NVIDIA After Jensen: What November 17 Can and Can’t Settle

NVIDIA After Jensen: What November 17 Can and Can’t Settle

Oh boy. NVIDIA is about to report one of the most consequential quarters in its history, and the revenue number might be the least interesting part of the call.

On November 12, NVIDIA announced that Jensen Huang had resigned as CEO, president, and a director. CFO Colette Kress is serving as interim CEO, and Harvey Jones has become board chair. His exit followed reporting about a relationship with a woman suspected by U.S. officials of links to Chinese intelligence, then reports of a separate relationship with an NVIDIA executive. Those are reported claims; the company did not give a reason for his resignation. Four days before the transition, the board opened an independent investigation into whether he had been targeted by a foreign intelligence effort. It said then that it had no reason to believe national security or company secrets had been compromised. That is a statement about what it knew at the time, not the conclusion of a completed investigation. Independent directors have since retained outside counsel.

There are a lot of lurid stories here. I don’t know what the investigation will find, and I am not going to turn a question about exposure into an allegation of criminal conduct. Investors do have to price the possibility that new facts change the legal, export-control, or customer picture. They also have to price the possibility that the board closes this out without finding a compromise of sensitive information. Those outcomes are miles apart.

The November 17 results are due around 1:20 p.m. Pacific, with the call at 2. My starting assumption is that demand will remain strong. NVIDIA’s August guidance called for fiscal third-quarter revenue of $108 billion, plus or minus 2%, after a $96.2 billion second quarter. The company guided to roughly 74% gross margin and assumed no China data-center compute revenue. A beat against that baseline would be impressive. It would not answer what happens to leadership, the product roadmap, or trust.

Four ways this can trade

Strong earnings, credible clarity. This is the best near-term setup for the stock. Revenue and next-quarter guidance beat; Kress and Jones explain who owns the product roadmap, customer relationships, export compliance, and engineering retention; and the board gives a specific process and timetable for the investigation without pretending to know its outcome. If an eventual review finds no compromise, that removes a large tail risk. The market can go back to arguing about growth and valuation. I would expect NVIDIA to outperform the broader AI trade in that case, though a one-day move will still depend on what was already priced in.

Strong earnings, no new clarity. This may be the most plausible outcome for Tuesday. A serious investigation is unlikely to wrap up on an earnings schedule. If management says operations are fine but cannot give investors a credible way to judge security exposure, succession, or employee retention, a blockbuster print could meet a shrug. The stock might even fall on good numbers as the multiple absorbs a new governance discount. I would listen for whether management answers the operational questions it can answer, instead of treating every question as privileged.

Strong earnings, worsening evidence. If credible findings indicate compromised information, export-control failures, misconduct with legal consequences, or material misstatements, earnings become backward-looking very quickly. I am describing a risk scenario, not saying any of those things happened. Customer procurement reviews, government restrictions, litigation, and key employee departures could compound one another. That would hurt NVIDIA first, but the impact on the rest of AI depends on whether buyers delay projects or simply move orders to other suppliers. A $108 billion quarter would offer little comfort if the future demand curve or license to serve customers had changed.

A weaker print or guide, whatever the investigation says. If orders, gross margin, or the next-quarter outlook soften at the same time, investors will have to decide whether they are seeing a temporary transition problem or an AI spending reset. This is the scenario that would travel furthest through semiconductors, data centers, power, memory, and credit. A governance scare is company-specific until customers and competitors change their behavior. A demand miss is everybody’s problem.

Notice what I did not put in those scenarios: a price target. We do not know the market’s positioning or the evidence the board will be able to disclose by Tuesday. I have more conviction about the direction of the reaction than its size. More credible clarity and less national-security risk should help. More ambiguity or substantiated legal exposure should hurt. The operating print decides how much room the stock has to absorb either one.

The people question will outlast the call

Jensen did not write CUDA alone, design every chip, or personally install every rack. NVIDIA is an organization with a deep bench, and the board says its leadership team expects the product roadmap and customer commitments to continue. Still, the founder was a remarkably effective architect, recruiter, salesman, and public face for a complicated platform. Replacing the quarterly keynote is easy. Replacing the judgment that connected hardware, networking, software, and customers over decades is harder.

The risk I would watch is not a dramatic mass resignation next week. It is the slower loss of principal engineers and product leaders who have other places to go. An interim CEO can stabilize the business, but employees will want to know who makes technical calls and whether the investigation will keep reopening. If attrition hits architecture, CUDA libraries, networking, or field engineering, the damage would appear in development schedules and customer experience before it shows up in a quarterly income statement. NVIDIA should tell us how it is keeping those people and how decision-making works during the search.

CUDA is not a one-person product

I do not think Huang’s departure by itself breaks CUDA. NVIDIA’s toolkit includes compilers, libraries, profiling tools, and compatibility guidance. Developers have written code and built workflows around it for years. That installed base is a real switching cost. More importantly, the moat is the whole system: accelerators, interconnects, systems software, tools, and support that make a cluster productive. Someone buying a vast GPU fleet cares about working software as much as peak chip specifications.

But “CUDA wins forever” is lazy too. AMD and Anthropic announced plans for up to two gigawatts of AMD systems and joint ROCm work, with initial deployment planned for 2027. Google says its TPU business is expanding beyond its own workloads, including systems for customers’ data centers. Those are company plans, not proof that either platform has displaced NVIDIA at scale. They show large buyers have both the money and the incentive to make the software portable. Google’s TorchTPU effort is aimed at reducing the friction of using PyTorch on TPUs. If the cost of moving a workload falls, NVIDIA has to defend price and performance more often.

So CUDA is not in immediate trouble from a CEO exit. Its premium is in trouble if NVIDIA loses execution discipline while customers gain credible alternatives. I would watch developer support, product cadence, delivered performance per dollar, and how often frontier labs choose a second accelerator for production workloads. A benchmark slide is less useful than a customer repeat order.

NVIDIA spending has become part of the AI trade

The other uncomfortable question is how much demand NVIDIA is helping finance. Its latest 10-Q describes $36 billion in AI cloud commitments as of July 26, generally over six years. The arrangements let cloud partners buy NVIDIA infrastructure while NVIDIA commits to capacity that can be sold to third parties. The same filing describes prospective third-party financing platforms intended to mobilize more than $500 billion over time, subject to agreements. That is not $500 billion of NVIDIA spending, nor is every commitment a bad debt. It does mean reported chip sales and future ecosystem support have to be analyzed together.

Suppose Kress slows guarantees, equity checks, or capacity backstops. My first reaction would be positive on capital discipline. My second would be to ask which customers depended on that support to place the order. If independent lenders step in because end demand is strong, great. If orders evaporate without NVIDIA’s balance sheet nearby, the market has learned something less pleasant about marginal demand. The call should separate customer purchases supported by their own cash flows from purchases that need unusual financing. We may not get that detail, but the question belongs on the table.

This matters to the broader AI trade unevenly. A financed neocloud could feel a pullback immediately. A hyperscaler with a large operating business may buy the same compute directly or use its own chips. A utility with a signed power contract does not have the same risk as a speculative data-center developer awaiting a customer. Basket selling can ignore those distinctions for a while (markets are excellent at finding the sell button before the footnotes), but eventually cash flow separates the winners.

Are the hyperscalers in trouble?

Some of them will overbuild. I do not think they are all in trouble in the same way. Microsoft reported $45.2 billion of quarterly capital expenditure and said customer demand still exceeded capacity. Amazon reported $46.0 billion of quarterly AWS sales and $17.8 billion of AWS operating income, while its trailing free cash flow had turned negative as AI property spending rose. Alphabet reported $51.4 billion of quarterly capex and $11.1 billion of Google Cloud operating income; it also expects depreciation and data-center costs to weigh on results. These figures use different definitions and periods. They show profitable distribution engines taking on enormous investment obligations, not an industry with no revenue.

The hyperscalers can win twice: sell infrastructure to model builders, then use cheaper intelligence inside search, ads, productivity software, and retail. They can also lose twice: bid up compute today and give away tomorrow’s efficiency gains to customers under competition. The question is whether incremental revenue and margin from AI exceed depreciation, power, leases, and the cost of replacing short-lived chips. Revenue growth by itself does not answer that. Return on the next dollar of invested capital does.

I am convinced the technology is powerful. As useful models spread through ordinary workflows, I expect productivity gains to become visible in aggregate data. That can take years because organizations have to change how work gets done. It also does not guarantee that the companies buying the first generation of hardware capture most of the value. Customers may keep much of the surplus as prices fall. The model labs, software distributors, chip vendors, and cloud platforms will fight over the rest. Some excellent businesses can make disappointing stocks if investors pay today for profits that accrue somewhere else.

That is why I want two scorecards from this earnings call. First, the familiar one: revenue, gross margin, next-quarter guidance, customer concentration, and cash conversion. Second, the one forced on us by Jensen’s exit: the investigation process, security and export controls, senior-engineer retention, succession, roadmap ownership, and the economics of NVIDIA’s support for cloud customers. A strong first scorecard with a weak second one leaves the stock vulnerable. A credible second one could let an already powerful business earn back trust.

I expect NVIDIA to sell an astonishing amount of compute on Tuesday. I am less certain what multiple the market will pay for it on Wednesday.

Sources and method: Information is current as of November 14, 2026. Company figures, dates, and stated plans come from the linked NVIDIA releases and SEC filing and the linked Microsoft, Amazon, Alphabet, AMD, and Google materials. The four scenarios and expected market reactions are my analysis, not company guidance or probability-weighted forecasts. Allegations under investigation are not findings of wrongdoing.

Disclosure: This article is for education and general information, not individualized investment advice. NVIDIA and other AI-related securities carry business, valuation, governance, financing, regulatory, and market risks.