Is There Enough Capital to Fund the AI Build-out?

Issue XIII of Around the Corner traced how AI and electrification capital expenditure is already showing up in Treasury yields, hyperscaler bond issuance and new issue credit spreads, and introduced the governance problem this creates for weighted average cost of capital and internal hurdle rates. The question that follows directly from that finding is a simple one, even if the answer is not: is there enough capital in the world to actually fund the AI build-out, or is the industry approaching a genuine ceiling on what global capital markets can supply?

This issue works through that question in four steps. First, we establish a defensible total cost for the build-out from published, independent research. Second, we set that figure against the size and depth of the capital markets that would need to supply it. Third, we ask the harder question this comparison invites: not whether the money exists in aggregate, but whether it can be committed, underwritten and disbursed fast enough, in large enough concentration, without crowding out everything else those same markets are already funding. Fourth, we trace that pressure through to the banking system and ask what it means for the small and mid-market businesses that have no other source of credit.

01   The Bill: Pricing the Build-out

Estimates of the total AI infrastructure build-out have proliferated over the past eighteen months, and they have also converged. McKinsey puts cumulative global data centre investment at close to seven trillion dollars by 2030, of which some five point two trillion is attributed specifically to AI workloads rather than conventional cloud. JPMorgan's research independently arrives at a comparable range, projecting cumulative AI infrastructure spending of five to seven trillion dollars by the end of the decade, and identifies power — not capital — as the more binding near-term constraint on how quickly that spending can be deployed. Goldman Sachs, using a bottom-up model anchored to Wall Street's forward revenue expectations for data centre chip sales, projects cumulative AI capital expenditure of seven point six trillion dollars between 2026 and 2031, rising from an annualised $765 billion in 2026 to $1.6 trillion by 2031.

Three independent methodologies, built on different assumptions about chip cycles, data centre design and power availability, land within a tight band of five to eight trillion dollars this decade. For the purposes of this analysis, we treat seven trillion dollars as a reasonable central planning estimate for the cumulative build-out to 2030, consistent with McKinsey's headline figure and the midpoint of the JPMorgan and Goldman ranges.

What matters more than the precise total, however, is its shape over time. The build-out is not a single lump sum to be raised once. It is an annual claim on capital that roughly doubles across this decade, from under one trillion dollars a year today to over one and a half trillion dollars a year by 2031. That distinction — between a cumulative stock and a rising annual flow — turns out to be the whole story.

02   The Capital Markets: Size and Depth

Set against those figures, the capital markets that would need to supply this financing look, at first glance, more than adequate. The global bond market stands at somewhere between $127 and $156 trillion US dollars outstanding, with the Bank for International Settlements placing total global debt outstanding above $156 trillion as of August 2025. Global equity market capitalisation reached approximately $158 trillion in 2025, having grown nearly nineteen per cent in a single year. Private credit has reached somewhere between $2 and $3.5 trillion in assets under management, and industry forecasters expect it to approach four to five trillion dollars by the turn of the decade.

On this reading, a seven trillion dollar build-out spread across five years looks entirely modest next to capital markets measured in the hundreds of trillions. If the question were simply whether enough capital exists in the world in aggregate, the answer would be an unambiguous yes — and this issue could end here. It does not, because the size of the stock is the wrong test. No borrower draws on the full stock of global capital markets at once, and no single claim, however large, is settled against total outstanding debt and equity. What matters is the annual flow of new capital that must be raised, underwritten and allocated within a single year, in competition with every other legitimate claim on that same flow.

03   A Flow Problem, Not a Stock Problem

Reframed around annual flow, the picture changes considerably. The IEA's World Energy Investment 2026 report puts total global energy investment at $3.4 trillion for calendar year 2026 — a record — of which $2.2 trillion is directed toward grids, storage, renewables, nuclear, efficiency and electrification, and the remaining $1.2 trillion toward oil, gas and coal. Grid investment alone is expected to reach $550 billion in 2026. This $3.4 trillion figure is not AI specific. It is the pre-existing, ordinary annual cost of running and modernising the world's energy system, and it is rising independently of anything the AI industry does.

Layer the AI build-out's annualised capital requirement — currently around $750 billion and rising toward $1.6 trillion by 2031 — on top of that existing and growing energy investment cycle, and the two are no longer separable. A meaningful share of AI capital expenditure is itself power and grid investment, competing directly with the IEA's $550 billion grid figure for the same transformers, turbines and skilled labour. And energy is only one competing claim. Sovereign borrowing continues at scale, with long-term US Treasury issuance alone reaching $4.8 trillion in 2025. A parallel reindustrialisation and defence capital cycle, distinct from AI but drawing on the same pools of project finance, private credit and infrastructure equity, is accelerating across the United States, Europe and allied economies. None of these claims pause to make room for the other.

This is why the size of the underlying stock of capital markets is largely beside the point. A $150 trillion equity market or a $130 trillion bond market does not commit fresh capital to a single sector at anything like that scale in any given year. Global long-term fixed income issuance across all borrowers and all purposes ran at roughly $11.5 trillion in 2025. An AI infrastructure claim that rises from under $1 trillion today toward $1.5 trillion annually by 2031 is asking for an ever larger share of a primary issuance market that must simultaneously absorb sovereign refinancing, ordinary corporate borrowing, the existing energy transition, and a resurgent industrial and defence capital cycle — all in the same twelve-month window, underwritten by the same finite pool of banks, insurers, pension funds and asset managers.

04   Early Signs of Strain

This is not a purely theoretical concern. Issue XIII documented the early market evidence: rising Treasury yields, a sharp increase in hyperscaler bond issuance volumes, and widening new issue credit spreads across both government and corporate debt markets — all consistent with a market beginning to price the strain of funding AI and electrification capital expenditure alongside its other obligations. Individual transactions have already stalled over exactly this concern. Blue Owl Capital withdrew from discussions to help finance a $10 billion data centre for Oracle and OpenAI in late 2025, with reporting at the time citing concerns among lenders about Oracle's rising debt load. BlackRock's own mid-2026 outlook frames the defining question of the period as a contest between scarcity and abundance, pointing explicitly to capital — alongside power, skilled labour and physical inputs — as a resource now under genuine competitive pressure rather than an assumed constant.

None of this means capital disappears. It means the marginal cost of accessing it rises, allocation becomes more selective, and the advantage shifts toward capital providers able to originate, underwrite and hold assets directly on their own balance sheet, rather than relying on the ordinary competitive cycle of public market issuance and refinancing that every other borrower is also drawing on in the same year.

05   The Banking Channel: Where the Squeeze Actually Lands

The comparison so far has been conducted almost entirely in the language of capital markets — bonds, equities and private credit — because that is where hyperscalers, sovereigns and the largest infrastructure sponsors go to raise money directly. Most of the real economy cannot do this. The overwhelming majority of small and mid-market businesses are unrated, and unrated borrowers do not issue public bonds or commercial paper. The OECD's Financing SMEs and Entrepreneurs 2026 Scoreboard finds that the stock of SME lending across nearly fifty economies remains broadly stagnant, and that banks continue to apply stringent lending terms and elevated borrowing costs relative to pre-pandemic levels. In the United States alone, the Federal Reserve's own Small Business Credit Survey found that roughly two thirds of small businesses that applied for financing in 2024 received less than the full amount they sought.

For this segment of the economy — which is by far the largest employer — there is effectively one channel: the banking system. Banks are not, however, an independent reservoir of capital sitting apart from the markets examined above. They are an intermediation layer funded by exactly the same two sources: retail and commercial deposits, and wholesale issuance into the same bond and money markets. Aggregate deposits at US commercial banks reached a record $19.5 trillion in February 2026, and global banking sector assets in total are estimated to have passed $180 trillion.

This is the transmission mechanism that connects a Wall Street financing question to a high street lending question. If demand for capital from AI infrastructure, sovereign issuers and the reindustrialisation cycle pushes up the marginal cost of funding in the bond and money markets, banks face that same repricing on both sides of their own balance sheet. Banks facing a higher marginal cost of funds do not absorb that cost evenly across their loan book. They tend to protect margins and capital ratios first on their largest, most liquid and most easily collateralised exposures, and ration credit at the margin to the borrowers who are hardest to standardise, price and securitise — which is precisely the small and mid-market segment already showing signs of a stagnant loan stock and tightening terms. The regional banking stress of 2023 illustrated the same mechanism in acute form, as smaller lenders pulled back first from exactly this segment. A rise in the cost of capital for banks, driven by competition for capital at the top of the market, is therefore not a contained wholesale market story. It is a direct channel through which the AI capital squeeze reaches the small and medium-sized businesses that have no alternative source of funding.

06   Monard's View: A New Fixed Income Channel for Infrastructure

Monard's reading of this evidence is that capital available to fund the AI and electrification build-out will become progressively more scarce over the remainder of this decade — not because the world's aggregate savings pool is too small, but because the traditional origination and underwriting channels that convert that pool into deployable annual capital were built to fund a diversified economy, not one facing a rising $5–8 trillion concentrated claim layered on top of an already record energy investment cycle, a resurgent sovereign borrowing programme, a parallel reindustrialisation and defence build-out, and a banking system funded from the same stretched pool of deposits and wholesale capital.

Monard's answer is to change which pool of capital this build-out draws on in the first place, rather than compete for a larger share of pools that are already stretched. A well-structured supply agreement — an offtake or power purchase style contract with a creditworthy counterparty, a take-or-pay or minimum volume commitment, a long tenor and properly drafted credit support provisions — is not merely a commercial arrangement. It is, in substance, a contracted and predictable cash flow of the same underlying quality as the investment-grade project bonds already used to finance LNG export terminals and contracted renewable generation. Structuring supply agreements explicitly to investment-grade fixed income standards from origination is what allows an infrastructure project to be funded directly out of the $135 trillion global bond market — the single deepest pool identified in this issue — rather than out of scarce bank balance sheet capacity or an increasingly congested corporate and sovereign issuance calendar.