Subscribe to The Value Anchor Topic / The Trend Catcher Topic —unlock the full historical archive and never miss a weekly pick again.

Executive Summary

NVIDIA Corporation(NVDA.US) is no longer content with simply selling chips. The company now aims to transform its chips into assets that can be priced, collateralized, and financed by Wall Street. NVIDIA Corporation(NVDA.US) announced a partnership with Apollo Global Management Inc(APO.US), BlackRock, Inc.(BLK.US), Blackstone Inc.(BX.US), Brookfield Asset Management Inc(BN.US), Goldman Sachs Group, Inc.(GS.US), and KKR & Co(KKR.US) to mobilize over $500 billion in third-party capital through an independent AI compute infrastructure financing platform. The goal: to accelerate the construction of data centers and “AI Factories” for leading AI labs, enterprises, governments, and AI cloud providers.

It is important to note that this $500 billion represents a planned fundraising pool to be gradually mobilized in the future. This is not money that has already been raised, nor is it revenue Nvidia can immediately recognize. Details such as final agreements, exact commitments, and the deployment timeline have yet to be released.

Despite heavyweight financial institutions being willing to put compute power on their balance sheets, both equity and credit markets increased Nvidia’s risk premium. These seemingly contradictory reactions actually point to a common underlying question:

Is AI compute power now truly long-term, cash-flowing infrastructure—or is Wall Street simply using more complex financing structures to turn chip orders, untested by end-demand, into near-term revenue?

To understand the debate, you have to view the $500 billion plan within Jensen Huang’s “Five-Layer Cake” framework.

From Industrial Map to Capital Structure: How Does $500 Billion Flow Through the “Five-Layer Cake”?

Nvidia CEO Jensen Huang suggested at Davos this year that AI is not a singular model or application, but a “cake” composed of five interconnected layers:

• Layer 1: Energy – Every token generated requires electricity, power transmission/distribution, land, and cooling. Grid capacity physically limits compute expansion.
• Layer 2: Chips – GPU, CPU, HBM, networking, and optical interconnects transform electricity into compute, determining token generation efficiency and cost.
• Layer 3: Infrastructure – Data centers, AI clouds, server clusters, and scheduling systems aggregate thousands of chips into saleable “AI Factories.”
• Layer 4: Models – Labs like OpenAI, Anthropic, xAI purchase and utilize compute for training and inference, translating compute into intelligence.
• Layer 5: Applications – Enterprise software, autonomous driving, robotics, healthcare, and fintech convert model capabilities into customer revenue—ultimately yielding the economic returns.

This $500 billion initiative does not add a “sixth layer” but instead installs a “capital pipeline” around the five-layer cake:

• Wall Street’s long-term capital primarily focuses on energy, chips, and data centers.
• Model companies purchase compute capacity via leasing or usage contracts.
• The top-layer application revenue pays for model services and compute, servicing project debt and delivering returns to capital.

Thus, the critical question for this structure is not whether GPUs can be sold—but whether top-layer applications can consistently generate enough cash flow to flow downward through the five layers. Financing can accelerate infrastructure buildout, but cannot conjure terminal demand out of thin air.

Nvidia’s Ambition: Turning CUDA Into “Credit Enhancement”

Traditional tech hardware rarely qualifies as solid collateral for long-term financing due to rapid iteration and high depreciation. When demand weakens, the value of second-hand equipment can plummet.

Nvidia must convince Wall Street that GPUs are now productive assets capable of reliably generating token-based revenue, not just ordinary electronics. Their argument rests on three pillars:

• Nvidia GPUs are usable across models and workloads.
• Equipment is transferable or leasable among multiple cloud providers and operators.
• CUDA software constantly optimizes and extends the useful lifespan and compatibility of older hardware.

If second-hand prices, lease rates, and utilization remain stable, GPUs can be leveraged for financing—much like aircraft, cell towers, or data centers—against future rental income.

Bank of America Securities notes in their research report, “Nvidia is underwriting asset value, not the loan”:

Nvidia does not need to lend all the capital itself; as long as the CUDA ecosystem maintains GPU residual value and liquidity, it essentially provides a layer of technology-backed credit to computing assets.

Put differently, Nvidia is evolving from chip supplier to a triple role: manufacturer, standard setter for compute assets, and initiator channeling Wall Street capital into AI industry buildout.

If this model works, CUDA’s moat expands beyond developer/software lock-in to become a lower financing-cost moat: data centers using Nvidia platforms may obtain loans more easily, and at lower cost of capital.

Why BofA Is Bullish: Handing Capital Pressure to Wall Street and Prolonging the AI Investment Cycle

From a bullish perspective, what currently constrains AI expansion isn’t demand orders, but capital, electricity, and construction capacity.

Large cloud providers can self-fund. But frontier model developers, “Neoclouds,” sovereign AI projects, and lower-rated enterprises often lack the balance sheets for massive investment. This $500 billion fund aims to lower their cost of capital, pulling forward years' worth of compute demand into shovel-ready projects.

Bank of America Securities considers this arrangement a clear positive in their report, primarily because:

“Having institutions like Apollo and Blackstone independently syndicate and organize capital disperses most of the credit risk across financial consortia and at the project level, rather than putting it all on Nvidia’s own balance sheet.”
BofA believes this supports their forecast of a $1.7 trillion AI systems market by 2030 and helps extend the current AI infrastructure supercycle.

Another crucial point: third-party financing also alleviates concerns over Nvidia’s dual role as both customer investor and chip supplier.

BofA estimates Nvidia’s current direct equity commitments in ecosystem partners at around $70 billion, including up to $30 billion invested in OpenAI, up to $10 billion in Anthropic, as well as holdings in Intel Corporation(INTC.US), Synopsys, Inc.(SNPS.US), CoreWeave(CRWV.US), NEBIUS(NBIS.US), Lumentum Holdings, Inc.(LITE.US), Coherent Corp.(COHR.US), Marvell Technology(MRVL.US), Corning Inc(GLW.US), and IREN Limited(IREN.US).

These investments span nearly the entire “Five-Layer Cake”:

• Energy/Data Center Layer: IREN Limited(IREN.US), Nscale, CoreWeave(CRWV.US), NEBIUS(NBIS.US)
• Chips/EDA/Interconnect Layer: Intel Corporation(INTC.US), Synopsys, Inc.(SNPS.US), Marvell Technology(MRVL.US), Lumentum Holdings, Inc.(LITE.US), Coherent Corp.(COHR.US), Corning Inc(GLW.US)
• Model Layer: OpenAI, Anthropic, xAI
• Application Layer: Autonomous driving company Wayve

$70 billion may seem large, but it’s just 15% of the $469 billion in free cash flow BofA expects Nvidia to generate from 2026-2027. If most incremental financing comes from external sources, Nvidia can continue funding the ecosystem while preserving cash for buybacks and shareholder returns—why BofA views the new platform as healthier than simple “vendor financing.”

Not Purely Off-Balance Sheet: $125 Billion Support Option Leaves Tail Risk

However, risks have not fully vanished for Nvidia. Jensen Huang stated on X that Nvidia reserves the right to provide up to $125 billion—roughly 25% of the platform’s scale—in potential support.

“Right to provide support” does not mean Nvidia must pay out $125 billion, nor does it represent an immediate liability; but it does mean risk is not completely shifted to third parties. The actual risk profile depends on the form of this support: minority equity, first-loss capital, repurchase commitments, minimum lease guarantees, or recourse guarantees—all of which have vastly different implications for free cash flow, credit rating, and buyback ability.

BofA’s report notes these conditions:

While it preliminarily judges most risk will rest with the consortium, it awaits final documentation and Nvidia’s August 26 earnings call for clarity.

If more cash support must be deployed, free cash flow earmarked for buybacks could be squeezed.

Is “Circular Financing” Valid? The Key Is Whether End-Users Actual Pay

Concerns about circular financing are not unfounded:

• Nvidia invests in model companies and AI cloud providers,
• Financial institutions lend to these same firms,
• Firms use proceeds to buy Nvidia systems—thus creating a closed-loop capital flow.

But a “closed loop” does not inherently equal a bubble. Aircraft leasing, telecom equipment, and mega energy projects all rely on a mix of vendors, long-term capital, and operators.

The true distinction between “infrastructure financing” and “manufactured demand” is whether there is an independent, sustainable, paying end user outside the loop.

If enterprises can boost revenue or cut costs through AI, app developers are willing to pay for model services, and model companies can pay compute leases in cash, then project financing will have real cash flow support. Leverage in this scenario lowers funding costs and speeds up supply building—the cake forms a virtuous cycle, with application revenue repaying layers below.

Conversely, if model revenue suffices only via related-party contracts and GPU orders are mainly supported by loose credit, then financing is merely front-loading future demand. As chips iterate, rentals decline, or funding costs increase, data center residual values and debt service both worsen—risk can flow back up to Nvidia via guarantees, buybacks, or leasebacks.

Thus, a rising CDS spread is best interpreted as investors repricing this tail risk, not as evidence of bubble or imminent collapse. CDS prices reflect both default expectations and hedging/liquidity/positioning—acting as warning signals, not verdicts.

Supercycle or Bubble? Five Key Metrics to Watch

How much Recourse Risk Does Nvidia Actually Bear?

• Will the final agreement cap Nvidia support at $125 billion, and what form does that support take? This will determine whether this model is “asset certification” or disguised vendor guarantees.

Who Are the Ultimate Buyers?

• Projects with long-term, take-or-pay, or minimum-commitment contracts from highly rated end customers provide higher-quality cash flows than those relying mainly on early-stage model startups.

Can GPU Residual and Rental Values Endure Product Cycles?

• Second-hand prices, lease rates, cluster utilization, and legacy hardware economics for new models will test the core hypothesis that “compute is an investable asset.”

Can Application Layer Revenue Catch Up with Infrastructure Capex?

• If model usage, corporate AI subscriptions, and inference income accelerate, the layers are already cash-flowing; if capex outpaces monetization, circular financing concerns will grow.

How Does Nvidia Allocate Its Free Cash Flow?

• If, after the platform launches, Nvidia still delivers robust buybacks/shareholder returns, risk is being diffused; if support/guarantees/leases keep swallowing cash, markets will revisit valuations and credit risk.

Conclusion: $500 Billion Is Not Demand Proof, But a Much Bigger Demand Test

From the “Five-Layer Cake” perspective, Nvidia’s Wall Street collaboration is not just about funding chip orders, but about packaging energy, chips, data centers, models, and applications into infrastructure assets held by global capital.

The bull case is clear:

Nvidia uses CUDA to underwrite compute asset value, third-party capital to lower customer financing costs, and faster AI factory buildout to cement its chip/network/software dominance.

If the application layer ultimately generates sufficient cash flow, the $500 billion will become an AI supercycle accelerator, expanding Nvidia’s moat from technical to financial ecosystems.

The bear case cannot be ignored:

The $500 billion remains a framework, not deployed capital; the final agreements are pending; Nvidia may yet assume support for some deals. If monetization lags capex, leverage won’t remove risk but will amplify overcapacity and asset value write-downs.

Thus, the real question is not “Will Wall Street lend?” but:

Can the top-layer applications feed the ever-growing assets and liabilities of the lower four layers?

Nvidia is not only adding $500 billion of leverage to the AI “Five-Layer Cake”—it is also adding a timer:

In the coming years, AI must transition from a compute-power narrative to a genuine cash-flow narrative.

Disclaimer: The content is provided as general information only and should not be taken as investment advice. All the contents shall not be taken as a recommendation to buy or sell any security or financial instruments. Any action you take resulting from information, analysis, or commentary on this article is your responsibility. Please consult your investment advisor before making any investments.