI infrastructure becomes, meaning that hyperscalers have to raise more debt, which in turn becomes more expensive. AI data centers are an inflationary force unlike any in history. www.wheresyoured.at/dead-money/

And a little under a month later, NVIDIA would bump its prices by more than 15%, partly as a result of memory costs, and partly because it has the entire tech industry by the balls.

So, as more AI data center debt gets issued, said debt becomes more expensive, because the larger the amount of debt any one thing takes up, the more competition it faces, and the more risk an investor carries by holding it. Once the debt is issued, it immediately flows into buying GPUs and associated hardware, slowly growing the cost of memory and hardware, all while increasing the competition for the specialist labor and materials needed to build data centers, such as spiking the cost of Copper, increasing the cost of construction by billions in the process.

In other words, the more you buy, the more you lose. The more money you raise, the more money you need. The more money you need, the more expensive that money becomes. And once you spend that money, everything you spent it on becomes more expensive, including raising more money in the future.

28012

Ed Zitron @edzitron.com · 29/09/2026

Oracle's November $25bn in 2025 bonds would be 41.5% more expensive if raised today, adding $6.89bn in extra interest. Coreweave's 2025/2026 short-dated debt would be 35.8% more, adding $1.2bn in interest. Both of them need billions more in debt in 2027. www.wheresyoured.at/dead-money/

Across the board, Oracle’s spreads between US Treasuries have effectively doubled, and its new yields range from a bad-yet-manageable 6.73% and 6.91% on its five and seven-year-dated bonds to astonishingly high 8%+ yield across anything longer than 10 years.

On a strictly cash basis, this means that Oracle’s debt would, if issued today, cost it another $6.89 billion in interest.

Last year, CoreWeave was already borrowing at ridiculously-high coupons of over 9%, but if that debt was repriced today, it would be paying at the very best rates between 11% and 13.22% — the kind of numbers you’d associate with a personal loan.

As you can see, repricing CoreWeave at today’s rates would increase its costs by 35.8%, adding $1.2 billion to the lifetime cost of the bonds for a company that already pays $640 billion a quarter in interest.

1683

Ed Zitron @edzitron.com · 29/09/2026

Despite this economic mismatch, hyperscalers are doubling down, and are expected to borrow $400bn next year, all as Treasury yields spike and debt markets sour thanks to Oracle's "Force Majeure" on its New Mexico data center. It only gets more expensive. www.wheresyoured.at/dead-money/

Per Morgan Stanley, AI-related debt issuance should be around $570 billion in 2026, with around $250 billion of that coming from hyperscalers, and the rest various different forms of high-yield debt shoved into either asset-backed securities or dodgy SPVs for AI data centers.

Things are only set to increase next year. Per Goldman Sachs, hyperscalers will fund more than a third of their AI investments with debt in 2027 — around $400 billion — with Jeff Pu of GF Securities putting the number a little higher at $419 billion, against estimated capital expenditures of around $1.14 trillion, specifically referring to Meta, Google, Amazon, Microsoft, and Oracle.

If we assume that other AI-related debt stays flat on the year, that puts us at $739 billion in AI data center debt in 2027, and if we assume growth matches hyperscaler debt issuance growth (around 67.6%), the number grows to around $939 billion in debt.

1764

Ed Zitron @edzitron.com · 29/09/2026

AI companies would need $725 billion in annual revenue for them to have both 10% operating margins and hyperscalers to have a 10% ROIC. We are not even close. None of the economics make sense for either side of the coin. www.wheresyoured.at/dead-money/

Yet the part that really worries me is about the so-called “application layer” — the companies paying the hyperscalers for AI compute — and how much revenue they’d need in totality to be able to justify that hyperscaler capex.

The answers are extremely grim. For hyperscalers to break even on their capex through 2027, their AI customers would have to make around $425 billion in annual revenue, and that’s if they had an operating margin of 10%, a number that includes training costs for OpenAI and Anthropic.

To be explicit, this chart measures how much revenue AI companies would need to have specific operating margins and for hyperscalers to have a specific ROIC. In other words, AI companies would have to make $725 billion in annual revenue to have both 10% margins and for hyperscalers to have a 10% ROIC.

For some context about how far we are from these numbers:

OpenAI estimates it will have $36 billion in revenue in 2026.
Through the first half of 2026, Anthropic had around $16.3 billion in revenue, and if we assume that it’s growing faster than OpenAI, that puts its annual revenue around $40 billion for 2026.
Cursor allegedly hit $4 billion in annualized revenue ahead of its acquisition by SpaceX, but that most decidedly does not mean $4 billion in revenue.
Perplexity is allegedly sitting at around $750 million in annualized revenue, but was at $250 million at the start of the year, making its annual revenues likely somewhere in the $350 million range.
Cognition recently hit $1 billion in annualized revenue ($83 million a month) as of September 25, 2026, but never defined what that meant. Considering that The Information had it at around $900 million a month beforehand, I think it’s likely that its revenues sit at around $300 million to $400 million.
The Information also notes that Cognition expects to burn $800 million this year.
Per The Information, OpenAI and Anthropic represent 89% of all AI startup revenues.
Even if Anthropic and OpenAI doubled their revenues and every single one of these “annualized” figures represented the true annual revenue of the companies, we’d be sitting at an embarrassing $157 billion, or roughly $268 billion short.

1757

Ed Zitron @edzitron.com · 29/09/2026

Data from Goldman Sachs shows that hyperscalers aren't even breaking even on their current and future spending from 2026 and 2027, let alone making a modest profit. Without Anthropic and OpenAI, they barely scratch $100bn in annual AI revenue. www.wheresyoured.at/dead-money/

Last week, Goldman Sachs’ Ryan Hammond got a little more specific, noting that hyperscaler capex estimates were now over $1.1 trillion in 2027.

These revised capex plans also came with a new and deeply-worrying analysis, taking the average of estimated AI capex for 2026 and 2027, and calculating how much annual AI revenue hyperscalers would need to break even on their capital expenditures for just those two years.

To just break even, hyperscalers need $308 billion in annual AI-specific revenues, and for a 10% Return On Invested Capital (calculated based on estimates of depreciation and operating expenses), they’d need $417 billion.

As discussed above, they are — including Anthropic and OpenAI — currently $124.2 billion short of break-even, or $243 billion short of break-even without their revenues, or $233.2 billion to $351 billion short for a measly 10% ROIC.

Now, keep in mind that A) the 2027 capex has yet to be spent and B) that, at least in theory, Anthropic and OpenAI will spend more next year…if hyperscalers are able to build the capacity necessary for them to do so, and they’re able to raise the money to pay them.

1734

Ed Zitron @edzitron.com · 29/09/2026

Goldman Sachs says that hyperscalers need $308bn in annual AI revenue to break even on their 2026/2027 capex. I estimate that Oracle, Microsoft, Amazon, and SpaceX have AI revenues of $183bn, of which nearly two-thirds come from Anthropic and OpenAI. www.wheresyoured.at/dead-money/

To put that in perspective, Microsoft had around $34.4 billion in AI revenue in fiscal year 2026, of which 70% was OpenAI’s compute spend. Per Barclays estimates, Amazon will have $31.6 billion in total AI revenue in 2026, 73% of which will come from OpenAI and Anthropic, and per UBS estimates, 54.3% of Google’s AI compute sales come from them too, with an undefined amount of Vertex AI model sales coming from Anthropic on top, for a total of around $65 billion in AI revenue, which sounds a little high.

Adding all those together gets us to around $131 billion in AI revenues for Google, Microsoft and Amazon, of which $82.2 billion (62.7%) are from Anthropic and OpenAI.

As of its latest quarter, SpaceX had (when you strip out Twitter’s ad revenues) around $2.194 billion in AI revenue, or $8.7 billion on an annualized basis, but I’ll bump that up to $25 billion on the year to include its full $1.25 billion a month from Anthropic and $920 million a month from Google, though I’ll add that both have 90 day outs. If we assume that Anthropic’s discounted compute for that quarter meant that it accounted for only $500 million of SpaceX’s AI revenue, this puts us at approximately $32.8 billion in AI revenue for SpaceX, with (as I believe Google will rent the compute directly to Anthropic) 79.3% of that coming from Anthropic.

While we don’t know Oracle’s actual AI revenues, it disclosed in its last quarter that its CPU and GPU revenues were at $6.5 billion for the quarter, or around $26 billion a year in revenue. Because I’m feeling nice, I’m going to say that Oracle has approximately $20 billion in annual AI revenue, but due to a lack of information it’s tough to say how much of that is OpenAI, though I’d imagine we’re looking at at least $8 billion or more given the progress of Stargate Abilene and the (as confirmed with sources) H100 and H200 GPUs currently rented to the AI lab. As a result, I think it’s fair to say at least 50% of Oracle’s AI revenues are from OpenAI.

This puts us at $183 billion in annual AI revenue for Google, Microsoft, Amazon, Oracle and SpaceX, with $118.2 billion, or at least 64.3%, coming from Anthropic and OpenAI.

1826

Ed Zitron @edzitron.com · 29/09/2026

The AI trade is speculative, depending on the survival of two unsustainable AI labs, themselves dependent on unsustainable startups for revenue. And now multiple banks have estimated that hyperscalers need $3 trillion+ in annual AI revenue to justify their capex. www.wheresyoured.at/dead-money/

I’ll admit it’s vindicating to see so many people suddenly jump on the “how much money do hyperscalers need to justify their capex?” train, even if not a single one of them bothers to give me credit. Per Callum Williams of The Economist, Google, Amazon, Meta, Microsoft, Oracle, and SpaceX will need somewhere in the region of $1.29 trillion in annual AI revenue to get a 10% return on invested capital for their capex through the end of 2027, with the amount rising to $2.87 trillion if this farce continues through 2030.

210417

Ed Zitron @edzitron.com · 29/09/2026

This customer concentration is a big problem for OpenAI and Anthropic, with the 80% of their revenues coming from 1% of customers — most of those being, themselves, unprofitable AI startups subsidizing their users' token spend. Without VC, it all tanks. www.wheresyoured.at/dead-money/

And that’s absolutely what’s happening, suggesting that the “AI boom” is more like five or six large companies (hyperscalers) feeding money to two companies (NVIDIA and Broadcom) so that they can feed money to two companies (Anthropic and OpenAI) who then feed that money back to them whenever capacity comes online. I estimate that there’s around $22 billion of global, non-Anthropic/OpenAI compute demand, and an indeterminately-large chunk of that is coming from AI startups that can only afford to pay for the compute as long as venture capital continues to fund them…

210115

Ed Zitron @edzitron.com · 29/09/2026

This all comes back to a point I've been making for a while now: We've treated NVIDIA's chip sales as reflective of big AI demand, when in reality $200bn+ of the GPUs are in storage, and those that are installed are mostly rented to Anthropic and OpenAI. www.wheresyoured.at/dead-money/

In other words, the talking point that NVIDIA’s GPU sales are proof of actual demand for AI services or, indeed, that hyperscaler growth is a result of all those capital expenditures is a complete lie. In reality, at least half of all those chip sales — and I’d add in Broadcom’s TPU sales too (see my Hater’s guide for more) — are being made years before anything actually happens with the chips, making the trillion-plus dollars spent on capex so far seem somewhere between optimistic and utterly incoherent.

Microsoft, Google, Amazon, Meta, Oracle, and far too many other companies have been hoarding hundreds of billions of dollars of AI chips that they either (to quote Microsoft CEO Satya Nadella) can’t plug in or simply want to have in supply for reasons that I find tough to imagine.

venture capitalist that “...some companies are hoarding colossal amounts in case they come to a point at which they don’t have enough chips to provide the computing capacity,” as if there’s been any shortage of NVIDIA chips, outside of the illusory one created by hyperscalers buying them years in advance.

So, we’ve got a situation where Microsoft, Google, Amazon, Meta, SpaceX, CoreWeave, and every imaginable neocloud is sitting on hundreds of billions of uninstalled GPUs (and increasingly TPUs). Whenever more capacity comes online, it’s immediately sold to OpenAI or Anthropic, who make up anywhere from 70% to 80% of all AI revenues and compute demand, creating the illusion that revenue growth is “coming from demand for AI compute” rather than said demand coming from two companies that have been fed over $217 billion in the last nine months, with the vast majority of it coming from Google, Amazon, Microsoft, and NVIDIA themselves.

As I discussed last week, If “demand is outstripping supply” because of millions of customers begging for AI compute, that’s very different to “demand outstripping supply” because two or three (including Meta) customers are taking up most or all of the capacity. Not to repeat myself, but…

Similarly, if “demand is outstripping supply” because lots of capacity is coming online and a diverse subset of customers is buying it, that’s vastly different to if capacity is coming on slowly, and the vast majority of it is being given straight to OpenAI, Anthropic, or Meta.
And that’s absolutely what’s happening, suggesting that the “AI boom” is more like five or six large companies (hyperscalers) feeding money to two companies (NVIDIA and Broadcom) so that they can feed money to two companies (Anthropic and OpenAI) who then feed that money back to them whenever capacity comes online. I estimate that there’s around $22 billion of global, non-Anthropic/OpenAI compute demand, and an indeterminately-large chunk of that is coming from AI startups that can only afford to pay for the compute as long as venture capital continues to fund them…

19716

Ed Zitron @edzitron.com · 29/09/2026

Last week, Fidelity published a report that said the AI trade had become "dead money," with GPU leases and token spending flat or down, and the FT's Bryce Elder reported research that said that 50% of the GPUs sold in 2026 and 2027 wouldn't be installed. www.wheresyoured.at/dead-money/

Last week, Fidelity Director of Global Macro Jurien Timmer said that “the [AI trade] has been dead money for more than three months,” citing that both token expenditures and GPU lease rates were all “flat to down,” citing specifically rental rates for H100 and A100 GPUs. While the counterargument might be that Blackwell GPU rental rates aren’t included, as I discussed last week, it’s questionable how many B200, B300, or other Blackwell chips are actually available for rent, as it appears that anywhere from $200 billion to $300 billion of NVIDIA’s sales since 2022 are sitting in warehouses or unplugged in data centers waiting for power.

The Financial Times’ Bryce Elder took the ball and ran with it, and found research that backed up what I’d been saying, emphasis mine:

Morgan Stanley measured the gap earlier this week by estimating the shortfall in available power, concluding that more than half of the GPU servers sold between 2026 and 2028 might not have anywhere to be plugged in.
Yet Elder makes the point, based on research from Jefferies, that there’re far more problems than simply not having enough power:

In the longer term, power availability is still the bottleneck — along with labour. And transformers. And cooling equipment. And backup generation. As Jefferies says: “The gap between planned capacity and physical execution remains the central issue.”

110219

Ed Zitron @edzitron.com · 29/09/2026

Anthropic fans don’t even try and defend the company anymore they just look at you like this

1450

Ed Zitron @edzitron.com · 29/09/2026

Based on my own reporting on OpenAI's audited financials from 2025, it appears that Anthropic was a worse business (at least in that year), spending $2.75 make $1 versus OpenAI spending $2.60 to make $1. So many people told me this company was "more profitable"! www.wheresyoured.at/exclusive-op...

OpenAl Lost $38.5 Billion In 2025
2025 — OpenAl Had $13.07 Billion In Revenue, $34 Billion In Costs and Expenses, and $20.92 Billion In Losses, with a net loss attributable to the company of $38.53 Billion
• Revenue: $13.07 billion
• Cost of Revenue: $7.5 billion
• Research and Development: $19.18 billion
• Sales and Marketing: $5.73 billion
• General and Administrative: $1.57 Billion
• Total Costs and Expenses: $34 billion
• Loss from Operations: $20.92 billion

The Al lab spent $7.33 billion on compute and infrastructure last year, a threefold surge from 2024, accounting for more than half of its
$12.65 billion in total operating expenses.

21636137

Ed Zitron @edzitron.com · 29/09/2026

Reuters got some of the IPO docs. Anthropic is a total dog of a company. Spent $12.6bn to make $4.6bn in revenue in 2025, $7.33bn of which was compute costs. Operating loss of $8bn. Losses getting worse year over year. Amazing stuff www.reuters.com/business/fin...

Summary
Companies
• Anthropic IPO could value the company at more than $2
trillion
• Anthropic lost $42 billion in 2025
• Al lab spent $7.33 billion on compute and infrastructure last year
• Operating loss widened to $8.06 billion in 2025 from $2.98 billion in 2024
• Execs, industry analysts have expressed concerns over harms of Al tech
Sept 28 (Reuters) - Anthropic is making a massive bet that Al will transform the global economy more profoundly than industrialization, electricity and the internet, according to its IPO prospectus seen by Reuters.
But the cost to get there will be staggering. Anthropic reported a net loss of $42 billion in 2025, and plans to spend $518 billion on cloud, computing and infrastructure obligations in coming year, according to the
prospectus.

17341431395

Ed Zitron @edzitron.com · 25/09/2026

Analysts from UBS, Barclays and Wells Fargo expect — by which I mean they are setting expectations — that Anthropic and OpenAI will account for at least $444 billion of hyperscaler earnings in the next three years. Without that spend, the revenue isn't there. www.wheresyoured.at/concentratio....

That spike in costs comes from Stephen Ju of UBS’ estimates, and even if you think that’s a little high, I would estimate that the $250 billion of commitments made by OpenAI alone on Microsoft Azure will likely mean Microsoft is expecting tens of billions more than $52.5 billion in FY27 and beyond.

I also need to express how much more money this is than these companies are already spending on compute.

In 2025, OpenAI spent (per my own reporting, assuming 50% of sales and marketing was compute expenses) a little over $29.5 billion on compute. Per The Information’s reporting, it spent $12.1 billion (with no affordance for sales and marketing) in the first quarter of 2026, and while we don’t know how much it spent in Q2 (when revenues grew by $1 billion quarter-over-quarter), it’s fair to assume that it’ll spend another $12 billion or so a quarter for the rest of the year, for a total of $48.4 billion, which is less than the $50 billion it said it expected to spend on compute in 2026.

Per Barclays and UBS, OpenAI is projected to spend $15 billion on AWS and $12.5 billion on Google Cloud in 2027, with Wells Fargo estimating it will spend $22.9 billion for the first two quarters of 2027 making it reasonable to assume at least $45 billion, for a total of $72.5 billion… which, even then, seems a little low based on what it’s already on track to spend in 2026.

Then you have to add in another $30 billion from Oracle’s $300 billion, five-year-long deal with OpenAI, which the Wall Street Journal reports is expected to drive $30 billion in revenue starting in 2027, though my own research found that it could be more than $50 billion or $60 billion

517331

Ed Zitron @edzitron.com · 25/09/2026

To read this post in full and to support my work, subscribe to Where's Your Ed At. Subscriptions start at $7-a-month, and you get an ad-free version of the main newsletter too, as well as full access to the massive WYEA Premium Archives. edzitronswheresyouredatghostio.outpost.pub/public/promo...

0363

Ed Zitron @edzitron.com · 25/09/2026

The insanity of the AI capex boom is that it only works (barely) during perfect conditions. Those do not exist, and I believe it's likely that a large chunk of AI capacity never gets built, and AI-related debt never gets paid back. www.wheresyoured.at/premium-the-...

The point of this two-part Hater’s Guide To AI Debt was to show you exactly how ridiculous this all is, and how wrongheaded and dangerous the surge of investment was from the very, very beginning.

As I said at the beginning, everything has been priced for perfection, but the truth is that everything requires perfection for this to work out. This is an industry that needs hundreds of billions of dollars’ worth of debt to build the most ambitious infrastructure projects of all time for a customer base predominantly made up of unprofitable, unsustainable companies, and under even the best case scenarios, said customers will have to be able to afford hundreds of millions or billions of dollars a year in cloud compute.

I honestly don’t believe the vast majority of data center capacity actually gets built, and in turn do not expect the vast majority of data center debt to get repaid.

I do not believe anybody is taking the severity of this problem seriously — I believe that upwards of 70% of all data center debt is left unpaid, and due to the continual use of non-recourse loans, it’s likely that many investors are left with whatever scrap GPUs can be sold for, because once the market is flooded, they’ll only be good for what copper and gold can be stripped from them.

I try not to be too alarmist, but I can’t see how we avoid this becoming one of the greatest and most-destructive asset bubbles in history, one that precision-targets asset managers and big banks that believed that every number would always go up, and that “all useful compute will be sold.”

Even before a crisis begins, the cost of building one of these things is increasing at such an alarming rate that even if the most-recently-funded projects get built, they face an impossible payback schedule even if they have consistent clientele.

I am genuinely shocked that so few people are discussing the brittle economics of what is meant to be the next industrial revolution. In the end I think it might end up being completely unprecedented — like if the Great Financial Crisis left thousands of ultra-expensive apartment buildings with cupboard-sized rooms under the belief that human beings were about to get dramatically smaller.

It sounds ridiculous, but in retrospect, I think that idea will seem magnitudes smarter than buying so many NVIDIA GPUs.

18717

Ed Zitron @edzitron.com · 25/09/2026

While individually these price increases aren't a massive deal, when they happen all at the same time, they meaningfully increase the cost of AI infrastructure -- and thus, how much debt needs to be raised, and where the breakeven point is. www.wheresyoured.at/premium-the-...

While, individually, an increase in the price of one of these commodities might not have much (or any) of an effect on the overall cost of building AI infrastructure, if they’re going up in price at the same time, you end up with a massive problem.

And that’s exactly what’s happening. Trans-pacific shipping costs more than three times what it did at the start of the year. Electricity is more expensive. The price of diesel is through the roof, meaning that last-mile shipping costs more, not to mention the cost of running heavy, gas-guzzling construction equipment.

Copper costs more. Sulfur, which is used to process copper and produce silicon wafers, has seen dramatic price increases. Helium has shot up in price.

And that’s without mentioning the other inflationary pressures. Labor, for example. The cost of living is going up, and so, it’s likely that the skilled workers that are building these data centers — those with skills that are scarce, and thus have a lot of bargaining power with construction firms — will demand higher wages.

Every time the cost of construction goes up, hyperscalers — who, again, I remind you, have long passed the point where they can pay for their capex through their reserves or cashflows — are forced to turn to the debt markets, borrowing more money at a time when the cost of debt is steadily ratcheting up.

With each price increase, the break-even point of an AI data center becomes further out of reach. Operators need to charge more, while ensuring maximum utilization, and from companies that (I remind you) are deeply unprofitable and cannot survive without constant inflows of capital.

1213

Ed Zitron @edzitron.com · 25/09/2026

Logistics are another pain point. Shipping stuff, whether from overseas or domestically, has shot up in price, and the rising cost of diesel means it's more expensive to run heavy construction equipment. www.wheresyoured.at/premium-the-...

Think for a minute about what a hyperscale data center actually is. It’s thousands of tons of material — from the structure to the servers — that all need to be shipped from somewhere, and that somewhere is often the opposite end of the earth.

And boy, that’s getting expensive. According to the latest data from Drewry’s World Container Index, shipping a 40ft container from Shanghai to the Port of Los Angeles costs $7,838 today. In February, that same container would have cost you around $2,200.

Put another way, in seven months, the cost of shipping stuff from China to America has more-than tripled, with the cost of sea freight now nearing its pandemic-era peak.

Things are similarly grim when you look at air freight. The cost of jet fuel is up more than 100% year-on-year — although, from what I’ve read, the actual hit to importers isn’t quite as bad, in part due to excess capacity on routes from China to the US.

Then you actually get to the point where you’re building something. A building site will have lots of heavy equipment, all of which requires fuel — often diesel — to function. A bulldozer, for example, might burn anywhere from four to 26 gallons of fuel per hour, depending on the model and the load.

A construction site will have on-site generators, excavators, and cranes, all of which burn fuel (and, again, most likely diesel).

So, let’s talk about it. In mid-2023, a gallon of diesel cost around $3.80. Today, it’s around $6.50 — or nearly twice as expensive. Truckers are now paying over $1,000 to fill their tanks.

Admittedly, the fuel used for on-site construction equipment is slightly cheaper, as operators can buy red diesel, which is exempt from federal and state fuel excises. While that means that construction firms aren’t paying as much as truckers, it does nothing to change the underlying price of the commodity, which has nearly doubled in price in three years.

So, let’s do some math. We’ll assume that the various equipment on a hyperscale data center site consume 50,000 gallons of diesel per month (which, based on my research, is almost certainly on the low end of things). We’ll say that diesel costs $6.50 per gallon, and we’ll subtract the federal tax on diesel, which amounts to 24.4 cents per gallon.

For the sake of argument, we’ll assume that this hypothetical data center project is based in Texas, which charges a flat tax of $0.20 per gallon of diesel. This means that said construction firm will be paying $6.046 per gallon.

At current rates, that means the construction firm will be paying $302,700 per month, just to keep its heavy machinery and generators running. If we assume that the project takes 24 months to complete (an optimistic guess) and that the diesel consumption remains constant throughout that period, we’re looking at a total bill of $7,264,800.

If we take the price of diesel from mid-2023 and subtract the same taxes, that same monthly bill would have been $167,800 — or $4,027,200 total.

This all sounds fairly trivial when we think of the scale of a hyperscale data center project. A few million dollars here. A few thousand dollars there. The problem is that this all adds up, and it’s not inconceivable that the collective increases in the price of shipping and fuel for on-site construction equipment will fundamentally change the amount of money that an AI data center requires.

1293

Ed Zitron @edzitron.com · 25/09/2026

The further you look across the AI infrastructure supply chain, the more you see precipitous rises in the cost of essential commodities like helium (used in semiconductor manufacturing) and sulfur (also used in chip lithography, as well as mining). www.wheresyoured.at/premium-the-...

Before I get into things, I just want to say that even though it seems like I’m going off the rails here, I’m not. We’re going deep into the AI supply chain to look at the things that are going up in price, because I believe that basic market conditions far beyond the control of hyperscalers will result in them either having to raise vastly more in debt than they already have (or plan to do), or to curtail some of their more ambitious plans.

Which brings me to sulfur (or, if you’re British, sulphur). This foul-smelling compound plays an important role in the AI supply chain, and wouldn’t you know, it’s also something that’s shot up in price by a factor of ten in recent months.

Sulfur (or, more specifically, sulfuric acid) is an essential material used in semiconductor manufacturing, where it’s used to clean and etch silicon wafers.

It’s also one of the main chemical inputs used when refining the kinds of critical metals that AI data centers can’t exist without. Metals like nickel and... uh... copper. To make one tonne of copper metal, you need around 3-3.5 tonnes of sulfuric acid.

Anyway, you’d never guess which region accounts for 24% of global sulfur and around half of all seaborne sulfur? That’s right, the Persian Gulf! And you’ll never guess what narrow stretch of water said sulfur has to pass through in order to reach global markets? The place that, thanks to Donald Trump’s ill-advised war on Iran, is now bunged up like an arsehole after a particularly heavy wine and cheese party?

That’s right! The Strait of Hormuz!

Don’t worry, we can always get sulfur from other places, right? Oh, shit. China’s banned exports of sulfuric acid to keep the stuff available for domestic users. And now Russia has too!

In short, there’s now a global shortage of a material that’s required for everything from metal production to semiconductor manufacturing (and fertilizers!), and prices have gone absolutely horseshit as a result. In 2024, you could get a ton of the stuff for $46. At one point in July, spot prices for sulfur surpassed the $1,000 watermark.

Which means that making GPUs, mining the copper used to build power transmission and server cooling systems, and, uh, farming is going to become more expensive as a result.

Helium is another essential component in semiconductor manufacturing where a huge chunk of global supply comes from the Gulf region, with 30% coming from Qatar’s Ras Laffan plant alone.

Anyway, Ras Laffan was struck by Iranian ordinance on March 2, 2026. A second attack followed on March 18, 2026, with Iran firing five ballistic missiles at the facility, of which one managed to slip past Qatari air defenses. This strike caused extensive damage that, according to Qatari officials, could take between three-to-five years to fix.

Even without that damage, you still have to ship that helium to the customer —- something that’s impossible now thanks to the Iranian blockade of the Strait of Hormuz.

Which means that helium prices are shooting through the roof — jumping between 118% and 125% in the quarter ending June 2026 alone.

Now, semiconductor production consumes a lot of helium — 21% to 24% — although that figure is misleading, considering that modern semiconductor facilities recycle much of the helium they use. One estimate suggested that the semiconductor industry only consumes between 2 and 5% of the global market for fresh helium.

In 2025, semiconductor manufacturers bought around $1bn of the stuff. That’s a lot — and regardless of the actual amount that the semiconductor industry consumes, we’re still talking about a commodity where the price nearly doubled over the course of one quarter.

3304

Ed Zitron @edzitron.com · 25/09/2026

Copper is essential for data centers. The problem is, it's not likely to get cheaper any time soon, in part because demand is outstripping supply, and mining the stuff is an expensive business. www.wheresyoured.at/premium-the-...

So, a few things worth noting:

Copper isn’t a particularly abundant resource. There’s far less of the stuff in the ground than, say, iron or aluminum.
We’ve found stuff that’s worth mining, but of the deposits that have been discovered but aren’t yet tapped, 65% of them are concentrated in five countries, two of which are Australia and the United States.
America’s large population (and higher population density compared to Australia) means that the odds of someone living near a ripe copper deposit are higher.
As a general rule, massive mining operations tend to be quite bad for property values, and so any mining projects tend to get held up with red tape and local objections.
Despite being endowed with lots of copper resources, new copper mines take forever to launch. When Rio Tinto started production at its Nuton venture (which supplies Amazon), it was the first new US copper source in more than a decade.
Westerners, however, are more than happy to allow mining to happen in other countries — especially those that are poorer and further away.

Anyway, a lot of these mines in Chile and Peru are old. Hell, a lot of copper mines in general are old. Which means that, in many cases, the best veins of ore have long been exhausted, and so they’re forced to go for the lesser, non-prime cuts of copper-infused rock.

Declining ore grades are something that the copper industry — and especially Chile’s state-owned Codelco — has been grappling with for some time now, with grades falling by 40% since 1990. And that’s a huge problem because processing lesser grades of ore requires a lot more energy.

What does it mean when ore is of lesser grade? Essentially, the ratio of copper to rock skews more towards the rock side of things. Processing copper requires the miner to pulverize the ore into small, millimeter-sized chunks, which can then be sorted and processed.

When there’s more rock than copper, it means that you’re expending more energy for less product. Copper, as a 1999 paper noted, was already the third-most energy intensive type of metal production. Chile’s Cochilo expects that, by 2034, the energy requirements of copper will increase by 20.2%, whereas the amount of actual copper produced will only increase by 8.3% in the same period.

Which means that copper — something that is essential to data centers, and that hyperscale AI data centers use an awful lot of, to the tune of multiple billion dollars — is only guaranteed to become more expensive.

2294

Ed Zitron @edzitron.com · 25/09/2026

Take copper, for example. At current prices, a gigawatt data center would have around $6.9bn of copper across its various components. In 2023, that would have been around $3.672bn. Copper has increased in price by 21% so far this year. www.wheresyoured.at/premium-the-...

According to the Copper Development Association (CDA), a hyperscale data center (a term which the CDA defines as any data center with more than 100MW of power capacity) could use as much as 50,000 tons of copper across its various components, from power transmission equipment to the stuff that prevents the hugely expensive GPUs from overheating as they process inference.

By contrast, the CDA claims a “conventional data center” (which it defines as a facility with “power demands in the range of 5 to 10 megawatts”), might use between 5,000 and 10,000 tons of copper.

If we want to keep things simple and, instead of using an example of a real-world facility, just use a hypothetical data center campus with 1GW of capacity, the amount of copper alone would contribute to $6.901 billion of the total construction costs.

It’s also worth noting that the price of copper has virtually doubled in the past three years. On September 23, 2023, a pound of copper cost $3.672 per pound, meaning that a hypothetical 1GW facility would require $3.672bn worth of copper.

Editor’s Note: If you’re short of cash and you live next to one of these gas-guzzling monstrosities, I have an awesome idea for a side-hustle.
Okay, so let’s assume that the price of copper continues its upwards trajectory and increases in price by 21% between the start of 2027 and this point next year.

That would mean that a pound of copper costs $8.35021, meaning that a 1GW facility would need $8.35 billion of the stuff— or an additional $1.45 billion compared to the current rates.

Again, we’re simply talking about the cost of the raw elemental material, and not any finished products. I told you, we’re going deep into the supply chain with this piece.

1334

Ed Zitron @edzitron.com · 25/09/2026

What if we don't have perfect conditions? Well, every time something goes up in price, a hyperscaler or neocloud has to borrow more money, and the more they borrow, the higher the interest rates go, and the more expensive the gear becomes. www.wheresyoured.at/premium-the-...

1302

Ed Zitron @edzitron.com · 25/09/2026

Even if construction goes to plan and your customer doesn't die, and you still manage to rent compute at full capacity, operational costs, debt, and depreciation prevent you from breaking even, even under perfect conditions. AI's economics are brittle. www.wheresyoured.at/premium-the-...

A source recently told me the all-in cost of an AI data center is now somewhere in the region of $40 million to $50 million a megawatt, so around $4 billion to $5 billion a year, which means we’re going to have to return to my Bastard Data Center Model from last year to illustrate the point.

At the Canopy Wave rate of $4.30-per-GPU-per-hour, that works out to roughly $17.7 million a megawatt. Factoring in 6-year-long depreciation of critical IT gear (read: GPUs and other hardware), electricity, and other operating costs, you’re left with approximately $946 million a year in costs against $1.31 billion of revenue, or around 27.89% gross margins for the first six years, growing to 74.97% every year thereafter…for six-year-old GPUs, rented at the same price.

Except, if we use the Canopy Contract, in year four that rate drops to $3.50, which drops your revenue to $10.71 million a megawatt, and your gross margin…to negative 11.34%.

But traveler, you forgot something! You can’t just build a data center. You have to borrow money to do so! At 70% loan-to-value, you’re required to put down 30% of the value of the eventual project — so $5 billion divided by 0.7, which works out to around $2.14 billion to borrow $5 billion. For the sake of this example, we’ll be putting interest at 6.5%, the loan at 10 years, and the first two years as interest-only during construction.

For the first two years, you rack up $650 million in interest payments, adding an effective $81.25 million to the annual payments in year three, for a total of $1.073 billion, giving you a gross margin (with debt) of negative 53.92%. In year two, payments improve to $1.027 billion and gross margins to negative 50.4%, and in year 3, you almost see daylight, with payments at $981 million, and gross margins at negative 46.91%.

You can do this!

Oh, shit, wait. In year four, your rate drops to $10.71 million a megawatt. While your loan payments decrease to $935 million, your gross margin drops to negative 129.25%.

To be clear, this is just an example, but said example is actually pretty optimistic. It assumes you will have consistent business for your 100MW data center, that your customers will pay on time and at a consistent rate, and that your operating expenses and debt servicing costs are stable.

That is most-assuredly not the case for many data centers or data center developers. For example, CoreWeave has multiple loans with floating rates, and at SOFR + 2.25%, that’s currently under 6.5%...until, of course, interest rates are hiked again.

Then there’s another problem: data centers are extremely expensive, and only getting more so over time.

1292

Ed Zitron @edzitron.com · 25/09/2026

Even ignoring the dire margins on compute, the biggest enemy of providers is time. Delays are an existential threat. Costs only ever go up. And your sole customer may not even exist by the time you're done. Even if they do, can they afford their bills? www.wheresyoured.at/premium-the-...

Construction Delays Cut Into Cashflows Needed To Pay Off Your Debt: Every single delay to your data center delays the cashflows you need to start paying off your debt, balloons your interest, and threatens any covenants on your loans that require you to have certain cashflows by a certain period.
I’ll add that this is particularly important to Delayed Draw Term Loans like CoreWeave’s, because the moment your data center generates any money, the Debt-Service Coverage Ratio kicks in, meaning that you have to maintain a certain amount of cash coming in or the rest of the money can’t be drawn, which you might need to finish the rest of the data center.
Construction Timelines Can Cut Into Revenue Guidance: Neocloud revenue guidance is set based on construction timelines, which means that investors set specific expectations based on relatively-volatile infrastructure milestones, and any delays mean guidance gets cut.
For example, back in Q3 2025 (per Paul Meeks of Freedom Capital Markets), CoreWeave’s construction partner Core Scientific had a delay of a single quarter on a data center delivery, and CoreWeave had to reduce its yearly revenue guidance by 3% (around $100 million).

You’re Planning Two Years In Advance, Which Means Your Customer Could Be Dead Before You Finish: If we assume those B300s are in an 8-GPU pod (around $525,000 each), that means you’ve got $533 million in guaranteed costs — and that’s before any price increases either from NVIDIA or the RAM manufacturers. Your costs are fixed, your payoff is fixed, but the future is volatile — with most AI startups being horribly-unprofitable, it’s certain that many (if not all) of them die, and if your customer, say, is a dodgy Dubai-based shell corporation, the chances are it’ll be in the grave far before you’ve installed your last GPU.

2332

Ed Zitron @edzitron.com · 25/09/2026

This complexity is a vulnerability. Data centers take years to build, which means there's plenty of opportunity for market conditions to change. As we saw with the cost of debt, things can materially change in a short amount of time. www.wheresyoured.at/premium-the-...

I’d argue that most people instinctually know this fact, will repeat it to each other, and then stop thinking about it entirely, and I really want to sit down and have a conversation about it, because it’s extremely bad.

A year-and-a-half ago, it was March 28, 2025. NVIDIA’s Blackwell GPUs had barely started shipping, and DeepSeek had thoroughly scared everybody into believing that the AI bubble might burst, which in retrospect was naive for the time (I made this mistake too).

A few days later, OpenAI would raise a $40 billion funding round, and the AI trade was back on target. The Fed had just decided to hold interest rates steady, and would go on to cut them three more times.

By September, everything was fun and exciting again, and a certain Ellison-owned tech firm would raise a whole shit ton of cash.

A few months later in November 2025, Oracle would issue $18 billion in bonds, with maturities ranging from 4.45% on the five-year-dated notes to 6.1% on the forty-year-dated. Back then, Oracle was the belle of the ball, with analysts a month previously saying they were “all a bit in shock” by its massive new revenue backlog, most of which came from OpenAI and would require building 7.1GW of data center capacity that, as I’ve established, would take years. In the month preceding, OpenAI had announced a flurry of multi-gigawatt deals, most of which didn’t exist, but the market was extremely excited to fund whatever crap was put in front of it as long as it had “AI” on the side.

By December 2025, the spreads (explained here) on Oracle’s debt were trading “like junk,” meaning that investors were buying and selling them at a price that said that if it were to issue more, it would have to be at the high yields associated with the junk bond market.

Since then, Treasury bonds have sold off and interest rates have been hiked with another due by the end of the year. Two months ago, Oracle’s credit rating was downgraded to BBB — one level above junk — by S&P Global, and the debt associated with the SPV behind its New Mexico data center for OpenAI has moved into “distressed” territory, meaning that it’s trading somewhere between 89 cents and 91 cents on the dollar

If the same bonds were issued today — with absolutely no reconsideration of Oracle’s financial condition — it would cost more than $2.18 billion in lifetime interest, or around $150 million more a year.

Yet things get worse when you factor in how the markets are currently pricing Oracle’s debt above treasuries — adding another $183 million a year in interest payments, or another $3.62 billion in interest, for a total of $333 million a year and $5.8 billion over the lifetime of the loan, otherwise known as paying around 32.6% more in interest for the same amount of money.

These calculations don’t factor in the generally-sour climate around data center financing, or that Oracle would face a potential downgrade if it raised again, or that even if it retained investment-grade credit it would still face investors that were deeply concerned about both AI and the company itself.

1313

Ed Zitron @edzitron.com · 25/09/2026

Last week, I talked about how AI debt is issued. This week, I'll talk about the function — the payoff, construction timings, and how costs can spiral far beyond what even hyperscalers can bear. And how 10 year Treasuries skyrocketing makes everything worse. www.wheresyoured.at/premium-the-...

2353

Ed Zitron @edzitron.com · 25/09/2026

Walked by here and they had saved some of you a table

Loser's Eating House
4.2 女文女女大（114）•田39 min
Bakery • $10-20
Closed • Opens 8:30 AM Fri
Saved in Nyc

2372859

Ed Zitron @edzitron.com · 24/09/2026

Larry Ellison when he needs to pay a data center lease

1044833

Ed Zitron @edzitron.com · 24/09/2026

Force Majeure is generally used as a kind of “act of god” measure for energy companies when wars or weather makes it impossible to deliver service. In this case Oracle is using it as a means of saying it shouldn’t pay in the event the data center isn’t ready. It is completely unprecedented.

Oracle Corp. is moving to shield itself from racking up expenses on a massive data center being built in New Mexico, adding a fresh wrinkle to a project beset by opposition and regulatory setbacks.
The technology giant sent the project's developer, a unit of Blue Owl Capital Inc., a notice citing force majeure, according to people familiar with the situation. Rather than trying to walk away as the site's main tenant, Oracle is attempting to put off payments should the data center dubbed Project Jupiter get derailed and fail to come online in 2028 as planned, the people said, asking not to be identified discussing private matters.

Even if the force majeure notice is intended as a precautionary step to win some wiggle room, it risks alarming lenders backing the project.
The debt tied to the development is already trading at stressed levels, below 90 cents on the dollar, according to a person with knowledge of the matter.
A group of about 20 banks provided an $18 billion loan to help fund the construction of the data center campus, one of several mega-debt deals that have helped bankroll the AI infrastructure boom.
The notice also stands to raise questions yet again about the resilience of data center leases.
Contracts often include a clause allowing customers to back out if service doesn't begin as scheduled. Earlier this year, Alphabet Inc's Google agreed to pay Elon Musk's SpaceX for computing power through mid-2029, for example, but reserved the right to terminate if it didn't gain access by a certain date.

2145956

Ed Zitron @edzitron.com · 23/09/2026

Fidelity’s Director of Global Macro saying that the AI trade has been “dead money” for three months, and that it’s inflationary, meaning (as I’ve discussed multiple times) that the more debt they raise the more expensive debt becomes and the more they build data centers the more expensive it gets

Post
...
Jurrien Timmer @TimmerFidelity
X.com
On the Al front, the trade has been dead money for more than 3 months now. The metrics I am following (token expenditures and GPU lease rates) are all flat to down. The price of memory (DRAM) seems to be the only thing that is still going up.
(1/2)
500
475
450
425
400
375
350
325
300
275
Al Metrics
Daily Data. Source: FMRCo, Bloomberg
• GS Al basket
387
359
56%
283
249
200
175
150
•Silicon Data LLM token expenditure index (50d chg)
•Silicon Data H100 GPU rental index (50d chg)
•Silicon Data A100 GPU rental index (50d chg)
Aug-25 Oct-25 Dec-25 Jan-26 Mar-26 May-26
Nov-24
Jan-25
Feb-25
Apr-25
Jun-25
Data as of 9/20/2026. Past performance is no guarantee of future results
Jul-26
70%|
60%
50%
40%
30%
20%
0%
-10%
-20%
-30%
-40%
-50%
-60%
Sep-26
A Fidelity
8:45 PM • 9/22/26 • 32K Views

Jurrien Timmer &
@TimmerFide... 2h S •
Yes, Al will change our lives (for the better l trust) and down the road it may well unleash a productivity miracle that raises the economy's non-inflationary speed limit enough to keep the world's mounting debt burden sustainable. But until then the buildout is inflationary with an unknown return for the companies who are investing trillions into compute. The demand fo capital (equity and debt) from corporates is rising at a $3.3 trillion clip.
(2/2)
$9,000
$8,000
$7,000
$6,000
$5,000
$4,000
$3,000
$2,000
$1,000
$0
-$1,000
-$2,000
$3,000
-$4,000
-$5,000
Crowding Out?
Monthly Data. Source: FMRCo, Bloomberg, Fed FoF Report.
non-financial corporate debt issuance (Sbi) equity share redemptions (buybacks + M&A)
Shading: US deb/GDP
inflation-adjusted
- total corporate net issuance (debt + equity)
housing bubble
+$7,924
bubble
Jun-20, +$6,597
+$3,462
dot.com bubble
Jun-00, +$3,233
+$3,999
Al capex
Jun-26, +$3,251
+$3,129
debt issuance + M&A
debt-equity arbitrage
+$123
-$197
debt-equity arbitrage
-$2.054
$2,175
-US Treasuries held by China
-$3.485
-US Treasuries held by Japan
—US Treasuries held by Fed's SOMA
—US Treasuries & agencies held by US banks
-$2,911
-$3,449
• $4,597
$3,602
$4,800
$2,400
$1,104
$1,200
$618
$600
1997
1999
2001
2003
2005
2007 2009
2011
2013
2015
2017
2019
2021
2023
$300
2025
2027
Data as of 9/20/2026. Past performance is no guarantee of future resuts
2029
•Fidelity
L71
50

221045207

Ed Zitron @edzitron.com · 23/09/2026

These are the reps of a man living in hell

Workout
Cable Bicep Curls
Sets: 4
Reps: 10
01
80 Ibs
15 reps
RIR
02
80 Ibs
15 reps
RIR
03
80 Ibs
10 reps
RIR
04
70 Ibs
15 reps
RIR
+ Add Set
Cable Hammer Curl
Sets: 3
Reps: 15
01
80 Ibs
15 reps
RIR

02
70 Ibs
14 reps
RIR
03
60 Ibs
15 reps
RIR
+ Add Set
+ Add Exercise
•••
RIR: 0
・・・
RIR: 0
BW
BW

13860

Ed Zitron @edzitron.com · 22/09/2026

I don't see how this ends well. At the very least, hyperscalers are going to be hit with massive depreciation and write-off penalties. Most of NVIDIA's revenues will never make the customer any money. The longer this continues, the greater the pain. www.wheresyoured.at/wherere-all-...

At the very least, hyperscalers are going to be burdened with brutal depreciation charges or onerous write-offs for years to come, whether their capacity turns into revenue or not. CoreWeave, Nscale, Lambda and every other neocloud is set on the highway to Hell, with ballooning debt that can only be paid via contracts that are dependent on a few AI labs and a company so capricious that it renamed itself after the Metaverse, burned $77 billion, then killed it two years later.

I don’t even know how to write what I’m thinking without sounding alarmist…but I don’t see how 90%+ of NVIDIA’s sales ever end up generating a single dollar of revenue, and considering the amount of project financing-backed data center debt deals, there’s very little that exists to protect investors if AI compute demand never arrives. I don’t know how we don’t see tens of billions of dollars of write-downs and dead data center debt deals with every investor involved losing every penny, nor do I see how big tech avoids admitting that they wasted all their capex.

I think everybody who invests in these things ultimately loses, ranging from embarrassment and terrible earnings for hyperscalers to genuine destruction for anyone that trusted the pablum that “all useful compute will be used.” Until that happens, more and more money will be sunk into further theoretical capacity, making the eventual collapse all the more gruesome.

None of this ever had anything to do with AI, and everybody who cheered Jensen Huang’s ascent in the belief it did is a mark.

What a fucking waste. I don’t enjoy finding this stuff out. I wish we’d have stopped doing this years ago.

Not that I think we will…but even if they bail out Anthropic, even if they bail out OpenAI, there is no way to magic up the trillions needed to justify the capex, or to prop up hyperscaler growth long term.

The longer this continues, the more promises are made, the more projects that are announced…the worse it’s going to be.

511816

Ed Zitron @edzitron.com · 22/09/2026

NVIDIA's (and Broadcom's) revenues are built on a speculative scramble for assets years before they're put in service, and not any actual demand for AI compute. Eventually, that compute will come online. It will be disastrous for NVIDIA and Broadcom. www.wheresyoured.at/wherere-all-...

So much money has been spent building and buying silicon for AI capacity that takes years to build, all as the AI industry tells us that right now there’s insatiable demand and that we’re fools to question it.

One of the core reasons that people believe that AI isn’t a bubble is because of NVIDIA’s perpetual quarterly revenue growth, which is branded, once again, as insatiable demand for AI compute, when it’s actually almost entirely-speculative purchases based on potential revenues, with said potential mostly driven by the compute spend from OpenAI and Anthropic, two unprofitable and unsustainable AI labs.

This is one of the reasons that Jensen Huang continues to funnel endless billions of dollars into circular financing — because the sense of ever-expanding demand for GPUs has become a proxy for ever-expanding demand for AI compute, even though it takes years for the first part to become the second, if it ever does.

NVIDIA has now sold at least $200 billion dollars’ worth of GPUs — multiple gigawatts-worth — that have yet to be ingested by the market, and hyperscalers have, through their obfuscation of operational capacity and refusal to disclose AI revenues, helped create one of the largest speculative asset bubbles in history.

Everybody who participated in this obfuscation owns part of what comes next.

Capacity will, eventually, come online at a scale that the market for AI compute cannot support, and it won’t be obvious until it’s way, way too late. I fear that every single model around existing and future data center construction and AI compute demand is wrong, and that every assumption we have about the underlying economics of AI is corrupted by the belief that there’s far more operational capacity than there really is.

If we believe there’s gigawatts’ worth of AI compute coming online every year, then we in turn believe there’s gigawatts’ worth of demand.

If there’s a gigawatt or two coming online every year, that’s a completely different story.

2624

Ed Zitron @edzitron.com · 22/09/2026

As I've said, more than half of NVIDIA's GPUs - and I'd argue most of the Blackwell GPUs sold since 2025 - are unproductive and being warehoused. NVIDIA and hyperscalers should be forced by regulators to disclose how many chips are actually in service. www.wheresyoured.at/wherere-all-...

I need to be more blunt here: the vast majority of companies that have bought GPUs have yet to turn them into meaningful revenue, if they’ve turned them on at all. Most of NVIDIA’s sales are sitting in warehouses, and that is a significant disclosure that NVIDIA should have already been forced to make.

It wouldn’t be too dissimilar to the last time NVIDIA got in trouble with the SEC back in 2022, when it failed to disclose that the revenue growth in its gaming segment was actually coming from cryptocurrency miners rather than gamers, which was considered “inadequate disclosure.”

While there’s a noted difference here — as NVIDIA’s customers are, ostensibly, buying their data center GPUs to put in a data center — the “demand” cycle for these chips, or really any AI chip, is entirely manufactured as a result of four or five large customers buying so many and Huang making statements about revenue generation that do not reflect reality.

Perhaps this doesn’t rise to the level of SEC action, but every single journalist and analyst should be asking Jensen Huang and every hyperscaler executive buying GPUs the following questions:

1635

Ed Zitron @edzitron.com · 22/09/2026

We are without question in an AI data center overbuild situation, and any and all new data center builds are a gamble that we'll have double or triple the demand we have today in 2028, 2029 or 2030. And right now, demand is much smaller than people think. www.wheresyoured.at/wherere-all-...

Right now, with more than half of NVIDIA GPUs yet to be turned into operational capacity, every single new data center being built is effectively a bet on whether AI demand is larger in 2028 or 2029 than it is today, because you’re going to be competing with all the other capacity coming online in the years preceding that have already broken ground.

Then there’s the problem of the upcoming flood of Blackwell GPUs, the vast majority of which have yet to be operationalized, meaning that anyone who bought them in 2025 is likely going to see them installed just as the first units of Vera Rubin come online, which will suppress prices even without there being significant available capacity, on top of the fact that there’s going to be a huge flood of them coming online in the next few years.

Honestly, I think it’s kind of laughable we’re even talking about Vera Rubin at this point. When are we going to see it at scale? 2030? C’mon now.

2946

Ed Zitron @edzitron.com · 22/09/2026

The reason that we have such 'compute scarcity' is that it's taking a great deal of time to come online, and the vast majority is sold to Anthropic and OpenAI, pumping backlogs across the industry and creating an illusion of demand fueled by VC and debt. www.wheresyoured.at/wherere-all-...

The massive backlogs across Google, Microsoft, Amazon, CoreWeave, IREN, Nebius, and Nscale come not from the incredible demand for AI compute but the incredible ability for Anthropic and OpenAI to sign contracts. For example, Nscale’s $45 billion deal with Anthropic along with a contract with Microsoft make up 85% of its $103 billion backlog, and Anthropic’s deal is contingent on yet-to-be-raised financing. These backlogs are regularly used to justify the massive AI data center buildout, when they’re more a function of Dario Amodei and Sam Altman’s DocuSign accounts.

You see, the ultimate problem is that the world outside of hyperscalers is — as a result of the obfuscation of data center capacity — under the belief that these companies are buying GPUs and then quickly turning that into cash versus buying these GPUs and quickly turning them into storage.

This, by the way, is the problem with hyperscalers not explicitly breaking out their AI revenue, because in doing so they create the (I’d argue deliberate) illusion that AI capex is creating revenue growth, which both tricks investors into buying their stock and tricks developers into building AI data centers, believing that capex quickly translates into revenue.

You can scoff about how investors or developers should “do better research” or “learn about stuff,” but remember that the vast majority of data points about data center construction are somewhere between misleading and outright fantasy. I’ve seen estimates of 12GW, 15GW, and as much as 20GW of capacity coming online in 2026, but based on everything I’ve talked about today, I think it’s farcical to believe that more than five to ten gigawatts of operational AI data center capacity actually exists.

2743

Ed Zitron @edzitron.com · 22/09/2026

The myth that hyperscalers sell is that "demand outstrips capacity" because of diverse customers - when in reality the majority of capex ($791bn) is unproductive, the "scarcity" comes from a majority of capacity being sold to Anthropic and OpenAI. www.wheresyoured.at/wherere-all-...

This situation is utterly obscene.

It’s very clear that at least $200 billion — if not more than $300 billion — of NVIDIA’s GPU sales have been made a year or years in advance, just as the company telegraphs it will make over $670 billion in revenue in its fiscal year 2028 (starting February 2027).

It’s also clear that Microsoft, Google, Amazon, Meta, SpaceX, and every neocloud are purchasing NVIDIA GPUs tens of billions at a time under the implicit knowledge that it will take years to build the capacity and connect the power to them, creating what amounts to the largest pre-order campaign in the history of capitalism, but also a material misrepresentation of the current AI buildout.

This leaves us with around $791 billion of capital expenditures unaccounted for, which is fairly disastrous, and if we assume that 50% of that is GPUs (across NVIDIA, AMD, Trainium, TPUs and any other custom silicon), that’s around $395 billion of silicon that’s been sold and is, I hope, sitting in a warehouse or an unpowered data center, as if they were just sold on paper, that’s…questionably legal accounting. I’m willing to believe that some share of that is also CPU infrastructure, storage, and other dollars not flowing directly to NVIDIA.

In any case, that’s a shit ton of undeployed silicon, and a very, very, very different picture to the one that both NVIDIA and the hyperscalers have been telling investors.

There’s a world of difference between “we’re buying a lot of GPUs and building a lot of data centers to make a lot of money” and “we’re investing in this stuff on the off chance it makes us money years in the future.”

I’ll break it down:

If investors and the general public believe Microsoft, Google, Amazon and Meta are bringing capacity online rapidly, capital expenditures are justified at their current rate, because it’s seen as spending money to make money.
If the truth is that the vast majority of these capital expenditures are going into Jensen Huang’s pocket and filling warehouses full of GPUs, that means that investors are being sold a line of shit about both revenue growth.

2774