The Quiet Trillion
Sovereign Stack | Money in the AI Ecosystem, Part 2: Debt Markets
By Oluwaseyi Ayodeji, Published on oluwaseyiayodeji.com | Sovereign Stack Newsletter
I am not a debt markets expert.
Two weeks ago I couldn't have told you what a special purpose vehicle was, or explained how a company can borrow tens of billions of dollars and never show it on its own balance sheet. I read Federal Reserve research notes, litigation risk memos, and bond market columns just to write this piece with any confidence. There are still questions in here I haven't fully answered. I'm publishing anyway, because where AI's real money sits matters too much to wait until I feel like an expert.
Here's what I found.
The trillion nobody's watching
Last week, in The 43 Percent Problem, I wrote about the venture money: two companies absorbing nearly half of all global startup funding in the first half of 2026. That's the loud money. It funds the labs and the headlines.
This week is about the quiet money, the second of three streams in this series. It's bigger than the venture money, and next week closes the loop with the third: the state's own money, and how China finances its AI buildout by simply deciding to, without needing a bond market's permission first. Keep that comparison in mind as you read this one, because debt markets and state capital solve the same problem, funding a trillion-dollar buildout, in almost opposite ways. Global AI-related debt issuance is projected to hit roughly $570 billion in 2026, nearly four times the prior year's pace [1][4]. Outstanding private credit tied to AI infrastructure had already crossed $200 billion by early this year, with another $800 billion expected over the next two [1]. Hyperscaler capital spending alone is expected to top $700 billion this year and cross $1 trillion by 2027 [4], and Goldman Sachs projects total AI and data center investment could reach $5.3 trillion by 2030 [1][5].
Meta's joint venture with Blue Owl Capital moved roughly $27 billion of data center financing into a separate vehicle [1][3]. Oracle, Microsoft, and Amazon have leaned on similar structures [3][4]. Life insurers now hold close to $1 trillion in private credit exposure tied to this buildout, drawn in because the long, contracted cash flows from data centers match their own long-dated liabilities [3]. Public pension plans in states like New York and Pennsylvania are showing up in the same vehicles, meaning ordinary retirement savings are now indirectly financing AI infrastructure most of those retirees have never heard of [3].
That quiet is deliberate. Debt markets aren't built for spectacle. They're built for discipline, and discipline doesn't trend on social media.
The question that bothered me most
Here's what stopped me while researching this: how does a public company legally borrow tens of billions of dollars and keep it off its own books?
The short answer is control. Standard accounting rules only require a company to consolidate, meaning report as its own, the debt of an entity it controls, typically by owning more than half of it. Hyperscalers have learned to structure data center financing through joint ventures where they hold a minority equity stake, often selling most of the entity to a private credit fund, while still locking in long-term contracts that guarantee the data center's output comes to them [1][3]. The debt sits on the joint venture's books. The hyperscaler gets the capacity it needs to train and run its models. Legally, none of it touches their own balance sheet, and their reported debt-to-earnings ratios stay clean even as the underlying buildout scales into the hundreds of billions.
Is that disingenuous? I went in expecting an answer and came out with more of an open question. Financial reporting standards exist to give investors a true picture of a company's obligations. Structures built specifically to sit just under a consolidation threshold satisfy the letter of that goal while testing its spirit. Regulators seem to be noticing too. The Federal Reserve has flagged that banks' direct exposure to AI-adjacent lending looks small, but their indirect exposure, through loans to the private credit funds structuring these very vehicles, is growing fast and is much harder to see [3]. Bond investors have started pulling back as well, with hyperscaler bond demand softening and coverage ratios thinning through the first half of 2026 [2]. Legal, but opaque. That combination is one of the threads I'm still sitting with, and I'd rather admit that than pretend I've resolved it.
Where Africa fits
Here's what should stop anyone who assumes this is purely a raising-money problem. Africa isn't short of capital. The Africa Finance Corporation estimates the continent holds over $4.4 trillion in investable domestic capital across pension funds, insurance, and sovereign wealth [9], and pension and insurance assets alone have crossed $1 trillion [6]. The African Development Bank puts the continent's annual infrastructure financing need at $130 to $170 billion [7]. The money to cover that gap several times over already exists inside African institutions today. Less than three percent of it is currently allocated to infrastructure [11].
What's missing isn't capital, it's plumbing: a mismatch between local currency capital and dollar-denominated projects, a shortage of bankable, fully-contracted projects that lenders can actually underwrite, and thin credit guarantee infrastructure outside a handful of markets [11][8].
Some of that plumbing already exists for power, just not yet for compute. Nigeria's InfraCredit provides local currency guarantees that let pension funds back infrastructure they'd otherwise avoid, most recently guaranteeing local currency debt for a portfolio of solar mini-grid projects [8]. Africa50 exists specifically to move projects from concept to bankable, then help finance them [10]. Nigeria's pension regulator already permits funds to allocate up to 15 percent of assets to infrastructure bonds [8]. Harith and the Africa Finance Corporation recently assembled a $3.3 billion power portfolio serving 30 million people across ten countries using exactly this model [10].
Building something similar for compute means pointing these same tools, credit guarantees, anchor offtake contracts, wider pension allocation rules, at data centers and power-for-compute instead of just power. It also means insisting on local currency debt matched to local currency revenue, so the financing doesn't become another chapter in Africa's long history of dollar-debt distress. Fast money and sustainable money aren't the same thing, and the American market's own opacity problem is a preview of what happens when that distinction gets ignored.
What I still don't know
I don't have a confident answer for how worried we should be about the off-balance-sheet question, or whether a compute-specific guarantee facility is a two-year project or a ten-year one. What I'm certain of is that the framework I've laid out here, borrowed tools plus local currency discipline, is a starting point, not a finished model.
If you work in structured finance, project finance, or African capital markets, I want two things from you in the comments. First, tell me where this framework is wrong. Second, and more useful, propose one that could actually work: a way to finance Africa's AI infrastructure buildout at scale without repeating the opacity problem the US market is now confronting, or the dollar-debt traps the continent has lived through before. This series gets sharper with people who've sat across the table from a lender arguing with it in public.
Next week: the third stream
Venture capital chases winners. Debt markets, as messy and opaque as they can be, still answer to investors who can walk away. China's model answers to neither. Next week closes this series by looking at how a state simply decides to finance its own AI infrastructure buildout, no bond prospectus, no private credit fund, no consolidation loophole to debate, and what, if anything, a state-directed approach could offer Africa that markets alone can't.
References
[1] “Mutual funds increasingly channel private debt from AI data centers,” Crypto Briefing. https://cryptobriefing.com/mutual-funds-private-debt-ai-data-centers/
[2] Robert Szczerba, “Bond Investors Push Back As AI Debt Heads Toward $570 Billion,” Forbes, July 2026. https://www.forbes.com/sites/robertszczerba/2026/07/17/bond-investors-push-back-as-ai-debt-heads-toward-570-billion/
[3] “Client Alert: Emerging Litigation Risks in Financing AI Data Centers Boom,” Quinn Emanuel Urquhart & Sullivan. https://www.quinnemanuel.com/the-firm/publications/client-alert-emerging-litigation-risks-in-financing-ai-data-centers-boom/
[4] “AI data center debt has climbed to the top of Wall Street’s credit risk watchlist,” Startup Fortune. https://startupfortune.com/ai-data-center-debt-has-climbed-to-the-top-of-wall-streets-credit-risk-watchlist/
[5] “Private Markets Are Expected to Have a Growing Role in Data Center Financing,” Goldman Sachs Research, June 2026. https://www.goldmansachs.com/insights/articles/private-markets-expected-to-have-growing-role-in-data-center-financing
[6] “Africa PPP 2026 to Spotlight Bankable Infrastructure, Domestic Capital and Project Delivery,” National Law Review. https://natlawreview.com/press-releases/africa-ppp-2026-spotlight-bankable-infrastructure-domestic-capital-and
[7] “African pension funds: The new driving force to finance infrastructure?” African Financial Summit Platform. https://www.afis.africa/en/african-pension-funds-the-new-driving-force-to-finance-infrastructure/
[8] “Local guarantees for local investors in infrastructure projects in Africa and Asia,” ImpactAlpha. https://impactalpha.com/local-guarantees-for-local-investors-in-infrastructure-projects-in-africa-and-asia/
[9] “Unlocking $4.4 trillion: African capital for African infrastructure,” CNBC Africa, April 2026. https://www.cnbcafrica.com/media/7775039115530/unlocking-44-trillion-african-capital-for-african-infrastructure
[10] “Pension funds power Africa’s infrastructure,” African Capital Markets News. https://africancapitalmarketsnews.com/pension_funds_power_africa_infrastructure/
[11] “Opinion: Pension funds are underused in financing Africa’s infrastructure,” ArmHarith. https://armharith.com/opinion-pension-funds-are-underused-in-financing-africas-infrastructure/
