How Much of the AI Economy Rests on Nvidia?

On November 11, 2026, Nvidia's board accepted Jensen Huang's resignation as president, chief executive and director, effective immediately. The company filed the news with the Securities and Exchange Commission the next day. Colette Kress, its finance chief since 2013, is interim chief executive and stays finance chief. Harvey C. Jones, a director since 1993, became what the company calls its "first-ever Chair of the Board." The board will consider internal and external candidates for the permanent job. Nvidia gave no reason for the resignation.

Huang co-founded Nvidia in 1993 and had run it ever since, so the first question is whether Nvidia can be Nvidia without him. That is a fair question about one company. For the economy that has grown up around Nvidia, it is the smaller question.

The bigger one is in the filings. In February, Nvidia's annual report said the company had been asked to help finance its customers' data centers, and added: "We have not entered into any financing arrangements." Six months later, its quarterly report showed up to $108.5 billion of guarantees on other companies' data center leases, $105 billion of it for a campus in Ohio whose tenant is an affiliate of OpenAI. It also showed $99 billion of equity stakes in other companies, and supply commitments that had more than doubled in a single quarter, to $279 billion. Nvidia's cash and marketable debt securities came to $56.6 billion.

Most people think the risk in Nvidia's leadership change is a key person. The filings show a key balance sheet. The AI build-out now leans on Nvidia's credit as well as its chips, and the executive who signed those filings, Kress, is running the company while the board searches. That changes what the next chief executive inherits, what anyone buying AI capacity should ask before signing, and what every board can learn from a handover with no notice.

The filing is short. It reports the board's decision, the appointments and the press release, and nothing else. It says the board appointed Kress, who is 59, "to serve as interim Chief Executive Officer, while the Board conducts a search for a new Chief Executive Officer." It discloses no separation terms. Nvidia's proxy statement in May had already said that "we do not have employment, severance, or change-in-control agreements with our executive officers."

The release adds two statements. Jones, for the board: "The Board has full confidence in Colette and the leadership team, and we expect the company’s operations, customer commitments and product roadmap to continue without interruption." Kress, for herself: she is "committed to building on the extraordinary foundation Jensen created over the last 33 years." Third-quarter results will come on Tuesday, November 17, "as previously scheduled."

That is all Nvidia has said, and none of it says why Huang left. The strategic questions do not depend on the reason. They depend on what Nvidia has promised, to whom, and who now looks after those promises.

Every annual report lists what could go wrong. Nvidia's, filed in February, has a section on people, and it does not name its founder. It says: "Additionally, we are highly dependent on the services of our longstanding executive team. Failure to ensure effective succession planning, transfer of knowledge, and smooth transitions involving executives and key employees could hinder our strategic planning, execution, and long-term success." The dependence it describes is on a team.

Where Nvidia's filings do describe a dependence on a few companies, they are customers. In fiscal 2024, one direct customer bought 13% of Nvidia's revenue. In fiscal 2026, two bought 36% between them. In the first half of fiscal 2027, to July 26, three bought 44%. Nvidia does not name them. Its quarterly report adds that an unnamed "AI research and deployment company contributed a meaningful amount of our revenue by purchasing cloud services from our customers."

Huang described the trend differently in August. "This time last year, one lab alone was driving the buildout; today, we have a golden age of new AI labs and startups, multiple frontier labs scaling in parallel," he said in the results release. Both can be true. The labs that use the chips may be multiplying while the share bought by Nvidia's largest direct customers grows. For Nvidia, that means bigger customers, each carrying more weight in its revenue. For the economy, it means a supply chain concentrated on both sides of the chip.

The February report already showed a small step in this direction. Nvidia had guaranteed some AI cloud companies' data center leases, with a maximum exposure of $3.5 billion that shrinks as the tenants pay their rent over five to seven years. And it said: "We have entered and may in the future enter into commercial arrangements, including long-term capacity purchase obligations and financial guarantees, and have been asked to offer financing arrangements to support our customers’ and partners’ buildout of datacenter infrastructure. We have not entered into any financing arrangements."

On August 17, Nvidia filed a current report on a campus in Pike County, Ohio. It had agreed to stand behind leases for about 4.25 gigawatts of computing load there, owned by SB Energy, with an affiliate of OpenAI as the tenant. "NVIDIA’s aggregate payment obligation is cumulatively capped at $105 billion for its initial commitment under the Agreements." If OpenAI's affiliate becomes insolvent or stops paying, Nvidia "will pay an amount generally equal to any shortfall between the guaranteed minimum value of a lease (as set forth in the Agreements) and amounts recovered through a replacement lease or sale."

The quarterly report nine days later set out the rest. The guarantees switch on phase by phase, as each of nine phases of construction is finished, the first in fiscal 2029. They cover "defined portions of lease and power payments and not the full cost of the site or all of the tenant’s obligations." They end when the lease terms run out, or upon "OpenAI achieving a satisfactory credit rating." And they came with a condition in Nvidia's favor: "In exchange for the guarantees, the site will exclusively host NVIDIA AI infrastructure, subject to limited exceptions." Nvidia also holds an option to back about 3.8 more gigawatts.

Consider what ends the guarantees: the leases running out, or OpenAI's own credit being judged strong enough to stand alone. Until then, part of the tenant's credit is Nvidia's. OpenAI has agreed to reimburse Nvidia for anything it pays, and the quarterly report adds the caveat that matters: "we may not recover amounts promptly or in full." By then the company's description of itself had changed too: "To support our customers’ and partners’ buildout of AI infrastructure, we enter into commercial arrangements, including financial guarantees and other forms of credit support, financing arrangements, and data center leases."

None of this is hidden. It is disclosed, capped and conditional, and Nvidia's obligations in Ohio start only as the sites are ready, expected from 2028. But it changes what Nvidia is. A chipmaker sells chips. A chipmaker that backs its customer's rent in exchange for an exclusive site is also, in part, an underwriter of its customers' credit.

Put the quarterly report's figures side by side. Nvidia held $56.6 billion of cash and marketable debt securities. Its maximum exposure on guarantees was $108.5 billion: the $105 billion for Ohio, signed in August, and $3.5 billion of older guarantees for AI cloud companies. It held $99 billion of equity investments, already paid for, and had committed $25 billion more. It had agreed to buy $36 billion of computing capacity from the cloud companies that bought its systems, and $279 billion of supply and capacity. Its own table of future commitments, which leaves out the guarantees, came to $366 billion, about 6.5 times its cash.

Each line has a business reason. The supply commitments, up from $119 billion a quarter earlier and $95.2 billion in January, are "primarily memory and manufacturing facilities" for products Nvidia expects to sell. The equity stakes, Kress wrote in August, "are focused on AI model makers, infrastructure financiers, and other private companies." In fiscal 2026 alone Nvidia put $17.5 billion into private companies and funds, and its annual report says: "These investments include AI model makers that purchase our products directly or through CSPs."

The cloud agreements show how the pieces connect. "AI clouds procure our data center infrastructure products and we commit to cloud service agreements, which the AI clouds can unilaterally stop providing to us and sell to third-party customers at more advantageous rates." And if they cannot sell it: "we have agreed to purchase that capacity." Nvidia sells the systems, stands ready to rent the output back, holds stakes in the model makers that rent it, and now guarantees one customer's lease. Its receivables tell the same story in a smaller way: $63.1 billion in July, 60 days of sales, up from 45 a quarter earlier, "due to extended payment terms on large, multi-quarter agreements with certain investment-grade customers."

This is not a company short of money. In the quarter to July 26 its revenue was $96.2 billion, up 106% on a year earlier. It raised $25.0 billion of senior notes in June and returned about $26.0 billion to shareholders in the quarter. Nor are the commitments debts. Most are purchase obligations that fall due over several years and are paid from the sales of the products they buy; at July's pace of sales, $366 billion is less than four quarters of revenue. The point is where the risk now sits. A slowdown at Nvidia's customers would no longer reach it through one door, lower orders. It would come through several: unsold capacity Nvidia must buy, stakes that lose value, receivables that take longer to collect, and guarantees that are called. The quarterly report lists the risk in its own words: "Customers or partners may fail to fulfill their financial commitments, secure necessary capital or infrastructure, complete projects on schedule or within budget, or experience financial distress or insolvency."

Kress joined Nvidia in September 2013, "after serving nearly 25 years in a range of finance roles at major technology companies." She was finance chief of Cisco's Business Technology and Operations Finance organization for three years, and spent 13 years at Microsoft, four of them as finance chief of its server and tools division. She signed the August 17 current report on the Ohio guarantees and the quarterly report that followed.

Her own commentary on that quarter explained the strategy in two sentences: "Securing land, power and shell for data centers has become the next critical phase in the AI infrastructure buildout. AI clouds and model makers are seeing extraordinary demand for AI infrastructure, yet many are growing faster than their balance sheets and long-term credit profiles can support." That is the case for the guarantees, in the words of the executive now running the company. Nvidia's customers want more computing than their credit can carry; Nvidia's credit fills part of the gap; Nvidia gets the business.

There is continuity in that. Few people know these commitments as well as the executive who signed the filings that disclosed them, and the board has said it expects "customer commitments" to continue without interruption. There is also a concentration. Until the board names a permanent chief executive, the person who sets Nvidia's strategy and the person who reports on its balance sheet are the same person. Research on interim successions gives a reason to want that period short, not a forecast for this one: Ballinger and Marcel found that interim chief executive arrangements "are associated with lower performance during the period in which the interim serves," an average across many companies. The filings do not say whether Kress is a candidate for the permanent job.

On the other side of the trade, the five companies at the center of the build-out are spending at twice last year's pace. In their latest quarters, Microsoft, Alphabet, Amazon, Meta and Oracle together spent $230.5 billion on buildings and equipment, against $105.8 billion in the same quarters a year earlier. Amazon still expects about $220 billion of cash capital spending this year. Meta raised its 2026 range to $135 billion to $145 billion, from $115 billion to $135 billion in January. Oracle expects fiscal 2027 to be higher than fiscal 2026, and Alphabet expects to "significantly increase" its investment this year.

Nvidia's name does not appear in the latest annual and quarterly reports of Microsoft, Alphabet, Meta or Oracle. Amazon's mention it once, as one of the companies named in a patent complaint. The buyers describe the dependence in general terms instead. Microsoft's capacity depends on "servers, including graphics processing units and other components," and "some of our products contain certain components for which there are very few qualified suppliers." Alphabet's supply of "specialized AI chips, is limited to a small number of qualified suppliers." Oracle has "significant commitments with suppliers of chips and other data center infrastructure."

One step down the chain, the dependence has a name. CoreWeave, which rents out computing to AI companies and in which Nvidia bought $2.0 billion of stock in January 2026, says that "all of the GPUs used in our infrastructure today are NVIDIA GPUs" and that "our current customers have contractually specified our use of NVIDIA GPUs." The biggest buyers can spread their bets and describe a category of supplier. CoreWeave is tied to a single supplier by its customers' contracts.

The buyers also build and buy alternatives. Alphabet sells its own chips: "Google Cloud generates product revenues primarily from the sale of TPU systems," and its latest design, Ironwood, is its "seventh-generation TPU." Amazon's arrangement with OpenAI "includes the use and performance of AWS chips," and the two added $100.0 billion to it in early 2026. Microsoft offers "custom-built silicon and strong partnerships with chip manufacturers."

Among chip sellers, AMD is the clearest rival. Its data center revenue was $8.5 billion in the quarter to September 26, 2026, up 97% from $4.3 billion a year earlier. In October 2025 it agreed with OpenAI to deploy 6 gigawatts of its GPUs, and it has issued warrants to OpenAI and Meta for up to 320 million AMD shares.

The alternatives are growing fast. So is the incumbent. Nvidia's data center revenue was $89.0 billion in the quarter to July 26, up 117%, about ten times AMD's data center revenue in its latest quarter, a segment that also includes server processors. The alternatives give buyers a bargaining position and a fallback. In 2026 they do not yet give the build-out a second foundation of the same size.

The dependence does not stop at the companies. By the Bureau of Economic Analysis's advance estimate, real GDP grew at an annual rate of 3.4% in the third quarter of 2026, and investment in information processing equipment and software contributed 1.09 percentage points of it, about 32%. In the fourth quarter of 2025, when the economy grew at a rate of just 0.2%, the same investment contributed 0.72 of a point, more than all of the growth. That investment reached 5.35% of GDP in the third quarter, up from 4.75% in the fourth quarter of 2025.

Three cautions belong with those numbers. The category covers all business spending on computers and software, not only AI. The contributions are gross: much of the equipment is imported, and imports are subtracted elsewhere in the same accounts, so the net effect on growth is smaller. And the accounts do not say whose chips the equipment contains. They measure the kind of investment Nvidia supplies, not Nvidia's share of it.

Investors have rewarded the supplier most. From January 3, 2023 to November 11, 2026, the last close before Nvidia announced the change, Nvidia's total return with dividends reinvested was 5.82 times that of the same money in the US market, ahead of AMD and of all five big buyers, which ranged from well ahead of the market to behind it. That is history, not a forecast. Whatever happens to Nvidia's leadership, a large share of recent US growth has come from the kind of investment it supplies.

There is a famous case of a company that seemed to be one person. On August 24, 2011, Apple's board announced that Steve Jobs had resigned as chief executive and named Tim Cook, then chief operating officer, in his place. Like Nvidia this week, Apple gave no specific reason; Reuters reported that Jobs resigned "without specifying a reason." In his letter, Jobs wrote: "I strongly recommend that we execute our succession plan and name Tim Cook as CEO of Apple." Cook had already run the company as interim chief executive in 2004 and for the first half of 2009.

The reaction was mostly supportive. One analyst told Reuters that day: "From a succession perspective they could not possibly identify a better candidate." Barclays' Ben Reitzes called Cook "a proven executive who can handle the pressure and knows how to run the inner workings of Apple in Jobs’ shadow." The doubt was narrow and specific. Macworld reported that "some analysts have questioned whether—despite his undergraduate design degree and years at Apple—he has the design chops that Jobs brings to the table," and recalled a 2009 remark: "Tim Cook’s the guy who makes the trains run on time." Apple's net sales were $108.2 billion in fiscal 2011 and $365.8 billion in fiscal 2021; net income went from $25.9 billion to $94.7 billion.

The lesson is not that every doubted successor succeeds. It is that the doubts measured Cook against Jobs's strengths, design above all, rather than against the job in front of him. As chief operating officer, Apple's release noted, Cook had been responsible for "all of the company’s worldwide sales and operations, including end-to-end management of Apple’s supply chain." The doubts were about design. His record was in operations.

Apple did it again this year, also by plan. On April 20, 2026, it announced that Cook would become executive chair on September 1 and that John Ternus, its head of hardware engineering, at Apple since 2001, would succeed him. That gave more than four months' notice, with the departing chief executive staying on the board.

Nvidia's case differs in the ways that matter. Its handover was effective immediately. Kress is interim, not a named successor. The departing chief executive left the board, and the board appointed its first-ever chair the same day. No succession plan has been disclosed. So the Apple comparison supports one point only: do not judge Nvidia's next chief executive by how closely he or she resembles the last one. It predicts nothing about who that will be or how Nvidia will fare.

Whoever takes the job will be judged on chips. The bigger decisions in the first years may be about money. The Ohio guarantees begin with the first construction phase, expected in fiscal 2029, and run with 20-year leases. Of the $279 billion of supply commitments, $92 billion falls due in the rest of fiscal 2027, $87 billion in fiscal 2028 and $88 billion in fiscal 2029. The deal to buy Hugging Face, for about $11.9 billion, is expected to close in the first half of 2027, under a pledge that the platform will continue "to support other silicon vendors."

Three questions will come with the job. How much of its customers' credit risk should Nvidia keep carrying? In August it announced partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR "to mobilize over $500 billion of third-party capital for the buildout of AI infrastructure over time, subject to definitive agreements," and its quarterly report warns: "These and other preliminary arrangements may not lead to definitive agreements." Should credit keep being traded for exclusivity, as in Ohio? And how should Nvidia plan for the markets it cannot sell into? Its outlook for the current quarter assumes no data center computing revenue from China, and its filing warns that "it is likely that additional unilateral or multilateral controls will be adopted."

For 33 years, those calls sat with a founder who also sat on the board. The next chief executive will make them alongside a board that has just appointed its first-ever chair. That is not a reason for alarm. It is a reason to put the rules on paper.

Nvidia reports on November 17. In August it guided to revenue of $108.0 billion, plus or minus 2%, for the quarter ending in October. The revenue line will get the headlines. The more useful lines are further down: whether the guarantees, the equity stakes, the capacity buy-backs and the supply commitments have grown again since July, whether any of the financing platforms has become a signed agreement, and how long customers are taking to pay.

It will also be the first time Nvidia reports results with one executive in both jobs. That is an opportunity. The clearest signal of continuity the company could give is not a statement of confidence. It is a complete account of its promises.

Put every promise in one table on November 17. Guarantees, equity investments and commitments, cloud buy-backs, supply commitments and receivables, each against its July figure and its cap. A statement that customer commitments will continue is a promise. A table of them is evidence.

Write the limits into board policy. Set a ceiling for guarantees and equity commitments, measured against cash and cash flow, and give it to a board committee to watch. The rules for lending Nvidia's balance sheet should not depend on who holds the chief executive's title, least of all during a search.

Separate the two jobs soon. An interim chief executive who is also the finance chief sets the strategy and reports on its cost. Naming an acting finance chief, or giving a deputy clear authority over the commitments, would restore a check that the old structure had.

Move customer credit to the platforms. The financing partnerships announced in August are the right destination for most of this risk: capital providers built to hold it, rather than a chipmaker. Report what has moved, and what has not.

Say what the search is for. Apple's boards had a plan before they needed one. Nvidia's board should say what the next chief executive is being hired to do, product, finance or both, and roughly when it expects to decide. A search without stated criteria invites the market to supply its own.

If you buy AI capacity, find out whose credit stands behind it. Ask your provider whether its capacity is backed by a supplier's guarantee or buy-back, and whether its contracts tie it to one chip family, as CoreWeave's do. Price the option to move between chips into the contract now, while you still have alternatives to point to.

If you supply the build-out, in power, construction, real estate or finance, read the guarantee, not the press release. A lease backed by Nvidia's credit support is a different instrument from one backed by the tenant alone. Know its cap, when it switches on and what ends it: a credit rating, a lease term, a default. Each changes what you are really owed.

If you sit on a board, ask two questions this quarter. Could your company hand over in a day, with a plan the board has already written down? And if your company lends its balance sheet to customers, through guarantees, stakes or long payment terms, is the limit written into board policy, or does it live in one executive's judgment?

We read Nvidia's 8-K and press release of November 12, 2026, its annual report for the year to January 25, 2026, its quarterly report and results for the quarter to July 26, 2026, its CFO commentary, its 8-Ks on the Ohio guarantees, its senior notes and the Hugging Face agreement, and its 2026 proxy statement. For the buyers, we read the latest annual and quarterly reports of Microsoft, Alphabet, Amazon, Meta and Oracle, and searched each for Nvidia's name. Capital spending is purchases of property and equipment from SEC XBRL data; quarters are year-to-date figures less the quarter before. We also read CoreWeave's and AMD's latest quarterly reports.

Economic figures are the Bureau of Economic Analysis's advance estimate for the third quarter of 2026, via FRED. Market performance is total return relative to SPY, a fund that tracks large US companies, from Tiingo adjusted closes; we show ratios, not prices. Apple's 2011 handover comes from its release, Jobs's letter and same-day coverage by Reuters, Macworld and Fortune; its 2026 handover from its 8-K. The study of interim chief executives is Ballinger and Marcel in the Strategic Management Journal.

What we could not see: why Huang resigned, which Nvidia did not say; who Nvidia's largest customers are; how much of the buyers' spending goes to Nvidia; and Nvidia's share of the investment in the national accounts. Nvidia's figures are as of July 26, 2026 and will be updated on November 17.

A chief executive can leave in a day. A guarantee lasts as long as the lease. When you weigh a company's key person, weigh its key promises too.