After Huang, Nvidia Has to Rebuild the Authority One Founder Held

Nvidia's next chief executive will inherit a management structure that ran through one person. Jensen Huang ran the company through about 60 direct reports, by his own account, and preferred giving feedback to the group over private meetings. That flat arrangement put a great deal of coordination within reach of one person. Replacing him means deciding where those decisions should go.

Chip research firm SemiAnalysis examined the cost of losing Huang in a report this week. The danger begins with the organization itself: a short route to the founder can become a queue at his successor's door. Spreading authority could help Nvidia grow beyond him. Doing so while preserving speed will require managers to resolve disagreements that once traveled upward.

Colette Kress, the finance chief, is serving as interim CEO. As South Shore Press reported Thursday, the board's search reaches outside Nvidia as well as within it. A permanent successor has yet to be announced. Kress must keep the business moving while the board decides how much of the old job one person should hold.

Huang helped establish Nvidia in 1993 and remained CEO for 33 years. There is no equally obvious replacement for that accumulated authority. A successor can receive the same title immediately; earning comparable latitude from employees and counterparties takes longer. The board should expect a transition in how decisions get made even if its eventual choice comes from inside.

SemiAnalysis argues that a founder can defend an expensive experiment as part of the company he intends to build, while a hired executive starts with a record still to establish and a board judging every allocation of capital. That difference may encourage clearer spending tests. It may also encourage abandoning a project before its usefulness can be measured. Nvidia's annual report dates its CUDA software platform to 2006 and the AlexNet image-recognition breakthrough on its GPUs to 2012, six years later.

Nvidia's filing puts $25 billion against future equity investments, including stakes in AI model developers. Such commitments make the distinction concrete. A return can arrive through demand for Nvidia's products as well as through the investment itself. Management will need to explain that reasoning to shareholders. Treating every investment as a self-contained financial asset could miss part of the purpose.

Buying into Nvidia's future also means trusting its account of where computing is headed. A customer committing capital today must believe the equipment will remain useful as the technology advances. In SemiAnalysis's telling, Huang's long tenure gave that promise a recognizable author. On Thursday, Nvidia shares fell 4.3 percent, while AMD and Broadcom each rose more than 5 percent.

Confidence has practical consequences for the roadmap. A customer may seek firmer assurances before funding the next expansion if leadership is unsettled. The next CEO will need to demonstrate continuity through delivery. Nvidia's $29 billion in cloud-service commitments, which support research including its own open models, underline how closely its development plans depend on infrastructure arranged years ahead.

The supply side gives this confidence a price tag. Nvidia's future commitments totaled $366 billion at July 26, of which $279 billion covers supply and capacity. The filing identifies memory and manufacturing facilities as the main uses of that supply spending. Suppliers can plan around business of that scale, while Nvidia gains a way to prepare for demand before finished systems exist.

Huang's exit leaves the agreements in place. A supplier asked to expand further must assess the next management team's judgment, and SemiAnalysis argues that more cautious negotiations could give rivals room to compete for future capacity.

Open models are another test of patience. At CES in January, Huang said they had "really revolutionized artificial intelligence." In July he promoted a policy letter urging Washington against premature limits on downloadable models, writing, "The world needs both frontier closed models and frontier open models." Nvidia's Nemotron family gives the company a product behind that advocacy.

SemiAnalysis casts open models as a way for a chip seller to widen the market for computing, which gives funding them a commercial logic even when outsiders use them freely. Nvidia is losing a forceful advocate who could connect that logic to its hardware business. His successor will have to justify the spending while deciding how prominently to defend openness in public.

On Nov. 4, three House committee chairmen sought a briefing from Huang about Alice Huang, who formerly ran Singapore chip customer Megaspeed. Two newspapers, citing anonymous sources, reported that U.S. intelligence officials believe China's main spy agency recruited her. Beijing denies she is a spy, and China's Foreign Ministry called the allegation "unsubstantiated." Nvidia says it provides customers with no classified information.

On Nov. 8 the board announced an independent investigation into reports of a foreign intelligence effort targeting Huang. He told employees, "I was unfaithful to my wife," without identifying anyone. Reports have connected the affair to Alice Huang. Singapore's Home Affairs ministry says its investigation concerned suspected false declarations involving servers and that she was never investigated for espionage. On Thursday, Attorney General Todd Blanche said, "There's no conclusions reached by the Department of Justice or by me or by (FBI) Director (Kash) Patel around who she is or is not." The allegations remain unproven. South Shore Press previously examined the stakes for the industry.

Congress faces its own deadline. House and Senate leaders are negotiating the defense authorization before Democrats, projected to have won both chambers, take over in January. House chairmen are pressing to put chip restrictions in that bill. House Speaker Mike Johnson said Thursday that he sees "a developing consensus in Congress" on limiting China's access to advanced chips.

Under the Remote Access Security Act, export controls would also cover access supplied through cloud computing. Its House vote was 369 to 22 in January; it remains with Senate Banking. That approach reaches beyond the shipment of physical hardware. Nvidia's next leader faces a policy dispute over where its processors can be used even when they stay outside China.

The criminal cases concern alleged physical diversions. Manhattan prosecutors charged Supermicro co-founder Wally Liaw and two others in March. They say a Southeast Asian customer purchased roughly $2.5 billion in servers and served as a route to China. Liaw has entered a not-guilty plea, and a co-defendant entered a guilty plea in October. Separately, Keelung prosecutors charged an Nvidia Taiwan manager and others over 130 servers containing B300 chips. Prosecutors say 74 arrived with buyers in China.

Policy risk also reaches Nvidia's suppliers. China's pause on its U.S. gallium and germanium export ban ends Nov. 27, and a Nov. 9 Chinese notice that moved other export-control pauses to Jan. 10 left that date unchanged. Nvidia's annual report warns that new export limits could "provoke responses from foreign governments, including China, that negatively impact our supply chain." Huang's departure leaves the next chief executive responsible for this exposure alongside the company's commitments. Nvidia can retain technical ambition while distributing authority more widely. Its durability will depend on whether the board gives new leadership enough room to make long investments.