THE SOLVENCY LEDGER

A No-Spin Assessment of How Deep the AI Investment Bubble Has Actually Gone — And How to Prepare Without Panicking

Jul 01, 2026

COGNITIVE-LOON | Restoration of Perception Series

Solvency. From the Latin solvere — to loosen, to release, to pay off. A solvent business is one that can pay what it owes when the bill comes due. A solvent household is one that can absorb a bad month without the whole structure coming apart. It’s an old word for an old problem: can you cover what you’ve promised?

Right now, the most important economic question on the planet is not “is AI real.” It clearly is. The question is whether the money built on top of it is solvent — and what happens to the rest of us if it isn’t.

This is not a doom post. Nobody sat down and planned this. No single villain is steering it. It’s the emergent shape of a lot of individually rational decisions — a chip company wanting a customer, a customer wanting cheap capital, an investor wanting exposure to the biggest story of the decade — stacking into something none of them fully control. Understanding that matters, because the point of this piece isn’t blame. It’s preparation. Grandmother’s second instruction was do your best — and doing your best right now means seeing the shape of the thing clearly.

Layer One: Surface Reading

“AI is booming, and some jobs are being lost along the way.”

That’s the version most people are getting. It’s true, and it’s incomplete.

Layer Two: The Blind Spot

The blind spot isn’t the boom. It’s the plumbing — the fact that an enormous share of the money changing hands in AI right now is circling between the same handful of companies, dressed up each time as new revenue.

Here’s the shape of it, verified: Nvidia has committed up to $100 billion to OpenAI, tied to OpenAI buying Nvidia chips with that same money. OpenAI has separately signed roughly $300 billion in cloud capacity from Oracle, tens of billions in chips from AMD in exchange for AMD equity, and further commitments to Microsoft, Broadcom, and Amazon. Analysts now estimate OpenAI’s combined chip and cloud commitments across these vendors could exceed a trillion dollars in value. Nvidia also owns a large stake in CoreWeave, a cloud provider that in turn buys Nvidia chips and sells capacity back to OpenAI.

Money goes out, comes back dressed as revenue, and goes out again. Nvidia pays into OpenAI, OpenAI pays Microsoft for compute, Microsoft pays Nvidia for the hardware behind that compute — the circle closes. None of this is illegal or secret. It’s disclosed in earnings calls and press releases. But it means a meaningful slice of the AI economy’s apparent growth is the same dollars passing through several hands before anyone asks whether an actual paying customer at the end of the chain wants what’s being built.

The Bank for International Settlements — the institution that functions as the central bank for the world’s central banks — flagged exactly this in its June 2026 Annual Economic Report. The BIS warned that AI spending is accumulating financial vulnerabilities that could amplify a future shock and spread from markets into the wider economy. Their specific worry isn’t that AI fails technically. It’s that a growing share of the borrowing behind it happens through structures that don’t show up cleanly on the borrowing companies’ balance sheets, which means the true leverage in the system is larger than what investors can currently see.

Layer Three: Reframe

The honest reframe is this: this is not 2008. There is no equivalent yet of millions of individual subprime mortgages bundled and mis-rated. What we have instead is a much smaller number of enormous, interlocking corporate bets, financed increasingly by debt and by “shadow” financing vehicles that sit outside normal bank oversight. The BIS specifically flagged the spread of “circular financing,” where chip makers and cloud giants take equity stakes in the AI labs that then commit to buying their chips and compute, funneled increasingly through hedge funds and private credit vehicles that face lighter scrutiny than banks.

That’s a different shape of risk than 2008 — more concentrated, in fewer hands, but for that same reason potentially faster and sharper if it turns, because reliance on non-bank lending channels means a downturn could unwind into a faster crash than a traditional banking crisis. Nobody planned this shape either. It emerged because building AI infrastructure at this speed requires more capital than even the largest companies generate from operations, so debt and creative financing filled the gap. That’s not a conspiracy. It’s what happens when an industry tries to grow faster than its own cash flow allows.

A Brief Satirical Interlude: The Ministry of Circular Commerce

[Monty Python voice, please]

NARRATOR: In the grand tradition of British bureaucratic sketches, let us visit the Ministry of Circular Commerce, where a chip company, a cloud company, and an AI lab sit around a table.

CHIP COMPANY: I shall invest one hundred billion pounds in you.

AI LAB: Splendid! I shall spend it all on your chips.

CHIP COMPANY: Marvellous! That’s one hundred billion in new orders for me. My shares are up.

CLOUD COMPANY: I say, might I borrow eighteen billion to build the data centre that houses these chips?

AI LAB: Of course — and I shall sign a three-hundred-billion-pound contract to rent capacity from you over the next five years.

CLOUD COMPANY: Excellent! That’s revenue on my books for the next five years, whether or not anyone actually needs the capacity.

NARRATOR: And thus, through the sheer motion of money changing waistcoats, the entire room agreed that everyone was fabulously, unambiguously rich. Whether any of the money had left the room was, at this stage, considered an impertinent question.

[SFX: distant sound of a bond rating agency clearing its throat]

Facts, No Spin

Verified, sourced, dated. No adjectives added.

Job cuts, May 2026: US employers announced 97,006 job cuts, the highest May total since 2020, up 16% from April. Employers cited AI as the primary driver for roughly 40% of those cuts — the first month AI has outranked every other stated reason, including “market and economic conditions.” Year-to-date AI-attributed cuts hit 87,714 through five months, already ahead of the 54,836 recorded across all of 2025.
Same period, hiring: US payrolls still grew by 172,000 in May, beating forecasts. The labor market overall is not contracting — the AI-cited layoffs are concentrated heavily in technology (38,242 of May’s cuts came from tech alone).
Hyperscaler spending: The five largest hyperscalers (Amazon, Microsoft, Alphabet, Meta, Oracle) are on pace to commit more than $1 trillion combined to AI infrastructure across 2025–2026, a figure the BIS itself cites in its June report. Moody’s Ratings separately projects roughly $785 billion in 2026 hyperscaler capex, closing in on $1 trillion in 2027.
The bond dimension: Hyperscalers issued more than $100 billion in corporate bonds in 2025 to help fund this buildout. Columbia University economist Stijn Van Nieuwerburgh estimates the total AI build-out could cost roughly $8 trillion over six years.
The BIS warning, in its own framing: major historical technology booms — canals, railroads, the early internet — attracted more capital than near-term returns ultimately justified, and those episodes “ended with an eventual reversal in investment, inducing economy-wide recessions.” The BIS is not predicting this outcome. It is saying the pattern rhymes and the exposure is now large enough to matter for the whole economy.
China’s price pressure: A documented and growing list of Western companies have shifted meaningful AI workloads to cheaper Chinese open-weight models to control costs — Coinbase (to Zhipu AI’s GLM and Moonshot AI’s Kimi, cutting its AI bill roughly in half), Airbnb and Uber Eats (to Alibaba’s Qwen), and Microsoft, which disclosed it is testing a fine-tuned DeepSeek V4 as a lower-cost option inside its own Copilot Cowork product. Reuters has quoted senior Microsoft and Coinbase executives openly saying cheaper models can handle a large share of routine enterprise work.
Power, the real bottleneck: the IEA projects global data-centre electricity demand will more than double by 2030, to roughly 945 terawatt-hours — more than Japan’s entire current electricity use — with AI-optimized data centres alone more than quadrupling. In the US, data centres are expected to account for nearly half of all electricity demand growth through 2030.
The IPO signal: OpenAI has reportedly delayed its planned IPO from late 2026 to 2027, holding out for a $1 trillion valuation rather than list lower amid a shakier public market. The report knocked SoftBank’s shares down roughly 12–13% in a single session, since SoftBank has a $40 billion bridge loan tied to an OpenAI exit.

None of these facts, alone, says “the bubble is popping.” Together, they say the system has grown large, leveraged, and interconnected enough that a real disappointment in AI revenue growth would not stay contained to tech stocks.

Three Dimensions

Individual. Somewhere right now, a person in consulting, administration, or a call center is being told their role is “being restructured due to AI efficiencies.” Some of that is genuinely true. Some of it, multiple economists quoted above caution, is companies using AI as convenient cover for cuts they’d have made anyway for other reasons. Either way, the lived experience is the same: uncertainty, and the need to build financial and skills resilience faster than the news cycle demands.

Institutional. Hyperscalers are not reckless children. They are sophisticated firms making a calculated bet that being early and enormous in AI infrastructure is worth the debt load. That bet may pay off. But the BIS’s specific worry — that guarantees embedded in off-balance-sheet financing structures could trigger unexpectedly — is a technical, structural concern, not a moral judgment on any single company.

Civilizational. Forty kilometers from where I’m writing this, the Bronze Age rock carvings at Tanum have sat through roughly three thousand years of economic systems rising and dissolving — trade networks, empires, currencies, entire modes of exchange, all gone, while the stone stayed. Civilizations that survive transitions aren’t the ones that avoid speculative booms. They’re the ones that keep enough slack in the system — enough stored grain, enough diversified skill, enough community trust — to absorb the correction when it comes. That is the actual lesson of every boom-bust cycle in economic history, from railway mania to the dot-com crash to 2008: the bust is rarely optional. Whether it’s catastrophic or merely painful depends entirely on how much slack existed beforehand.

Grandmother’s Algorithm

Pay attention. Do your best. Pay it forward.

Pay attention means reading reports like the BIS’s — not because you can trade on them, but because understanding the shape of systemic risk lets you make ordinary, boring, non-speculative decisions early: an emergency fund sized for months, not weeks; a resume and skill set that isn’t a single point of failure; a household budget that doesn’t assume the current job or the current market conditions are permanent.

Do your best means resisting both the impulse to panic and the impulse to dismiss this as hype-chasing doom content. The facts above are real and dated. What they mean for your specific life depends on your specific circumstances, and no article — including this one — can tell you what to do with your savings or your career. That’s a conversation for a financial advisor or a trusted professional who knows your actual situation, not a Substack post.

Pay it forward means sharing accurate information with people who are anxious but haven’t seen the actual numbers — not to scare them, but so they can prepare calmly, the way you’d stock a pantry before a storm whose exact strength you can’t yet know.

2008 was, at its core, a crisis of hidden risk in millions of small transactions. What’s forming now is a crisis of concentrated risk in a small number of enormous transactions, some of it circular, a meaningful share of it now flagged directly by the world’s central bank for central banks. That doesn’t guarantee a crash. It means the slack matters more than usual right now — for households, and for the people making policy.

COGNITIVE-LOON

Discussion about this post

Ready for more?