Nvidia: the $500 billion that does not exist yet

On 10 August 2026, Nvidia announced a plan to mobilise more than $500 billion for artificial-intelligence infrastructure. The chipmaker attached six names that gave the number immediate weight: Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR.

At first glance, the announcement looks like the creation of an enormous financing pool. Nvidia’s own release describes something different. It refers to memorandums of understanding, third-party capital to be mobilised over time and partnerships that remain subject to definitive agreements. It allocates no commitment to any individual partner. The split between equity and debt, the deployment schedule and the price of any guarantees remain unknown.

The $500 billion target may eventually be reached. As of 10 August, that capital had not been raised, committed or even defined through a public calculation method.

Five points to retain

• Nvidia signed six memorandums of understanding, not six definitive financing agreements.
• BlackRock, Brookfield and KKR already operated Nvidia-linked platforms before the announcement.
• The headline figures mix equity, debt, co-investment, assets under management and future investment capacity.
• Jensen Huang has described a potential Nvidia backstop of up to $125 billion without publishing its terms.
• The financing need is credible. The $500 billion figure cannot yet be audited from the information available.

Five words that change the meaning of the announcement

The Nvidia release contains five essential qualifications:

1. the partners signed memorandums of understanding;
2. they intend to mobilise capital, a broader verb than raise or commit;
3. the target is third-party capital, supplied by investors, lenders and managed vehicles;
4. the objective is to be pursued over time;
5. the partnerships remain subject to definitive agreements.

The release supplies no partner-level amount, deployment schedule, target return or debt-equity split. Reuters likewise reported that the financial terms, individual investment commitments and timetable had not been disclosed.

That omission prevents an answer to the simplest question: how much new money have the six institutions legally committed to provide?

One number, several accounting systems

The partners’ releases use billions of dollars to describe very different things.

Adding those numbers would produce a spectacular and misleading total. Their periods, scopes and financial meanings differ. Some refer to committed capital, others to an equity target, debt capacity, asset value or a development pipeline. Several programmes may also overlap.

The nature of each dollar must be traced before any attempt is made to add them together.

BlackRock: how $30 billion can become $100 billion

BlackRock offers the clearest example.

In September 2024, BlackRock, Global Infrastructure Partners, Microsoft and MGX launched the Global AI Infrastructure Investment Partnership. The platform sought to unlock $30 billion of private equity and mobilise up to $100 billion of total investment potential when debt financing was included. Nvidia was already providing technical expertise.

Nvidia and xAI joined in March 2025, when the platform was renamed the AI Infrastructure Partnership. The same $30 billion and $100 billion targets remained in place. The Kuwait Investment Authority later became AIP’s first non-founder financial anchor investor.

Nvidia’s 10 August 2026 release explicitly acknowledges that continuity. Larry Fink describes the new partnership as a deepening of the existing relationship, including through AIP.

The mechanism is transparent in principle: $30 billion of equity can support up to $100 billion of investment after debt is added. The ratio is 3.3. It is not necessarily the final leverage ratio of each underlying project, but it shows why capital mobilised and capital contributed are not synonyms.

One question remains unanswered. Are the $100 billion already targeted by AIP included in Nvidia’s new $500 billion headline? The release does not say.

Brookfield: from $5 billion committed to $100 billion of assets

Brookfield takes the same logic further.

In November 2025, the asset manager launched a $100 billion global programme with Nvidia and the Kuwait Investment Authority. Its flagship fund targeted $10 billion of equity commitments and had received $5 billion of capital commitments at launch.

Brookfield said that the fund, additional co-investor capital and prudent financing could acquire up to $100 billion of AI-infrastructure assets. The maximum figure was therefore twenty times the capital actually described as committed at launch and ten times the fund’s equity target.

That does not mean Brookfield intends to lever every asset tenfold. Co-investors can provide additional equity. It does mean that the $100 billion figure combines several layers of funding with different risks, costs and degrees of certainty.

Brookfield had also announced partnerships in France and Sweden representing up to $30 billion of combined AI-infrastructure investment. Whether any part of those programmes is included in Nvidia’s total is unknown.

KKR: $10 billion had already been committed

On 11 June 2026, KKR, Nvidia, Vistra and the Kuwait Investment Authority launched Helix Digital Infrastructure. The company began with more than $10 billion of long-duration capital commitments and remained open to other eligible institutional investors.

Nvidia was both a founding investor and a strategic partner. KKR said its anchor investment was being funded through its balance sheet and managed vehicles. Helix was designed to invest across hyperscale data centers, power generation, transmission and connectivity.

Two months later, Nvidia’s $500 billion release presented the KKR relationship as an extension of Helix. As with BlackRock, the continuity is explicit.

Here, the more than $10 billion is described as committed, giving it more substance than a fundraising objective. The allocation among KKR, Nvidia, KIA and Vistra is not public. Nor is it clear whether the amount is included in the $500 billion target.

Apollo and Blackstone were already financing compute

Apollo was not starting from scratch. In January 2026, Apollo-managed funds and affiliates led a $3.5 billion capital solution supporting Valor’s acquisition and lease of $5.4 billion of compute infrastructure to an xAI subsidiary. The assets included Nvidia GB200 GPUs and the transaction used a triple-net lease. Nvidia was an anchor limited partner in Valor Compute Infrastructure.

Apollo said it had deployed more than $40 billion since 2022 across a broader category of next-generation infrastructure that included compute, digital platforms and renewable energy. That figure cannot be attributed to Nvidia.

Blackstone already controls a huge global data-center platform. On 30 June 2026, the manager said it expected to invest or commit roughly $100 billion across its own portfolio by the end of 2026. It valued its global platform, including facilities under construction, at $165 billion and identified another $160 billion of future development pipeline.

Two months before Nvidia’s announcement, Apollo and Blackstone also launched a Broadcom compute-financing platform with an initial $35 billion transaction. That transaction concerned Broadcom accelerators, not Nvidia GPUs. It demonstrates that the same managers were already creating financing markets around more than one chip supplier.

Both readings matter. Apollo and Blackstone clearly have the capacity to execute large transactions. A substantial part of their financial machinery and project pipelines was already operating before 10 August.

Goldman Sachs: the blank cell in the table

Goldman Sachs is the least quantified partner in Nvidia’s announcement.

David Solomon describes two roles, investment and distribution, and refers to building a credit market around Nvidia compute. The release does not say whether Goldman will deploy balance-sheet capital, infrastructure funds, private-credit vehicles or primarily its structuring and placement capabilities.

Goldman has pools of capital that could fit the project. In January 2026, the Qatar Investment Authority signed an MoU targeting $25 billion of commitments to Goldman Sachs Asset Management funds and co-investment opportunities. The agreement names AI, digital infrastructure and private credit among several priority sectors.

Nothing in the QIA-Goldman release allocates those $25 billion to Nvidia’s programme. Counting the amount toward Nvidia’s total would be speculative double counting.

A $125 billion backstop would change the scale of Nvidia’s risk

Potentially the most consequential detail does not appear in the official release.

Jensen Huang said on X that Nvidia would have the option to backstop up to $125 billion, equal to 25% of the potential deals, according to Reuters. No published document specifies its duration, beneficiaries, triggers, covered assets, seniority or the compensation Nvidia would receive.

Backstop can describe several mechanisms. Nvidia might guarantee lease payments, debt service, residual value, a purchase commitment or only a portion of losses beyond a threshold. The accounting and economic exposure would differ radically from one contract to another.

The maximum can still be compared with Nvidia’s current balance sheet. As of 26 April 2026, the company reported $50.3 billion of cash, cash equivalents and marketable debt securities, plus $30.2 billion of marketable equity securities. A $125 billion backstop would equal:

• approximately 2.5 times the $50.3 billion of cash and marketable debt securities;
• approximately 1.55 times the combined $80.5 billion after marketable equity securities are added.

This comparison measures scale, not a probable loss. A backstop could be contingent, deployed over several years, spread across multiple transactions and supported by future cash flows.

The same 10-Q nevertheless shows the distance from Nvidia’s existing disclosed arrangements. The company had $27 billion of investment commitments subject to contingencies. Its maximum loss exposure to equity-method infrastructure funds was $2.3 billion. Facility-lease guarantees provided to partners carried maximum gross exposure of $3.5 billion in exchange for warrants, with $712 million placed in escrow.

A capacity of $125 billion would therefore move Nvidia into a different category as a financial guarantor. It cannot be treated as a routine expansion of the existing arrangements without seeing the contracts.

Why Nvidia is organising financing for its own customers

Nvidia says the platforms are intended to create dedicated capital pools at attractive rates for its customers. That formulation directly connects financing availability with demand for Nvidia infrastructure.

The company identifies the same dependency in its 10-Q. The availability of data centers, energy and capital is described as crucial to future revenue. Nvidia warns that limited access to capital can delay customer deployments or reduce their scale, particularly for less-capitalised companies.

The economics are straightforward. A lower cost of capital allows more customers to buy or lease Nvidia compute. Asset managers obtain projects and contracted cash flows to finance. Nvidia sells more hardware and software. Customers avoid carrying the entire upfront investment themselves.

The partners say that the platforms will be independent and that each project will be underwritten on its own merits. The contracts will determine how meaningful that independence is: eligibility criteria, guarantee pricing, first-loss exposure, customer selection and conflict management all matter.

This structure should not automatically be labelled circular financing. l0g has already examined the capital loops inside the AI market. The new development is the institutionalisation of the mechanism. Nvidia is trying to turn financing for its own ecosystem into an asset class that can be distributed at very large scale.

A $500 billion financing need is plausible

The announcement’s lack of precision does not make the order of magnitude absurd.

Goldman Sachs Research estimates that the leading technology companies could spend $5.3 trillion on compute and data centers from 2025 through 2030. Private infrastructure funds held almost $400 billion of dry powder as of September 2025, while private real-estate funds held roughly $600 billion more.

Another estimate, cited by the Structured Finance Association, puts global data-center capital expenditure through 2028 at $2.9 trillion. Roughly $1.5 trillion would need to be financed outside hyperscaler cash flows. These are market forecasts, not commitments.

The serious question is not whether $500 billion of financing needs can be found. It is how much genuinely new capital will be supplied, how much leverage will be used, who the borrowers will be, how losses will be allocated and what mobilised means in Nvidia’s accounting.

What is established

• Six memorandums of understanding were signed.
• The definitive agreements had not been executed when Nvidia published its release.
• No partner-level commitment, timetable or debt-equity split is public.
• BlackRock’s AIP, Brookfield’s AI Infrastructure Program and KKR’s Helix already involved Nvidia before the announcement.
• BlackRock explicitly describes a $30 billion equity target capable of mobilising up to $100 billion when debt is included.
• Brookfield described $5 billion committed, a fund targeting $10 billion of equity and the capacity to acquire up to $100 billion of assets with co-investors and financing.
• Apollo and Blackstone had already launched compute-financing transactions worth tens of billions of dollars.

What remains hypothetical

• The $500 billion will actually be mobilised.
• The total represents new and additional capital.
• Existing programmes will not be counted again.
• Nvidia will provide a $125 billion backstop.
• The six partners will contribute comparable amounts.
• Each platform will retain a meaningful share of the economic risk.

Public questions for Nvidia and the six partners

The documents available as of 11 August 2026 do not answer the questions below. l0g is putting them publicly to the companies concerned and will incorporate any response supported by verifiable evidence.

1. To Nvidia: what precise definition of “capital mobilised” produces a figure above $500 billion?
2. To Nvidia and all six partners: what legally binding amount has each party accepted, on what timetable and with what split between equity and debt?
3. To BlackRock, Brookfield and KKR: are amounts previously announced through AIP, the AI Infrastructure Program and Helix included in the $500 billion, and if so, in what proportion?
4. To Apollo and Blackstone: what amounts and vehicles are reserved for the Nvidia programme, separately from the already announced Valor / xAI and Broadcom transactions?
5. To Goldman Sachs: how much will come from its balance sheet or managed funds, and how much will reflect structuring and placement with third-party investors?
6. To Nvidia: what exactly would the potential $125 billion backstop cover, what would trigger it, how would Nvidia be compensated and how would the exposure be accounted for?
7. To all partners: who would bear first losses, GPU residual-value risk, customer credit risk and the risk of technological obsolescence?
8. To all partners: what criteria will ensure that the platforms remain independent when they finance customers buying Nvidia infrastructure?

A financial infrastructure before it is a sum of money

The $500 billion figure first sends a message to Nvidia’s customers: financing should no longer be the constraint on purchasing compute. It also sends a message to institutional investors: GPUs, data centers, usage contracts and power infrastructure can be assembled into financeable assets.

The six partners give that ambition real credibility. Their funds, balance sheets and distribution networks can move hundreds of billions of dollars. The public documents also show that part of the architecture already existed and that the largest headline amounts rely on multipliers between equity, co-investment and debt.

Nvidia did not raise $500 billion on 10 August 2026. It announced the construction of a machine that might eventually mobilise that amount. The financial risk sits in the difference between those two sentences.

That raises another public question: how can a data center issue securities backed by its revenue without being treated as an asset-backed security by the US regulator?

---

Main sources

• Nvidia, announcement of the six platforms, 10 August 2026.
• Nvidia, Form 10-Q, balance-sheet date 26 April 2026.
• BlackRock, launch of GAIIP, 17 September 2024.
• BlackRock, Nvidia joins AIP, 19 March 2025.
• BlackRock, KIA joins AIP, 3 June 2025.
• Brookfield, launch of the AI Infrastructure Program, 19 November 2025.
• Helix, launch with KKR, KIA, Nvidia and Vistra, 11 June 2026.
• Apollo, Valor / xAI financing, 7 January 2026.
• Blackstone, 2026 mid-year investment perspectives, 30 June 2026.
• Goldman Sachs and QIA, MoU with a $25 billion target, 20 January 2026.
• Structured Finance Association, the data-center financing ecosystem, 23 July 2026.
• Reuters, the announcement and potential backstop, 10 August 2026.

Method and limitations

• Evidence cut-off: 11 August 2026. The article uses no current security price, valuation multiple or investment recommendation.
• The figures in the table are not added together because they cover different periods, vehicles and layers of financing.
• Amounts disclosed by BlackRock, Brookfield, KKR, Apollo, Blackstone and Goldman Sachs describe their own announcements. They do not prove inclusion in Nvidia’s target.
• The $125 billion backstop comes from a Jensen Huang statement reported by Reuters. It appears neither as a definitive commitment in Nvidia’s release nor as a liability in the 26 April 10-Q.
• Goldman Sachs Research and Morgan Stanley estimates, the latter cited by the Structured Finance Association, measure future financing needs. They are not capital already raised.
• The article separates published facts, objectives, capacity and hypotheses. Its conclusions should be revised if definitive agreements or guarantee terms become public.