Roko's Basilisk
Posts
Broadcom Becomes The House

Broadcom Becomes The House

Plus: Meta ships Muse, Xbox chases a billion daily users, and Zhipu weighs its own chip.

Roko's Basilisk
July 09, 2026

Here's what's on our plate today:

🧪 Broadcom quietly became the insurer of the AI debt boom.
📰 Meta's free image generator, Xbox's billion-user bet, and Zhipu's chip ambitions.
🛠️ Three things worth trying: SemiAnalysis, The Information's data center database, Vast.ai.
📊 Poll: what is Broadcom really doing here?

Let’s dive in. No floaties needed.

See what enterprise-ready AI support looks like

How are leading teams getting AI support to work? We're breaking down the playbook, live July 9.

Save your spot

What’s next is almost here.

On July 16th at 1PM ET, beehiiv is going live with a look at the future of publishing, audience growth, and digital business.

What started as a newsletter platform has evolved into something much bigger: a place where creators and brands can grow, monetize, and own their audiences without stitching together half the internet to make it work.

The next chapter starts live at the Summer Release Event.

Join us to see what’s coming next.

RSVP now.

_{*This is sponsored content}

The Laboratory

TL;DR

The AI boom found its insurer, and it's the company selling the chips.

$570B debt wave: Morgan Stanley forecasts AI-linked debt issuance nearly doubling in 2026, as hyperscalers whose capex hits $700B stop funding the buildout from cash.
Broadcom's quiet role: its new AI XPV Platform with Apollo and Blackstone targets 20 gigawatts of compute through 2028, launching with a $35B tranche for Anthropic's TPU leases.
The guarantee is the product: Broadcom's credit support turns fast-depreciating custom chips into collateral pension funds will hold, priced off its balance sheet, not a startup's.
The circularity problem: a supplier guaranteeing debt that buys its own chips blurs real demand, echoing vendor financing from past tech busts.
The stakes: Broadcom can honor its promise only while AI demand stays strong, the one scenario where nobody needs it to.

How Broadcom became the quiet insurer of the AI debt boom

For most of the past decade, artificial intelligence expanded on a simple assumption: the companies building it were so wealthy they could finance their ambitions themselves. Microsoft, Alphabet, Amazon, and Meta generated enough cash to pay for the largest infrastructure buildout in corporate history without asking anyone else's permission.

AI’s next phase is being financed on an unprecedented scale, with hundreds of billions of dollars in debt reshaping how the world's largest compute infrastructure gets built. Photo Credit: Reuters

That assumption is beginning to disappear as AI becomes a debt-financed industry, and debt always asks the same question: Who guarantees that the money will come back? Increasingly, that answer is not a bank, an insurer, or a government. It is a chip company from Palo Alto that most people outside the industry could not describe.

A forecast and a launch, one day apart

Two announcements, made one day apart in June, revealed how quickly AI financing is changing. On June 9, Broadcom announced the AI XPV Platform alongside Apollo Global Management and Blackstone's credit and insurance arm, a financing vehicle designed to enable more than 20 gigawatts of computing capacity for frontier AI labs, including Anthropic and OpenAI, through 2028. Its first transaction was a $35B tranche supporting Anthropic's expansion of more than one gigawatt of compute. Bloomberg had reported days earlier that Apollo and Blackstone were assembling the debt to purchase Google's custom TPU chips, which Anthropic will lease rather than own, with Broadcom providing credit support on the senior portions of the financing.

A day later, Reuters reported Morgan Stanley's forecast that AI-related global debt issuance would more than double to nearly $570B this year, with $236B already raised by the end of May, roughly four times the pace of 2025. The bank expects hyperscaler capital expenditure to reach $700B in 2026 and pass $1T in 2027. Companies that once relied almost entirely on their own cash flows are now among the largest borrowers in the investment-grade bond market. Together, the two announcements explain not just how much money the AI industry plans to borrow, but why lenders are increasingly willing to provide it.

What the arrangement actually does

To understand why Broadcom matters, it helps to look beyond the headline and into the deal's structure. Although Apollo and Blackstone are supplying the financing, Broadcom plays the role that ultimately gives lenders the confidence to participate. The transaction is structured around a special-purpose vehicle (a standalone legal entity created to hold a project's assets and debts), which borrows billions of dollars, purchases the chips, and then leases them to the AI lab, using the resulting lease payments to repay its creditors.

Although Anthropic is the ultimate user of the hardware, the lenders are really evaluating the transaction through two sources of security: the lease payments that will accrue over the life of the agreement and the value the chips would retain if they were ever sold. It is that second assumption that makes the financing unusually complicated, because custom AI accelerators are unlike the kinds of assets that have traditionally supported large debt markets. They are built for highly specific workloads, lose economic value as newer generations enter the market, and have yet to establish the kind of deep, liquid resale market that gives lenders confidence in more conventional forms of collateral.

Broadcom enters the picture by standing behind that residual value. If the chips are eventually resold and the proceeds fall short of expectations, the company absorbs the difference, allowing lenders to evaluate the transaction against Broadcom's balance sheet rather than Anthropic's. In practical terms, that single commitment changes the character of the financing, transforming highly specialized hardware into an asset that large institutional investors such as pension funds are willing to finance and helping make what could become one of the largest private credit transactions ever completed.

The pattern is not isolated. Blackstone separately committed $5B to a Google-backed TPU infrastructure venture in May, and Meta pioneered the adjacent template last October, placing its $27B Louisiana data center campus into a joint venture majority-owned by Blue Owl Capital, keeping the debt off its own books.

When a chipmaker starts selling certainty

That arrangement is both innovative and potentially uncomfortable, because it introduces a question that financial markets have encountered before. Broadcom is not a passive guarantor; it is the company whose chips the borrowed money buys. Its second-quarter results, published a week before the platform launch, showed AI semiconductor revenue of $10.8B, up 143% year over year, and the company guided to AI semiconductor revenue of $56B for fiscal 2026, with Tan projecting more than $100B in fiscal 2027. When a supplier guarantees the financing used to purchase its own products, the arrangement begins to resemble vendor financing, a structure that can make it more difficult to distinguish between demand created by customers' own purchasing power and demand supported by the supplier's balance sheet. Fortune drew a similar comparison in its analysis of NVIDIA's investment in OpenAI, arguing that such arrangements can blur the line between independent market demand and sales enabled by the vendor's financing.

The defense is real and should be stated plainly. Broadcom's guarantee is backed by one of the most profitable balance sheets in the industry, generating $10.3B in free cash flow in a single quarter. Chief executive Hock Tan framed the platform as a response to a structural shift, saying "the demand for AI compute is fundamentally reshaping the global economic landscape." If demand holds, the guarantee is never called, Broadcom collects strategic influence for free, and the arrangement looks less like risk-taking than like a hardware company monetizing its own credibility. The trouble is that guarantees only reveal their cost in the scenario they were designed to make everyone stop worrying about.

The machinery that makes the promise hard to test

Assessing the risk is not straightforward because Broadcom's guarantee cannot be viewed in isolation; its value depends on the assumptions underlying the entire AI financing ecosystem, from how quickly chips lose value to whether a meaningful secondary market for AI hardware ever develops. The useful life of the underlying chips is itself contested: investor Michael Burry accused hyperscalers last November of stretching depreciation schedules beyond what rapid chip cycles justify, and CNBC has documented how customers already walk away from facilities housing aging hardware. A residual value guarantee written on assets whose residual value nobody can yet observe is closer to actuarial guesswork than underwriting. Regulators have noticed the broader pattern without naming the guarantor: the IMF's April Global Financial Stability Report flagged that parts of the AI value chain have increasingly relied on circular financing, while assessing the current impact on stability as modest. Private credit, where these deals live, largely sits outside the disclosure rules that govern banks, so the true concentration of guaranteed exposure is visible mainly to the parties within it.

Who insures the insurer?

The implications extend well beyond Broadcom itself. Every period of rapid technological expansion eventually reaches a point where someone has to absorb the risk that everyone else would rather avoid, and that responsibility has traditionally belonged to banks, insurers, or governments operating within well-defined regulatory frameworks. The AI industry is beginning to construct its own version of that safety net, relying instead on the balance sheets of its most successful companies to reassure lenders that unprecedented sums of capital can be deployed with manageable risk.

For now, that arrangement appears self-reinforcing. Broadcom's willingness to guarantee billions of dollars in financing rests on the strength of the AI market, while the continued expansion of that market depends in part on financing enabled by such guarantees. As long as demand for AI infrastructure continues to grow, the guarantee is unlikely to be called upon, allowing the entire structure to function as intended. Yet that is also its central paradox: the promise only becomes meaningful in the one scenario where the assumptions supporting it begin to unravel.

Morgan Stanley's forecast of nearly $570B in AI-related debt this year will almost certainly be remembered as a measure of how aggressively the industry chose to finance its expansion. The more revealing figure, however, may never appear in a quarterly report or regulatory filing: how much of that borrowing was possible only because lenders trusted a chipmaker's balance sheet as much as the technology it helped create. AI has often been described as the defining technology of this generation, but the durability of its infrastructure may depend less on the models themselves than on the financial promises quietly supporting them. Like every guarantee, those promises will seem almost invisible until the day they are asked to perform the task for which they were written.

Hire smarter with Athyna, save up to 70% on salary costs

Athyna connects you with top LATAM AI talent, fast

Meet vetted professionals in as little as five days, without long, expensive recruiting cycles.
Save up to 70% on salary costs when hiring AI engineers, product leaders, and data scientists.
Get AI-assisted matching plus human vetting, so your shortlist is tight, and your interviews are worth it.