$250 Billion in AI Bonds, 340 New Fintech Shells, and a Market That's Starting to Gag

$31.9 Billion in Junk. Literally.

Here's the number: $31.9 billion. That's how much high-yield, sub-investment-grade AI bond paper the US market has absorbed through July 8 this year. Not equity. Bonds. Debt. The kind that has to be repaid.

Yahoo Finance reported Thursday that AI-related bond issuance has now topped $250 billion in 2026, with July alone already featuring a $25 billion package from Amazon and data centre operator QTS marketing a further $2 billion mid-month. The investment-grade side, driven by Amazon, Alphabet, Meta, Oracle, and SpaceX, hit $218 billion through July 8, obliterating the $80.5 billion total from all of 2025. The market ate it. Mostly. But signs of strain are appearing in the secondary.

New SpaceX 6.65% 30-year bonds, priced at T+175, were trading above T+200 within days. Meta's 6.30% 2056 bonds drifted to their widest spread yet, 13 basis points wide of pricing. In high-yield, the biggest prints of the year traded below par this week. The bond market's polite way of saying: we are full.

The Equity Side Is Doing Its Own Thing

Over in VC, the numbers operate at a different altitude altogether. Global venture funding hit a record $510 billion in H1 2026, per Crunchbase, surpassing the $440 billion invested across the entirety of 2025. Q1 alone was $305 billion. OpenAI and Anthropic together absorbed $217 billion of that, which is 43 cents of every venture dollar raised in the first half of the year. Two companies. Forty-three percent. Charlie Munger called this kind of concentration "the folly of following the crowd into a burning building."

Nearly 40 AI startups reached unicorn status in H1 2026, with valuations ranging from $1 billion to $41 billion, according to Pitchbook and Crunchbase data. Prometheus, co-founded by Jeff Bezos, pulled a $12 billion Series B from JPMorgan Chase and BlackRock, the largest individual funding round in the cohort, reaching a $41 billion valuation. Core Automation, founded in 2026, hit a $1 billion valuation on a $100 million seed round. The company is months old.

The maths. A $100 million seed at a $1 billion valuation implies investors expect a 10x on a business with, presumably, zero revenue and a PowerPoint. At NASDAQ's average exit multiple of roughly 30x revenue for profitable software companies, Core Automation needs to hit $33 million in annual revenue just to justify its seed-round entry price. The average SaaS startup takes four to six years to reach $10 million ARR. The arithmetic is adventurous.

What the Shell Company Data Actually Says

Here's where it gets interesting. AIBD's analysis of Companies House data for SIC code 64.20 (Holding companies, not elsewhere classified) shows just 340 new company formations in Q3 2026 to date. That's a 94.8% collapse versus the prior period.

SIC 64.20 is not a glamorous code. It's the administrative wrapper that ambitious fintech and AI-finance ventures typically reach for when building holding structures ahead of a proper product launch. When new 64.20 formations spike, it's a leading indicator: founders are building the legal scaffolding for the next wave of AI-finance startups. When formations crater 94.8% in a quarter, it means one of two things. Either the scaffolding is already built and the boom is mature. Or founders have stopped starting.

Cross-reference against the VC data and the answer looks more like the latter. More than 70% of Q2 2026 global venture capital went to AI companies, but the money is brutally concentrated. Nearly 88% of AI-related funding in 2026 went to US companies. For a British or European founder incorporating a 64.20 holding company ahead of a fintech raise, the expected-value calculation is genuinely grim. It's not that fintech is dying. Rational founders may simply have done the sums and sat down.

Debt Is the Honest Market

Equity markets can sustain a story for years. Debt markets, eventually, cannot. The debt market for AI in 2026 is now so large it's starting to have a digestive problem. Debt financing for AI-related needs is accelerating in 2026, as Yahoo Finance put it, "testing the limits of investor demand."

Nassim Taleb would recognise the structure: convex exposure on the upside goes to equity holders; the fixed obligation on the downside sits with bondholders. Hyperscalers issuing 30-year paper at 6.65% are telling you something about what they think the AI build-out actually costs over a full economic cycle. It's not cheap. And it does not stop needing to be serviced if the AI revenue cycle disappoints.

This is the dynamic absent from VC pitch decks. The equity boom and the debt boom are happening simultaneously, layered on the same underlying assets. Together AI closed an $800 million Series C at an $8.3 billion valuation in July, led by Aramco Ventures. QTS is borrowing another $2 billion to build the data centres that will host the models. Joulent, a Houston energy infrastructure firm, just secured $1.75 billion in strategic financing to power those same data centres. The whole stack is leveraged.

The Formation Tell

Back to those 340 companies. They're not nothing. They're the people who still believe. But 340 against a prior period of multiple thousands is a compression that deserves a name. Call it the sobriety index. The frothy phase of fintech-holding-company creation, where every ex-banker with an LLM API key filed a 64.20 and called a VC, is over.

What replaces it is the phase Taleb describes as the "barbell": extreme caution at the speculative end, concentrated bets on the genuine infrastructure monopolists. Nvidia at $4.84 trillion market cap, 90-92% share of the AI accelerator chip market, forward P/E near 25. Not cheap. But not vapour either.

The AI bond market crossing $250 billion is the real tell. When the asset class needs debt capital at this scale, in these maturities, at these spreads, the story is no longer "this might be big." The story is "this is a capital-intensive industrial build-out and the winners will be whoever is still standing when the debt matures."

One prediction: the secondary market for 2026-vintage AI high-yield bonds underperforms by 15-20% against investment-grade within 18 months, as bifurcation between hyperscaler infrastructure credit and speculative datacentre paper becomes impossible to ignore. Bondholders will feel this before equity holders admit it.