Nvidia after Huang: where AI and its supply chain go next, and why small businesses stand to gain

As of November 12, 2026, early afternoon ET. This article is educational. It is not a recommendation to buy, sell or hold any security.

On November 11, Jensen Huang resigned as President, CEO and a director of Nvidia, the company he co-founded in 1993. The board accepted his resignation effective immediately. Colette Kress, Nvidia's CFO since 2013, is interim CEO and remains CFO. Harvey Jones, a director since 1993, was named what Nvidia's press release calls the board's "first-ever Chair", and the board says its search will consider internal and external candidates. (Nvidia Form 8-K, filed November 12, 2026, reporting events of November 11; press release, November 12.)

The news hit hardest in the middle of the AI economy: the chipmakers, GPU clouds and suppliers that build the machinery. But the bigger story is where all that machinery is going. Very large sums are being spent to build AI capacity. In our view, the companies paying for it are not the ones who will feel most of the benefit, and some of the biggest winners of the AI buildout are at the other end of the chain: small business owners, who can rent capabilities that no small company could ever afford to build.

This article starts with where AI and its supply chain are heading and why small businesses stand to gain, then turns to Nvidia itself: what happened, what it means for the companies around it, and who could lead it next.

Where AI is heading: from building it to using it

The first years of the AI boom were about training: building ever-larger models in huge data centers, mostly on Nvidia GPUs. The next phase is increasingly about inference: running those models millions of times a day to answer questions, draft documents, sort email, flag fraud and handle customer requests. Training happens in a few places. Inference happens everywhere AI is used.

That shift changes the supply chain in ways that matter more than any one CEO:

• More kinds of chips. Training has rewarded Nvidia's general-purpose GPUs and its CUDA software. Inference rewards speed and cost per answer, which opens room for others: AMD; custom chips designed with Broadcom; Google's TPUs, Amazon's Trainium and Microsoft's Maia; and specialists such as Cerebras. In July 2026, for example, CrowdStrike announced it would run its Falcon AI Detection and Response models on Cerebras's inference hardware, and Cerebras would use CrowdStrike's Falcon platform to secure its own operations (CrowdStrike press release, July 22, 2026).
• More capacity. Cloud providers, neoclouds and data-center builders are adding capacity on multi-year plans. Chip programs are set years ahead, and Nvidia's own roadmap has been reported as set through 2028. The scale is striking. Anthropic's IPO filing reports about $900.9 billion of committed purchases and leases as of and after September 30, 2026, substantially all of it for compute and infrastructure, running on Amazon's Trainium, Google's TPUs, Nvidia GPUs and AMD GPUs. Of that, about $451.6 billion is expected before the end of 2030, and substantially all the rest between 2031 and 2045. The total also leaves out other compute arrangements with no fixed dollar amount, which Anthropic estimates at a further $150 billion to $175 billion.
• Power is the next bottleneck. Chips can be made faster than power plants, grid connections and turbines can be built. More and more often, an AI buildout waits on electricity, not on silicon.
• More competition in most layers. There are now several chip makers, several large clouds and several leading model providers. A few chokepoints, such as leading-edge manufacturing and high-bandwidth memory, still have few suppliers. For the businesses at the end of the chain, more competition among suppliers is generally good news.

The AI supply chain, layer by layer

Nvidia designs its chips; it does not make them. Its products depend on a long, specialized chain:

• Contract manufacturing: leading-edge chips are fabricated by a foundry partner, primarily TSMC.
• Advanced packaging: joining the GPU to its memory has been a capacity bottleneck in its own right.
• High-bandwidth memory (HBM): a few memory makers supply it, and capacity has typically been booked well ahead.
• Networking and systems: NVLink, switches and full rack-scale systems ship with each generation.
• Power and data centers: the facilities that host the chips need large power connections with long lead times.

Those long lead times favor continuity. As one commentator noted on CNBC, a new CEO does not need to make fast decisions on chip plans. They also limit how quickly anyone can switch suppliers: Futurum's analysis, as reported, is that AMD could not absorb a large shift of demand quickly, because supply, not interest, is the limit.

Seen from the top down, the whole chain forms a stack:

Three points stand out:

• The biggest customers are also rivals. Alphabet, Microsoft and Amazon are widely reported to be among Nvidia's largest buyers, and all three design their own AI chips. Customers that can build their own may find it easier to shift. Anthropic's IPO filing shows how tangled this has become. Alongside Nvidia's own commitment of about $10.0 billion (covered in the neocloud section below), its $26.0 billion private placement includes Google at about $10.0 billion, AMD at about $2.5 billion, Broadcom, Micron and Microsoft at about $1.0 billion each, and TSMC Partners at about $500 million. Nvidia's chip rivals, its customers and its own suppliers are buying into the same AI lab. Google's and AMD's purchases partly fulfill investment commitments they had already made to Anthropic.
• The money flows down the stack. It starts as small monthly bills for apps, AI assistants and security. App vendors and model providers pay the clouds, the clouds turn that into large, lumpy orders for chips and servers, and the bottom of the stack turns it into multi-year power and lease contracts.
• The risk flows the other way. The large, lumpy bets sit in the middle and at the bottom of the stack. The top of the stack, where most businesses sit, pays as it goes.

Small business owners: among the biggest beneficiaries

Most small businesses will never buy anything from Nvidia. They buy AI the way they buy other software: by the seat, in a monthly subscription, or by the request through an AI provider's API. In our view, that makes them strong beneficiaries of the buildout.

• Someone else pays for the factories. The cloud providers, model labs and their suppliers carry the capital spending, the debt and the risk. A small business rents the result for a monthly fee and can cancel it.
• AI arrives inside tools they already use. Google Workspace includes Gemini features, and Microsoft 365 includes Copilot Chat, with fuller Copilot features sold as a paid add-on. Direct subscriptions such as ChatGPT Business are another route. Payroll, bookkeeping, payments, point-of-sale and website tools are building AI into the workflows owners already pay for.
• Competition works in their favor. In our view, competition among chip makers, clouds and model providers, and the shift toward faster, cheaper inference chips, should push the cost of running AI down over time. How much of that reaches a small business depends on its vendors, some of which have strong pricing power where they own a workflow such as payroll or payments.
• Small teams can do more. Drafting proposals, answering routine customer questions, reconciling books and summarizing documents are tasks AI can now help with. For a business with a handful of people, that extra capacity can matter more than it does at a large company.
• One step removed from the turmoil. A leadership change at Nvidia lands first on the companies that buy and sell GPUs, not on the apps a small business uses every day.

There are real caveats:

• The main risk is indirect. If cloud providers slow their spending on AI infrastructure, new AI features may arrive later or cost more.
• AI makes mistakes. Anything it drafts for a customer, the books or a tax filing needs a person to check it.
• Prices can rise. AI features are often sold as paid add-ons or higher tiers, and vendors that own a workflow can raise prices.
• Security matters more. More AI use means more data moving through more tools, and more for attackers to target.
• Where the data goes matters. Before turning on an AI feature, it is worth knowing what the vendor keeps, for how long, and whether your data is used to train its models.

For a business owner, a few practical questions:

• Which AI features are already included in the software you pay for?
• What does each vendor do with your data?
• Where would you be locked in if a vendor raised prices?
• Which tasks in your business are repetitive enough to hand to AI, with a person checking the result?

What happened at Nvidia: what we know, and what we don't

Confirmed in Nvidia's own filings and releases:

• On November 8, Nvidia's board said it was "initiating an independent investigation in light of recent reporting that founder and CEO Jensen Huang has been the target of a foreign intelligence effort", overseen by a committee of independent directors with outside counsel. The board said it had "no reason to believe that either national security or NVIDIA secrets were compromised."
• On November 10, the independent directors said they had retained Joon H. Kim, a former acting U.S. Attorney for the Southern District of New York, and his firm, Wachtell, Lipton, Rosen & Katz, to conduct the investigation.
• Huang resigned on November 11, and the board accepted the same day.
• Kress is interim CEO and CFO, Jones is Chair, and a CEO search is under way.
• In the release, Jones said the board expects "the company's operations, customer commitments and product roadmap to continue without interruption." The release says nothing about financial guidance, and it does not disclose severance terms or name a search firm.
• Third-quarter results are still scheduled for Tuesday, November 17, with a call at 5 PM ET.

Reported, but not confirmed by the company:

• In a November 4 letter, the chairs of three House committees, Brian Mast, John Moolenaar and Rick Crawford, asked Huang for a joint closed-door briefing on November 18 about his ties to a Chinese Nvidia customer (House Select Committee on the CCP press release). We have seen no confirmation that Huang accepted, or that the date holds.
• A press report cites billions of dollars of chip shipments through Chinese companies. We have not read the full report.

We separate these on purpose. The first list is what the company has put on the record. The second could change.

Neoclouds: when your supplier is also your financier

A neocloud is a company that rents out AI computing power, built mostly on Nvidia GPUs, to model developers and enterprises. Neoclouds compete with the large cloud providers, but they are newer and far more specialized, and they tend to be heavily financed. Well-known examples include CoreWeave, Nebius and IREN.

Nvidia has done more than sell these companies chips:

• Investor. Schedule 13G filings, as reported, show Nvidia holding about 11.5% of CoreWeave and about 9.3% of Nebius. The Nebius figure includes a pre-funded warrant.
• Customer. It has a reported order to buy unsold CoreWeave capacity through 2032, and a multi-year cloud contract with IREN.
• Financial backstop. By agreeing to buy capacity, Nvidia makes it easier for these companies to borrow against their contracts and keep building.

Reported counts of Nvidia's commitments vary widely by method. Reuters counts about $165 billion. Yahoo Finance counts about $530 billion, because its figure also includes guarantees and multi-year purchase deals. Either way, Nvidia has become a meaningful financier of the AI buildout, not only its main supplier.

What a CEO change does, and doesn't, do to this web:

• Signed agreements survive. The CoreWeave capacity order, as reported, can be ended for material breach or insolvency. A change of CEO is neither.
• The next deal is the open question. Decisions on new investments, new backstops and who gets scarce GPUs now sit with an interim CEO and a board that is running an investigation. The Information has reported, citing people with knowledge of the discussions, that Kress pushed back on a proposed $250 billion Nvidia credit backstop for OpenAI. In August, according to that report, Nvidia split the deal in two and has so far guaranteed only the first part, about $100 billion of credit support. The same report says Kress has worked to keep Nvidia's financial risk from concentrating in a handful of companies. We have not read The Information's article itself; these details are as relayed in other coverage.
• One new commitment is already on paper. Anthropic's amended IPO registration statement, filed with the SEC on November 10, says Nvidia has agreed to buy about $10.0 billion of Anthropic's non-voting Class N stock at the IPO price. It is part of a $26.0 billion private placement scheduled to close immediately after the offering. The placement needs Anthropic's stockholders to approve it, which Anthropic says it expects to obtain before closing, but the IPO itself does not depend on the placement closing.
• It isn't only Nvidia. Under a July 2026 hardware supply agreement, Anthropic buys or accesses AMD's Helios rack-scale systems, and AMD committed to buy up to $5.0 billion of Anthropic's Class N stock, available in steps as AMD GPU delivery milestones are met. AMD's $2.5 billion in the placement satisfies part of that commitment. A chip seller funding its own customer is the same pattern as Nvidia and the neoclouds (Anthropic S-1/A).
• Not every drop this week is about Nvidia. Nebius's drop was driven mostly by its own results and capital spending guidance: it raised its expected 2026 capital spending to $27 billion, from a previous range of $20 billion to $25 billion, and said it expects 2026 revenue at the low end of its range (CNBC, November 12).
• The financing is not all on Nvidia. Nebius's shareholder letter puts 2027 capital spending at $40 billion to $45 billion and says prepayments and contract financing cover 75% to 85% of it. That is customer money committed in advance, a partial answer to the worry that this circular financing could suddenly unwind. It is only partial, because some of the customers prepaying are themselves funded or backstopped by Nvidia or the large cloud providers.

The broader lesson is about circular financing. When a supplier funds its own customers, sales and financing can reinforce each other on the way up. They can also unwind together if the supplier pulls back.

Who is exposed, and how

It helps to sort the companies around Nvidia by what ties them to it. The tie decides what moves them.

The groups overlap, and the ties differ even within a group. Microsoft and Alphabet sit in several camps at once: they sell apps to businesses, and they also buy chips at scale for their clouds. For example, Google, not Nvidia, has reportedly backstopped the lease payments of a key TeraWulf tenant.

What the market did on the morning of November 12. In morning trading, as Nvidia fell, shares of two chip rivals, AMD and Broadcom, rose. That is a few hours of trading, not a trend, and moves like these can have several causes.

Washington and China: what could change

Nvidia's leadership change comes at a busy time in Washington, and in talks between Washington and Beijing. Here is what the public record shows as of November 12, and what it does not.

Company vs. person: what is and is not established

• Nvidia's 8-K gives no reason for Huang's resignation.
• The board's independent investigation concerns reporting that Huang "has been the target of a foreign intelligence effort." The board said it had "no reason to believe that either national security or NVIDIA secrets were compromised" (per the Nvidia board statement of November 8).
• We found no 2026 penalty, charge, license revocation or Entity List action against Nvidia as a company (per our review of DOJ and BIS releases and Nvidia's SEC filings). The 2026 chip-diversion cases charge individuals and intermediaries, and the DOJ releases do not name Nvidia (per DOJ, March 19 and October 1).
• Three House committee chairmen asked Huang for a closed-door briefing on Nvidia's security protocols against foreign espionage and on his contacts with an executive of a Chinese customer (per their November 4 letter). CNBC reported that the date is still being negotiated. No official body has made any finding of wrongdoing.

Policy tools and pending legislation

• Commerce can suspend or revoke an export license "without notice" (per 15 CFR 750.8). A civil penalty requires notice and a hearing first (per 50 U.S.C. 4819).
• None of the chip bills is law. The Remote Access Security Act passed the House 369-22 on January 12 and sits in Senate Banking (per the House Clerk and govinfo). The Chip Security Act and the AI OVERWATCH Act cleared House Foreign Affairs 42-0 and 42-2 but have had no floor vote (per govinfo bill status). The AI OVERWATCH Act would give Congress 30 days to review advanced-chip licenses to China (reported by Axios).
• Chairman Moolenaar called for passing export-control legislation "in the year-end national defense bill" (per his November 12 statement). The Senate has not passed its version of that bill (per govinfo). Reports differ on whether chip provisions will survive the House-Senate negotiations.
• Unfinished bills lapse on January 3, 2027. The AP has projected that Democrats will control both chambers in the new Congress (reported by AP via Spectrum News).

Remote and cloud access

• No U.S. rule bans a Chinese company from renting chips in a data center outside China. Current controls apply to the export of chips. A January rule makes screening of remote users a condition of H200-class licenses (per the Federal Register, January 15).
• One outlet, The Information, reported on November 11 that Commerce is close to finalizing a rule on remote access through data centers in other countries, including Malaysia, Singapore and Thailand. Commerce has made no public statement, and we have not seen the rule.

China's levers and the summit

• China extended its pause on the rare-earth export controls it announced in October 2025 to January 10, 2027 (reported by Shanghai Metals Market and Skillings, November 9). The controls it imposed in April 2025 remain in force (same sources).
• APEC leaders meet in Shenzhen on November 18 and 19 (per the Shenzhen government, citing Xinhua). The White House said both leaders intend to attend each other's summits (per the White House fact sheet, September 25). A Trump–Xi meeting there is expected (reported by The National).

How much China matters to Nvidia's revenue

• Customers headquartered in China, including Hong Kong, accounted for $19.7 billion of $215.9 billion in fiscal 2026, about 9%. In fiscal 2024 the share was about 20% (per Nvidia's 10-K; percentages are our own calculation). In our view, the decline reflects U.S. export restrictions on Nvidia's China sales far more than weaker demand. In the most recent reported quarter it was $7.9 billion of $96.2 billion, about 8% (per Nvidia's 10-Q).
• These figures count where the buyer is headquartered, not where the chips are used (per the 10-Q). Nvidia says it is "effectively foreclosed" from China's data center market. Licensed H200 shipments were under 1% of Data Center revenue (per the same 10-Q).

What it means for small businesses

In our view, most of these decisions land on chipmakers, cloud providers and large AI labs, not on the businesses that buy AI services from them. The things worth watching are slower: whether tighter export rules or a rare-earth dispute after the January 10 deadline raise the cost of building AI capacity, and with it the price of the tools small businesses rent.

CUDA: does the moat depend on the founder?

CUDA is Nvidia's software platform for programming its GPUs. Many investors consider it Nvidia's deepest competitive advantage. It is close to twenty years of code, libraries and developer habits:

• Installed base: hundreds of millions of CUDA-capable GPUs, per Nvidia.
• Developers: Nvidia says about 6 million developers build on CUDA. That is a company figure, not an audited one.
• Libraries: more than 400 CUDA-X libraries, such as cuDNN, NCCL and TensorRT, that AI frameworks rely on.

None of that leaves with one person. Ian Buck, whom Nvidia credits with creating CUDA, is still at Nvidia.

The real long-term questions are different:

• Portability is rising anyway. Tools such as Triton, now the default path in PyTorch's compiler, and framework support for AMD's ROCm have been lowering switching costs for some time. That trend predates the leadership change, and it is part of why buyers further up the stack have more choice.
• Investment pace. CUDA stays ahead because Nvidia keeps spending heavily on it. Whether a new leader keeps that pace is a fair question.
• Talent. Some analysts, as reported, have flagged the risk of senior engineers leaving once retention awards vest or a new CEO arrives.

For Nvidia itself: earnings versus valuation

When a company loses its founder, it helps to ask two separate questions:

• Is the business itself affected? Analysts publishing this week have, as reported, largely left their earnings estimates unchanged. Their reasons, as reported: chip programs take years, the roadmap is set, and customer contracts are signed with the company, not with its CEO. One business risk remains open: if the export questions turn out to involve the company, not only Huang, penalties or license limits could affect sales.
• What will investors pay for those earnings? Most revisions reported this week lowered the valuation multiple analysts apply to Nvidia's earnings, rather than the earnings themselves. In a November 6 note, before the leadership news, Bank of America valued Nvidia at 22 times its calendar 2027 earnings estimate (excluding cash). After Huang's resignation, Stifel cut the multiple it applies to fiscal 2028 earnings from 20 times to 13 times, as reported by Yahoo Finance and Investing.com. "The issue is the multiple," Stifel's note said, according to Yahoo Finance. A founder's exit with no announced succession plan, plus reported congressional questions about a Chinese customer, weighs on confidence. Melius Research, as reported, pointed to a precedent: Hewlett-Packard's leadership turmoil in 2010, after which a lower valuation lasted for years.

The succession: the top five, and how deep the bench really is

The board says it will consider internal and external candidates, and it has not named a timeline or a search firm. Below are the five names ranked highest by the average of two prediction markets, Polymarket and Kalshi, as of about 12:30 PM ET on November 12, counting names listed on both. These markets are thin and move quickly. Analysts mostly discuss the first three.

1. Ian Buck (insider). Vice president of Hyperscale and HPC, and head of product for hyperscale AI. He joined Nvidia in 2004 after a Stanford PhD, and Nvidia calls him the inventor of CUDA. He works directly with Nvidia's largest customers, including Microsoft, Google, Amazon, Meta and OpenAI, and has testified to Congress on AI.

• Strengths: unmatched technical credibility inside Nvidia, and direct continuity for CUDA and the software moat.
• Gaps: he is not one of Nvidia's executive officers, and there is no record of him running company-wide finances, supply or sales.

2. Colette Kress (insider, interim CEO). CFO since 2013, and now interim CEO as well. Before Nvidia, she held finance leadership roles at Cisco, Microsoft and Texas Instruments.

• Strengths: she is already in the seat, so there is no transition gap. Investors know her from every earnings call, and she knows Nvidia's customer commitments and financing in detail.
• Gaps: she has no engineering background, which some analysts say limits her long-term case. She is doing two jobs during an earnings season and a board investigation. She approved much of the financing web now under scrutiny, though she reportedly pushed back on its largest piece, the proposed OpenAI backstop. One analyst, Dan Ives, expects her to serve about a year before a permanent pick from outside.

3. Rene Haas (outsider, former Nvidia executive). Arm's CEO since 2022. He spent about seven years at Nvidia, running its Computing Products unit, before joining Arm in 2013. He also holds several SoftBank roles.

• Strengths: a sitting public-company CEO in semiconductors who knows Nvidia's culture and people. He is the outsider analysts cite most often.
• Gaps: he would have to unwind his Arm and SoftBank roles and would likely forfeit large unvested equity, so hiring him would be costly. Arm licenses chip designs, a very different business from Nvidia's.

4. Tim Cook (outsider). Apple's Executive Chairman since September 1, 2026, after 15 years as CEO. President Trump publicly suggested him for the job.

• Strengths: the best-known example of succeeding a larger-than-life founder, deep supply-chain experience, and experience running a company of Nvidia's scale.
• Gaps: he just took on a new role at Apple, and he is not a chip or systems architect. A White House endorsement may also complicate things for a board seeking independence during an investigation.

5. Jay Puri (insider, retiring). Nvidia's longtime head of worldwide sales, who retired from that role in August 2026 and stays on as a senior adviser.

• Strengths: he knows every major customer and partner, from two decades at Nvidia.
• Gaps: he chose to retire this year, and he is a sales leader rather than a product or technology leader. We found no analyst support for him in current coverage.

Other names come up: AMD's CEO Lisa Su, who runs Nvidia's top rival; Apple's hardware chief Johny Srouji, on single-source speculation; and Intel's CEO Lip-Bu Tan, who is listed on one of the markets.

How deep is the bench?

Nvidia is often said to have a deep bench. That is true for its functions, and it is not true for the top job:

What the record shows:

• No groomed successor. Nvidia has no chief operating officer or president. Its annual report on Form 10-K, filed in February 2026, named five executive officers: Huang, Kress, Puri, Shoquist and General Counsel Tim Teter.
• No disclosed plan. Nvidia's 2026 proxy statement never uses the word "succession". It says only that leaders work with the CEO to form a pool of candidates for promotion. The November 12 release cites no succession plan and names no search firm.
• A flat structure. Huang reportedly had about 60 direct reports. That suits a founder who makes the big calls himself, but it leaves no obvious second-in-command.
• A contrast. When Intel's CEO resigned in 2018, its release cited a robust succession process and a search firm. Apple's 2026 release on Cook's handoff cited long-term succession planning. Nvidia's cites neither.

Could Nvidia split the job?

One option the board could consider is a pair of co-CEOs: one leading the business and one leading the technology. We have seen no reporting that the board is weighing this. It is our own view.

In our view, the split would be business and technology, not sales and technology. Nvidia's customers buy on the roadmap, so the technical leader is effectively also its top salesperson. That points to a pairing like Colette Kress, for finance, operations, investors and the deal web, with Ian Buck, for technology, CUDA and the largest customers. Intel used interim co-CEOs, its CFO and a product chief, in 2024.

The case for it:

• It matches the bench. Nvidia has deep leaders in each function and no single groomed successor. Two leaders could cover what one founder did.
• It keeps continuity. Engineers and customers keep a technical leader they trust, and investors keep the executive they know from every earnings call.
• It buys time. It avoids a rushed outside hire while the investigation runs.

The risks, which are specific to Nvidia:

• No one holds the tiebreak. Huang ran an unusually flat organization and was its single strategic voice. The hardest calls, such as how much to keep financing neoclouds and partners, sit exactly where a business leader and a technology leader would disagree.
• The Chair would become the referee. Harvey Jones would effectively break ties while the board also runs an investigation. That is a heavy load for a non-executive chair.
• It can look like indecision. Naming co-CEOs could read as the board avoiding a choice, which would not help rebuild investor confidence.

Our read: a Kress and Buck pairing could be a credible short-term bridge. It could reassure investors more than a long, open-ended search, but as a permanent structure it is weaker. The signal to watch would be an 8-K announcing an interim title change, or a new chief operating officer or president role.

What past founder transitions suggest

What it means. Nvidia's functional bench is long-tenured and real, so the work of designing, building and selling chips does not depend on one person. The CEO bench is thin: no one was groomed, titled or disclosed as a successor. An insider such as Buck would signal continuity, especially for CUDA. An outsider would signal a reset, which can bring more uncertainty in the near term even if it proves right over time. Either way, the most important tests in the coming months are whether key technical leaders stay, and how quickly the board names a permanent CEO.

What this means for long-term investors

We don't make calls on individual stocks in articles like this. But the episode offers a few lessons that apply well beyond Nvidia:

• Key-person risk is real, even at the largest companies. Founder-led companies often benefit from one person's vision. They can also have thin succession plans.
• Separate the business from the valuation. A company can keep earning while investors decide to pay less for those earnings, and that change can last.
• Know where you're concentrated. Nvidia is one of the largest companies in the major U.S. stock indexes. Many investors own it through index funds and AI-themed funds without holding it directly. It is worth knowing how much a portfolio depends on one company, or one theme.
• Look at the whole chain. Capital spending and its risk sit with the builders, while the benefits spread to the users. A company's place in that chain shapes what can move it.

What to watch, with dates

The days right after Nvidia's November 17 earnings are crowded. The requested congressional briefing, Anthropic's IPO pricing and first trading, and the APEC meeting in Shenzhen, where a Trump–Xi meeting is expected, all fall on November 18 and 19.

Sources

• Nvidia: Form 8-K, Items 5.02 and 9.01, filed November 12, 2026, reporting events of November 11, with Exhibit 99.1, "NVIDIA Announces CEO Transition" (press release dated November 12, 2026), SEC EDGAR accession 0001045810-26-000082.
• Nvidia newsroom: "Statement From NVIDIA Board of Directors" (November 8, 2026) and "Statement From the Independent Members of the NVIDIA Board" (November 10, 2026), nvidianews.nvidia.com.
• Anthropic: Form S-1/A, Amendment No. 2, filed November 10, 2026 (SEC EDGAR accession 0001628280-26-072484).
• Analyst multiples: CNBC analyst-calls roundup, November 6, 2026 (Bank of America); Yahoo Finance, November 12, 2026, and Investing.com, November 11, 2026 (Stifel).
• Nvidia: Form 10-K for fiscal 2026 (filed February 25, 2026; executive officers as of February 20, 2026); Form 10-Q for the quarter ended July 26, 2026 (filed August 26, 2026); 2026 proxy statement (DEF 14A, May 12, 2026); Forms 8-K on the Puri retirement and Parker appointment (July 2, 2026) and the chief accounting officer (April 27, 2026). Executive biographies on the Nvidia Blog, Newsroom and Research sites.
• Apple: release on Tim Cook becoming Executive Chairman (April 20, 2026). Intel: releases on CEO transitions (June 21, 2018; December 2, 2024; March 12, 2025). SoftBank Group: Rene Haas biography.
• Nebius: third-quarter 2026 results and shareholder letter; CNBC coverage, November 12, 2026.
• House Select Committee on the CCP: "Moolenaar, Mast and Crawford Seek Briefing from Jensen Huang" (press release on the November 4, 2026 letter), chinaselectcommittee.house.gov.
• The Information: reporting on Kress and the OpenAI backstop, as relayed in other coverage; we did not read the original article.
• CrowdStrike: "CrowdStrike and Cerebras Partner to Power AI Detection and Response on the World's Fastest Inference", press release, July 22, 2026 (crowdstrike.com).
• Schedule 13G and 13G/A filings: Nvidia's stakes in CoreWeave and Nebius, as reported by StockTitan. IREN: its May 7, 2026 release on its Nvidia cloud contract. CoreWeave's capacity order: a CoreWeave filing, as reported by The Inference.
• U.S. Department of Justice: Releases on chip-diversion indictments, Mar 19 and Oct 1, 2026 (justice.gov)
• Rep. John Moolenaar: Statement on Huang's resignation, Nov 12, 2026 (chinaselectcommittee.house.gov)
• Office of the Clerk, U.S. House: Roll Call 13, H.R. 2683, Jan 12, 2026 (clerk.house.gov)
• Library of Congress: Bill status for H.R. 2683, H.R. 3447, H.R. 6875 and S. 4784, as of Oct–Nov 2026 (govinfo.gov)
• eCFR: 15 CFR 750.8; 50 U.S.C. 4819, current as of Nov 2026 (ecfr.gov; law.cornell.edu)
• Bureau of Industry and Security: "Revision to License Review Policy for Advanced Computing Commodities," Jan 15, 2026 (govinfo.gov)
• The White House: Fact sheet on the China state visit, Sep 25, 2026 (whitehouse.gov)
• Shenzhen Longgang District Government (citing Xinhua): APEC 2026 dates, Jan 6, 2026 (lg.gov.cn)
• Axios: House committee advances the AI OVERWATCH Act, Jan 21, 2026 (axios.com)
• CNBC: Live updates on Nvidia's CEO change, Nov 12, 2026 (cnbc.com)
• AP (via Spectrum News): Balance of power after the 2026 midterms, Nov 4, 2026 (spectrumlocalnews.com)
• The Information (via Investing.com): U.S. finalizing a rule to block China's remote access to AI chips, Nov 11–12, 2026 (investing.com)
• Shanghai Metals Market / Skillings: China extends its rare-earth export-control suspension, Nov 9, 2026 (metal.com; skillings.net)
• The National: China extends pause on rare-earth export curbs, Nov 9, 2026 (thenationalnews.com)
• The Next Web (summarizing WSJ and Bloomberg): Megaspeed imports and denial, Nov 10, 2026 (thenextweb.com)
• Reuters, CNBC, BBC, Axios, Forbes, Yahoo Finance, Investing.com, Benzinga, StockTitan and The Inference: reporting and analyst commentary, November 11–12, 2026. Analyst views are as reported by these outlets; we did not read the underlying research notes.