Part IV of the series “AI Infrastructure, Capital and Power”.
Part I: The Boldest Measure May Be the Safest·Part II: Apparently, $500 Billion Is Not Enough·Part III: What Must the Future Earn?
Three Months Later
On 10 August 2026, CNBC anchor Becky Quick convened an extraordinary gathering of American business leaders.
CNBC Roundtable, 10 August 2026: Read the transcript · Watch on YouTube
Jensen Huang, founder and CEO of Nvidia, sat alongside senior figures from BlackRock, Blackstone, Apollo, Brookfield, Goldman Sachs, and KKR. Together, they unveiled an initiative designed to mobilise more than $500 billion in capital over the coming years to accelerate the build-out of artificial intelligence infrastructure.
It was a remarkable moment. The institutions that shape global capital markets rallied behind the vision of a semiconductor manufacturer. What had hitherto been viewed primarily as technological progress was to be transformed into a new, institutionally financeable infrastructure asset class. AI compute was compared to electricity and telecommunications; data centres were no longer to be treated merely as technical facilities, but as long-lived, productive assets.
Huang had won over the technology industry. Now, he sought to bring Wall Street into the fold.
Three months later, the landscape has shifted fundamentally.
Following a leadership crisis, Jensen Huang has stepped down as CEO of Nvidia. Chief Financial Officer Colette Kress has assumed leadership on an interim basis. According to media reports, Kress had previously questioned some of the company’s more expansive financing proposals. She now faces the task of overseeing its next phase.
The technological revolution is by no means over. Yet the way it is financed may be entering a different phase—and with it, the valuations of companies whose growth depends on the pace of AI infrastructure deployment.
The Original Blueprint
In the first three installments of “AI Infrastructure, Capital and Power”, we examined how technology, power generation, industrial capacity, and capital markets have become increasingly intertwined. Our central thesis was straightforward: the AI build-out has long ceased to be a mere technology capex cycle. It is evolving into an integrated industrial and financial system.
Behind every modern AI data centre stand semiconductors, memory, optical interconnects, cooling systems, power generation, transmission grids, real estate, and vast construction outlays. Mobilising the capital required for this undertaking is an economic challenge in its own right.
On 10 August, Nvidia outlined a potential solution:
- BlackRock was positioned to unlock long-term institutional capital.
- Blackstone and Brookfield brought deep expertise in developing and financing physical infrastructure.
- Apollo represented the reach and flexibility of private credit markets.
- Goldman Sachs and KKR rounded out the architecture with capital markets access, structuring capabilities, and institutional distribution platforms.
At its core, the announcement was a framework agreement to build dedicated financing platforms; the headline $500 billion figure did not represent committed or deployed funds. Even so, the message was unmistakable: US capital markets were being mobilised to dramatically compress the timeline of AI infrastructure expansion.
Jensen Huang’s thesis rested on a pivotal premise: modern AI compute could be financed as a productive, fungible, and cash-generating asset. That premise formed the bedrock of the entire structure. If compute is indeed an economically viable utility with predictable revenue streams, it can legitimately be backed by long-term debt.
The crucial question, however, remains: how reliable and durable are those revenues?
When the Manufacturer Becomes the Financier
In traditional industries, a manufacturer sells equipment to customers who fund their capital expenditure out of internal cash flows or independent third-party financing. In the AI sector, a far more intricate arrangement has taken shape.
Nvidia sells high-performance processors and networking hardware. At the same time, it takes direct financial stakes in customers, technology partners, and infrastructure ventures that, in turn, drive demand for Nvidia’s own systems.
Strategic investments of this kind can clear supply bottlenecks, accelerate adoption, and catalyse entire markets. Difficulties arise, however, when the boundary between organically funded demand and vendor-supported demand begins to blur.
A customer ordering additional Nvidia clusters because Nvidia—or a financing vehicle backed by Nvidia—has provided the capital generates genuine top-line revenue in the near term. Yet the economic durability of that revenue depends entirely on whether the customer can subsequently generate sufficient cash flow to service the investment independently.
As long as demand accelerates, these vulnerabilities remain largely invisible. Should financing costs rise, hardware depreciate faster than modelled, or end-customer monetisation fall short, the very same mechanism can turn into a liability:
Vendor financing → accelerated orders → capacity-driven build-out → rising fixed commitments → valuation risk as refinancing tightens
The issue, therefore, is not whether vendor-backed financing is permissible or commercially rational. The real question is: how much of this growth would occur without the financial support of the technology supplier?
Four Billion Dollars as a Case in Point
Two transactions from the spring of 2026 illustrate the scale of these commitments.
In March, Nvidia invested $2 billion in Coherent, a leading manufacturer of optical technologies, as part of a broader strategic partnership to secure future production capacity. Weeks later, it committed a further $2 billion to Marvell Technology, targeting custom AI silicon and tighter integration with Nvidia’s NVLink interconnect architecture.
Both companies perform vital industrial roles; these were far from speculative financial bets. Nevertheless, they created deep financial interdependencies with key suppliers and ecosystem partners.
Here lies a practical constraint for any strategic recalibration. An abrupt exit from existing equity stakes could weigh heavily on the share prices of the companies involved and cast doubt over ongoing operational collaboration. Prudent management would therefore distinguish carefully between legacy commitments and new capital allocation.
Already committed funds and long-term supply agreements cannot simply be unwound overnight. New financing arrangements, by contrast, can be subjected to far tighter hurdles. This points toward a measured approach: maintaining existing strategic holdings, honouring contractual obligations, capping new balance-sheet exposure, and gradually tapering financial support over time. This is a scenario rather than an announced policy—yet it aligns with the logic of orderly risk reduction.
Colette Kress and the New Discipline
Following Jensen Huang’s departure, Nvidia faces an unusual leadership test: sustaining its technological cadence while subjecting its financing strategy to rigorous economic scrutiny.
As long-serving CFO, Colette Kress possesses an intimate command of Nvidia’s balance sheet, investment portfolio, and contingent liabilities. Reuters reported in November that she had previously acted as an internal counterweight to some of the company’s more aggressive financing proposals.
While this does not establish a formal pivot, it makes it plausible that economic discipline will carry greater weight in evaluating future commitments under her stewardship.
Such a shift could strengthen Nvidia over the long run. Fewer equity investments, lower contingent obligations, and a tighter alignment between capital deployed and organic cash generation would enhance balance-sheet resilience. In the near term, however, that same discipline could moderate top-line expansion.
Therein lies the central trade-off. Nvidia could emerge with higher earnings quality even as market expectations for revenue and profit growth are revised downward. For equity investors, this is no trivial nuance: a company can remain operationally dominant and still experience multiple compression if the market reassesses its long-term growth trajectory.
The Earnings Consequences
A more conservative financing posture would carry several distinct financial implications:
- Valuation volatility: Mark-to-market fluctuations on existing strategic investments can affect reported earnings without altering underlying operating profits from semiconductor sales.
- Demand deferral: If less external capital is channelled into incremental AI capacity, orders for GPUs, high-bandwidth memory, and networking equipment could be pushed out across quarters.
- Cost asymmetry: At the same time, substantial purchase and capacity commitments with foundry and supply-chain partners are already locked in. The cost base of an accelerated expansion cannot necessarily be scaled back as rapidly as new financing pledges.
Demand can decelerate faster than industrial cost structures adjust. Whether—and to what extent—this plays out will depend on underlying end-user demand, contractual flexibility, and the continued capital expenditure appetite of the major hyperscalers. A shift in financing policy need not trigger an outright revenue contraction, yet it can temper expected growth rates and heighten earnings volatility.
Back at the Roundtable
Imagine Becky Quick reconvening the participants of 10 August today. Jensen Huang would no longer occupy Nvidia’s seat; in his place would sit Colette Kress, perhaps accompanied by the Chairman of the Board.
Larry Fink, Jon Gray, Jim Zelter, Bruce Flatt, David Solomon, and Waldemar Szlezak would confront the same fundamental economic question as in August—but under markedly different premises. Back then, the focus was on mobilising capital at scale. Today, the conversation would centre on project-level unit economics:
- BlackRock would need to verify whether institutional clients can earn durable, predictable yields over multi-year horizons.
- Blackstone would assess which data centres can command sufficient tenant cash flows upon completion to justify construction costs.
- Apollo would scrutinise credit quality, collateral coverage, and—above all—the depreciation curve of the financed compute hardware.
- Goldman Sachs and KKR would evaluate which financing structures remain syndicatable under more conservative underwriting assumptions.
- Brookfield might highlight a crucial structural distinction: whereas GPUs face rapid technological obsolescence, power grids, land, and energy assets enjoy vastly longer economic lives.
These are not statements from the executives named, but a logical continuation of a dialogue whose economic parameters have shifted within a matter of months.
The defining question posed to everyone at that table would be simple: how much of the originally targeted $500 billion would you finance on unchanged terms if Nvidia no longer provided supplemental financial backing or residual-value support?
A candid answer to that question would reveal more about the true economic durability of the AI boom than any semiconductor manufacturer’s next quarterly guidance.
The Great Differentiation
The technological trajectory of artificial intelligence and the investment returns of the companies building it are two separate matters. Economic history offers ample precedent: railways, telecommunications networks, and internet backbone infrastructure transformed entire economies, even as many first-wave investors suffered severe capital losses.
Canary Wharf in London provides a vivid illustration. In the early 1990s, original developer Olympia & York collapsed under a combination of excessive leverage, a commercial property downturn, and inadequate transport links. Following financial restructuring and vital infrastructure upgrades, Canary Wharf nonetheless grew into one of the world’s premier financial districts.
Yet even that commercial success was not immune to broader shifts in the financial system. The 2008 financial crisis brought an abrupt end to the expansion of global bank balance sheets; the collapse of structured mortgage markets stripped the private repo system of assets previously treated as pristine collateral, as AAA-rated mortgage-backed securities lost both liquidity and acceptability. The European sovereign debt crisis, tighter capital and liquidity regulations, and later Brexit and hybrid working patterns further reshaped demand for large-scale banking headquarters.
The buildings remained standing—but financing terms, tenant demand, and asset valuations underwent profound repricing. That Brookfield, alongside the Qatar Investment Authority, eventually acquired Canary Wharf Group sharpens the parallel: Bruce Flatt leads an organisation well acquainted with both the rewards and the perils of long-duration infrastructure investing.
The enduring lesson is clear: infrastructure can prove indispensable over the long term while its initial financing structure proves unsustainable.
There is, however, a critical difference between Canary Wharf and an AI data centre. Office towers can be economically repurposed over decades following a balance-sheet restructuring. GPUs and their associated technical systems, by contrast, suffer far faster technological obsolescence. It is far from certain that the economic lifespan of a GPU cluster is long enough to permit a comparable financial workout. Multi-year debt financing requires that assets either generate sufficient cash flow during their useful life or retain meaningful residual value at maturity—both assumptions carry substantial uncertainty when semiconductor architectures advance at breakneck speed.
Consequently, a profound differentiation is beginning across the AI value chain:
- Semiconductor and networking vendors remain most exposed to the pace of incremental capacity additions.
- Memory suppliers and specialised component makers benefit from rising technical intensity, yet remain inherently cyclical.
- Power, grid, and physical infrastructure operators backed by long-term contracted revenues may offer far more resilient cash-flow profiles, provided contracts and counterparties hold firm.
- Enterprise AI adopters across software and industry stand to benefit over time from falling compute costs, even if certain infrastructure investors earn disappointing returns.
The decisive fault line no longer runs simply between technology and traditional industry.
It runs between businesses whose future cash flows are largely self-sustaining and those reliant on a continuous acceleration of the capital expenditure cycle.
What Must the Future Earn Now?
In Part III of this series, we asked: “What must the future earn?” Our conclusion was that what matters is not the absolute level of debt, but the cash flow available to service it.
November’s developments give that question renewed urgency. Sophisticated financial engineering can accelerate capacity deployment and distribute risk across markets—but it cannot manufacture economic profitability if the underlying assets fail to generate adequate returns on capital.
Jensen Huang’s premise that AI compute is evolving into a productive utility still merits serious attention. Yet the pace of expansion, the capital required, and the returns ultimately realised are not identical. For investors, that divergence represents both a risk and an opportunity.
The Nelson Moment
Admiral Nelson understood that a dominant fleet required far more than well-built ships. Behind it stood dockyards, ports, supply chains, seasoned crews, and a financial system capable of sustaining operations over years of conflict. Neglect those foundations, and one could win a battle yet lose the campaign.
The AI revolution relies on a similar foundation. Its industrial and financial machinery has already been set in motion. Jensen Huang’s departure does not halt that momentum—but it may mark the point at which participants recalibrate their speed, financing terms, and risk appetite.
A more disciplined Nvidia would not necessarily signify a retreat from the technological future. It could instead represent an effort to secure the economic foundations for long-term success.
For investors, the task is not to follow a grand technological vision uncritically. It is to test continuously whether expected cash flows justify the capital committed—and to adjust course when the evidence demands it. A fleet does not prevail merely by building ever more ships; it prevails by deploying its resources so that it remains fit to fight the next engagement.