Daylila

Finance News · Wednesday, 10 June 2026

01 · Briefing · what happened

The AI boom's loudest deals are stock sales. Its biggest ones are loans.

Finance News 5 min 80 sources

SpaceX and OpenAI grabbed the headlines with their IPOs. The real money funding the AI buildout is moving quietly through private credit, lease backstops, and margin loans — debt structured so the labs don't visibly carry the risk.

Key takeaways

  • The AI buildout's biggest deals this week were loans, not stock sales — Apollo and Blackstone are financing $35 billion of computing power for Anthropic, with Google quietly backstopping the payments.
  • These deals are structured so the credit risk lands on whoever is least visible: lease backstops, margin loans, and private-credit money that traces back to pensions and insurers.
  • The IPOs (SpaceX, OpenAI) get the spotlight and the upside; the debt underneath carries the obligation that has to be paid in cash on schedule, no matter what.

This week the cameras are pointed at the stock market. SpaceX is going public, with demand running about four times the shares on offer [65]. OpenAI filed its IPO paperwork. Anthropic is in the mix too. Three of the most valuable private companies in the world, all queueing to sell shares at once.

But the deals that actually pay for the AI boom are not stock sales. They are loans. And they were structured this week to keep the risk out of plain sight.

The $35 billion that isn’t called a loan

On Tuesday, Apollo and Blackstone — two of the world’s largest private-equity firms, which increasingly lend money rather than buy companies — agreed to finance a $35 billion expansion of computing power for Anthropic, the maker of the Claude AI [44]. The money buys custom chips from Broadcom and the data centers to run them. One gigawatt of capacity to start, enough electricity to power about 750,000 homes [44].

Here is the part worth slowing down on. Anthropic is not borrowing this $35 billion the ordinary way, where the loan sits on its books and the company owes it back. Instead, Anthropic leases the chips at five data centers, and Google quietly agreed to backstop the lease payments — to step in and pay if Anthropic can’t [60]. That backstop is what turns a risky lease into something lenders will treat almost like a $35 billion loan [60].

So the headline says Anthropic. The credit risk says Google. The capital comes from Apollo and Blackstone’s insurance and credit arms — which means, a few layers down, from pension funds and insurance policyholders whose money those firms manage. Three different parties, none of them obviously on the hook, all on the hook.

This is not unusual anymore. Meta did a $27 billion version with Blue Owl Capital last October to fund its biggest data center [44]. The pattern has a name in the trade — off-balance-sheet financing — and a plain-English meaning: arrange the debt so it doesn’t show up where people look.

When the quiet financing won’t close

The same day, a different deal went the other way. SoftBank, the Japanese investment giant, has been trying to borrow at least $6 billion against its stake in OpenAI — a margin loan, meaning a loan backed by shares you own rather than by the company’s earnings [57]. Those talks stalled this week, Bloomberg reported [57].

A margin loan is the cleanest example of the risk hiding in this structure. You pledge an asset whose price can fall. If it falls far enough, the lender can demand more collateral or sell the shares out from under you — at the worst possible moment, when everyone else is selling too. SoftBank may still get the loan later [57]. But a stalled $6 billion is a small signal that the lenders financing this boom are starting to look harder at what they’re being asked to hold.

You could see the same nerves in the stock market’s reaction to Super Micro, which makes the servers that go inside these data centers. The company announced $7 billion in stock-and-stock-linked financing to pay for hardware — and its shares fell 9% [29]. Selling new shares dilutes the ones already out there, so the price drops; that part is routine. The signal underneath is that even a company with $39 billion in fresh AI server orders [29] has to keep going back to the market for cash, because the orders cost more up front than they pay back.

The same money, told two ways

Watch how the same dollar gets described depending on who’s selling it. The IPOs are framed as a once-in-a-decade chance to own the future — and SpaceX’s offering is genuinely strange in a way worth flagging: only about 5% of its shares will actually trade publicly, a “float” so small that index funds may end up owning roughly 30% of it within weeks of listing [75][5]. A tiny float means a stock that swings hard on small amounts of buying, because there isn’t much of it to go around [5]. Excitement, structured.

The debt deals are framed as boring infrastructure plumbing. But the plumbing is where the leverage lives. Equity, the stock, takes the first loss if things go wrong but can’t force anyone’s hand. Debt has to be paid on schedule, in cash, or the lender acts. The louder, equity story carries the upside everyone’s chasing. The quieter, debt story carries the obligation — and it’s spread across pensions, insurers, and a web of mutual backstops between a few giant companies who keep showing up on both sides of each other’s deals [60][44].

Elsewhere: a currency defended, a cancer bet, gold holding

Away from AI, Indonesia’s central bank surprised economists with a rate hike on Tuesday, lifting its benchmark to 5.5% from 5.25% [12]. The aim wasn’t to fight inflation but to defend the rupiah, which has sunk to a record low against the dollar as foreign investors pull out of Indonesian stocks [12]. Jakarta has reportedly drained its foreign-currency reserves to a near-two-year low trying to hold the line [12]. Higher rates make a currency more attractive to hold — but they also make borrowing dearer for everyone at home. It’s the bind small economies face when global money heads for the exits.

In drug-making, Britain’s GSK agreed to buy the US cancer-treatment firm Nuvalent for $10.6 billion — its biggest deal in over a decade [18][6]. The logic is plain: GSK’s key HIV medicine loses patent protection in 2028, and the company is buying late-stage lung-cancer drugs to fill the coming gap [18]. Buying a finished pipeline is faster than growing one. It’s a wage-earner’s reminder that a lot of corporate dealmaking is just companies racing a clock that’s already ticking.

And gold held steady near recent highs as traders weighed the fragile Israel–Iran ceasefire against the risk that war keeps inflation sticky [26]. Gold pays no interest and earns nothing — people hold it precisely when they don’t trust the alternatives. That it’s holding firm, rather than falling back as tensions ease, says the market hasn’t fully relaxed.

02 · Lesson · why it matters

The risk doesn't disappear — it moves to whoever's hardest to see

When a deal is structured to keep the danger off the main page, the danger is still there. It just lands on someone who didn't agree to it and can't watch it.

A loan that nobody calls a loan

The cleanest way to understand today’s news is to ask one question about each deal: if this goes wrong, who actually pays?

Anthropic gets $35 billion of computing power. On paper it borrowed nothing — it just signed leases for chips. But Google promised to cover those lease payments if Anthropic falls short. And the cash came from Apollo and Blackstone’s credit arms, which manage other people’s money. Three names, one obligation, and the name on the headline is the one least exposed.

This is a real financial technique, not a trick — it’s how big infrastructure gets built. But it has a feature worth naming: it separates the reward from the risk. The company that gets the chips gets the upside. The risk gets parceled out to a backstopper, a set of lenders, and, a few layers down, the pensioners and policyholders whose savings those lenders invest.

The risk has to go somewhere

Here is the rule the whole thing runs on, and it’s older than finance: risk is conserved. You can move it, slice it, rename it, hide it. You cannot make it vanish.

When a structure makes risk seem to disappear, what’s really happened is that it got pushed onto someone with less information, less power, or less attention. The person who can’t see it doesn’t price it. The pensioner whose fund bought a slice of an AI data-center loan didn’t read the lease terms. They couldn’t have. That’s not their failure — it’s the design.

Watch how the same dollar gets two different costumes. SpaceX’s stock is sold as a once-in-a-decade chance, with so few shares trading that the price can lurch on small trades. The debt deals are sold as dull plumbing. But the plumbing is where the obligation lives — the part that has to be paid in cash, on schedule, or a lender takes action. The exciting story keeps the upside. The boring story keeps the bill.

Why the quiet structures are the ones to watch

There’s a reason the dangerous version is always the quiet one. Loud risk gets priced. If a company stood up and said “we’re borrowing $35 billion and we might not make it,” lenders would demand a high interest rate, and the cost would show. Structure the same risk as a lease with a backstop, and it looks safer than it is — so it gets funded cheaply, which means more of it gets funded.

That’s the trap. The harder a risk is to see, the more of it the system takes on, precisely because nobody’s charging the right price for it. You saw a small crack this week: SoftBank’s $6 billion margin loan — a loan backed by shares whose price can fall — stalled. Lenders started asking the question out loud. When the quiet financing gets harder to close, it’s usually because someone finally looked.

You’re closer to this than it looks

It’s tempting to read all this as a story about giants — Google, Apollo, SoftBank — playing a game at an altitude you’ll never reach. But the money in those private-credit funds is not the firms’ own. It’s insurance premiums and retirement savings, pooled and lent out. If you have a pension, an insurance policy, or a fund in your name, you are probably, in some small fraction, one of the lenders financing the AI buildout. You didn’t choose the data-center lease. You can’t see it on any statement. But your future is quietly attached to whether it pays off.

That’s the second, harder half of seeing the whole. The first half is noticing that risk doesn’t vanish, it travels. The second is noticing that one of the places it travels to is you — not as a spectator reading about other people’s deals, but as a node in the same web, holding a sliver of a risk you never agreed to and can’t watch.

No single seat sees the whole structure. The pensioner can’t see the lease. The lender can’t see every backstop the borrower has stacked elsewhere. Even the firms arranging these deals mostly see their own slice. That’s not a reason for alarm. It’s a reason for humility — to hold any confident story about who’s safe and who’s exposed a little more loosely, because the part of the risk that matters most is usually the part that was built not to be seen.

03 · Lab · your turn

Where Does the Risk Land?

Structure a $35 billion AI deal three ways and watch who carries the loss when it goes wrong — usually whoever's least able to see it.

Across the beats