Colette Kress Takes Over as Nvidia Interim CEO After Huang’s Reported Exit

Nvidia CFO Colette Kress has stepped into the role of interim chief executive following Jensen Huang’s November 12 resignation, according to CryptoPotato. This transition places her at the helm of Nvidia’s data center strategy right before the company’s scheduled November 17 quarterly results.

Nvidia’s infrastructure ecosystem spans AI processors, enterprise software, server manufacturing partners, and systems deployed by major cloud operators. This leadership shift unfolds as operators pour billions of dollars into expansion while simultaneously grappling with power and capacity bottlenecks.

Nvidia’s Leadership Transition Comes Before Quarterly Results

What changed at the top of Nvidia?

According to CryptoPotato, Huang stepped down from his positions as CEO and board member, with Kress assuming the top job on an interim basis. Kress has a tenure of more than ten years at Nvidia, while the board of directors actively evaluates both internal and external contenders for the permanent executive seat.

Read More: Jensen Huang Exits Nvidia After 33 Years; Shares Fall Over 10% Premarket

This succession puts a seasoned finance leader at the helm of an enterprise historically defined by Huang’s uniquely centralized management style. Core operational strengths—such as AI-chip demand, the CUDA ecosystem, and deep industry partnerships—remain pivotal assets during this period of change.

How did the transition unfold?

Reports from BeInCrypto indicate that directors initiated an inquiry on November 8 to determine if a foreign intelligence entity had targeted Huang. His subsequent resignation occurred on November 12, prompting Elon Musk to publicly endorse a petition advocating for his reinstatement on November 13.

The board retained the services of Wachtell Lipton, placing former acting US Attorney Joon H. Kim at the helm of the inquiry. With quarterly earnings slated for November 17, the company faces a major financial disclosure directly on the heels of significant leadership turmoil.

Who Controls Nvidia’s Next CEO Appointment?

What authority does the board have?

Based on the corporate bylaws outlined in recent governance coverage, executive appointments are determined by a majority vote of the directors currently in office. Following Huang’s departure, that threshold sits at six out of the 10 remaining directors, as shareholders lack a direct vote on the CEO position.

Although Musk holds no seat on the Nvidia board, his public endorsements lend high-profile visibility to the reinstatement movement. Ultimately, however, the official selection remains squarely in the hands of the directors managing the ongoing investigation and executive succession.

How much influence does Huang retain?

Nvidia’s 2026 proxy indicates that Huang controls roughly 3.58% of the company’s voting power, whereas institutional giants BlackRock and Vanguard each maintain stakes exceeding 7%. Employee sentiment following his departure was visibly demonstrated by an internal company post calling for his return, which amassed upwards of 1,600 comments.

Kress’s interim appointment equips the organization with critical institutional memory during this immediate pivot. As the board weighs options from both inside and outside the firm, the long-term leadership trajectory stays open leading up to the upcoming financial report.

Nvidia Data Center Products Extend Beyond AI Chips

What did Nvidia introduce for enterprises in 2021?

At Computex 2021, Nvidia unveiled its cloud-based Base Command platform alongside AI-optimized x86 servers via its certified systems initiative. TechRadar’s coverage of the rollout identified NetApp as a primary Base Command collaborator, alongside server manufacturers Asus, Dell Technologies, Gigabyte, Hewlett Packard Enterprise, and Lenovo.

The launch tightly coupled physical hardware with enterprise software packages. The machinery received certifications for AI Enterprise running on VMware vSphere, Omniverse Enterprise, and Red Hat OpenShift, alongside integration support for Cloudera data engineering and machine-learning platforms.

How did Base Command connect teams to larger systems?

Base Command granted enterprise groups cloud-enabled access to Nvidia DGX SuperPOD architecture by leveraging NetApp data management solutions. The system architecture was engineered to accommodate multiple data scientists, researchers, and technical squads operating on concurrent AI initiatives.

During the initial 2021 announcement, the early-access subscription carried a price tag of $90,000 per month. This historical offering highlights Nvidia’s strategy of bundling expansive computing power with development toolkits and partner software, pushing its business model well beyond the sale of standalone silicon chips.

Power and Cooling Shape Nvidia Infrastructure Expansion

Where are operators encountering constraints?

Industry reports from DataCenterFeed point to power and capacity limitations highlighted by major operators like Equinix and Digital Realty. Equinix logged $533 million in annualized gross bookings, while Digital Realty reported $855 million in bookings at 100% share alongside a $2.5 billion backlog.

Oracle’s cloud infrastructure supplies another pathway for customers to access Nvidia technology, having made Vera Rubin NVL72 racks and Vera CPUs generally available through OCI while also mapping out a multi-gigawatt Supercluster strategy.

How are server suppliers adapting their hardware?

Supermicro has engineered a factory-integrated 48U rack housing 256 Nvidia Vera CPUs, 22,528 Arm cores, and up to 400 TB of memory. Utilizing direct liquid cooling, the system is tailored to execute agentic-AI sandboxes side-by-side with GPU arrays, effectively broadening liquid-cooling demands to include CPU server racks.

Meanwhile, Crusoe utilizes a modular blueprint for its sixth-generation Spark module, packing up to 3.5 megawatts of compute power into a prefabricated unit. The apparatus utilizes a closed-loop liquid cooling framework that consumes zero water and maintains full compatibility with Nvidia GB300 NVL72 and Vera Rubin NVL72 rack designs.

AI Infrastructure Spending Reaches Tens of Billions

What scale has CoreWeave reached?

CoreWeave reported $3.66 billion in third-quarter revenue alongside a $132.5 billion backlog and an active power capacity exceeding 1.7 gigawatts. The firm revised its capital expenditure forecast for 2026 upward to a range of $37 billion to $39 billion, underscoring the massive financial commitments driving AI computing infrastructure.

The company maintains that it will eschew parent-level unsecured debt through 2027 under current market yields. Instead, management plans to rely on contract-backed financing agreements and advance customer prepayments, aligning its borrowing strategy directly with secured business revenues.

What investment plans has Meta outlined?

Meta projected its 2026 capital expenditures to fall between $135 billion and $145 billion. The tech giant also filed for approximately $35 billion via underwritten common and mandatory convertible preferred stock offerings, alongside a planned $60 billion at-the-market equity program—a transaction in which Nvidia intends to invest $3 billion.

Furthermore, Meta confirmed the completion of Phase 1 of its Prometheus supercluster located in New Albany, Ohio, which has been utilized to train its Muse models. The facility is scaling toward a capacity of roughly 1 million GPUs, and Meta notes that internal software coupled with battery reserves allows the facility to manage power load fluctuations of 30% to 40% smoothly.

DGX Spark Brings Nvidia Computing to Smaller Sites

What does the compact Nvidia system offer?

Reporting from Trak.in details the DGX Spark as a compact Grace Blackwell platform, originally brought to light under the moniker Project DIGITS. Nvidia states that the hardware yields up to 1 petaflop of AI computing throughput, backed by 128GB of unified memory and the capability to execute local models packing up to 200 billion parameters.

This desktop-form-factor appliance is aimed squarely at enterprise researchers, individual developers, and scientists. It introduces a decentralized computing tier to Nvidia’s broader infrastructure narrative, operating alongside massive cloud installations and enterprise server farms for localized model training and inference tasks.

What is Span proposing for residential computing?

Emerging startup Span is engineering compact XFRA server nodes tailored for deployment within residential districts. The enterprise asserts that aggregating thousands of these nodes could theoretically deliver computational output equivalent to a conventional 100-megawatt AI data center.

The core concept relies on utilizing pre-existing electrical grid connections in residential zones. Nvidia executive Marc Spieler noted that deploying units in locations already equipped with power makes pragmatic sense when acquiring large standalone power allocations proves difficult. While the DGX Spark and XFRA represent distinct products, both align with this broader distributed-computing philosophy.

Governance and Financing Risks Accompany the Buildout

Why does the succession matter for infrastructure execution?

Jensen Huang’s foundational role across Nvidia makes the ongoing executive search a critical factor for an enterprise intertwined with chip fabricators, hyperscale cloud vendors, and data center operators. Layering an internal investigation onto this transition introduces acute governance hurdles, even as massive infrastructure investments face continuous scrutiny regarding their ultimate financial return on investment.

CoreWeave’s distinct funding strategy accentuates how heavily capital expansion relies on prevailing borrowing costs and firm customer purchase commitments. On the distributed computing front, Span’s capacity benchmarks remain prospective projections, as the initiative relies heavily on successfully orchestrating thousands of smaller nodes across residential power grids.

What does the separate Singapore server case allege?

In an unrelated legal development, reporting from The Defiant highlights that Singapore authorities brought 41 charges against Shi Bingzhe and two associated corporate entities. These money-laundering accusations center on the procurement of 917 servers as part of an investigation into shipments directed toward Malaysia.

This legal proceeding introduces additional external scrutiny surrounding Nvidia-linked server commerce across the broader tech landscape. Crucially, these allegations target specific individual defendants and isolated server transactions, remaining entirely separate from the ongoing board-level inquiries concerning Huang.

November Results and 2027 Manufacturing Plans Come Next

Which events are on the immediate calendar?

Nvidia’s financial reporting scheduled for November 17 serves as the next official corporate disclosure milestone under Kress’s interim oversight. The executive search runs concurrently with this reporting cycle, with board members actively reviewing internal and external candidates.

Separately, the House Foreign Affairs Committee has formally requested Huang’s presence at a hearing on December 1, appending warnings of a formal subpoena. This creates a distinct regulatory challenge in Washington as the board works to navigate its leadership transition and internal probe.

What infrastructure production is planned for 2027?

Crusoe anticipates launching commercial production of its sixth-generation Spark modules out of its Brighton, Colorado facility by 2027. Their inherent support for Nvidia GB300 NVL72 and Vera Rubin NVL72 hardware anchors this manufacturing roadmap firmly inside the broader supply chain supporting modern AI data centers.

These factory-assembled modules integrate dense compute arrays and cooling mechanisms into a single deployable asset. They remain entirely separate from Nvidia’s own DGX Spark desktop appliance: Crusoe’s product is engineered for broad infrastructure deployment, whereas Nvidia’s compact machine targets localized AI research and development.

FAQ

What specifications does Nvidia DGX Spark offer?

Nvidia’s official specifications for the system include 128GB of unified memory and up to 1 petaflop of raw AI computing performance. Originally unveiled under the name Project DIGITS, this Grace Blackwell machine is capable of running local models containing up to 200 billion parameters.

Which server manufacturers appeared in Nvidia’s 2021 certified systems announcement?

The verified equipment vendors included Asus, Dell Technologies, Gigabyte, Hewlett Packard Enterprise, and Lenovo. Their AI-optimized x86 server hardware formed the foundation of Nvidia’s certified systems lineup launched during Computex 2021.

What was Base Command’s monthly early-access price in 2021?

During its initial 2021 rollout, the early-access tier for Base Command was priced at $90,000 per month. This package granted cloud-managed access to DGX SuperPOD computing architecture combined with NetApp data management tools designed for collaborative AI engineering teams.

Who can appoint Nvidia’s chief executive under its described bylaws?

Selecting or appointing a chief executive requires a affirmative majority vote from the directors currently active on the board—calculated as six out of the 10 remaining directors following Huang’s exit. Public shareholder petitions do not grant any direct voting power over the CEO position.

When are Nvidia’s next quarterly results scheduled?

The next quarterly financial update is locked in for November 17, 2026. Colette Kress will step up to lead this earnings disclosure in her capacity as interim CEO after transitioning from her long-standing CFO post.

When has the House committee asked Huang to appear?

The House Foreign Affairs Committee issued a request for Huang to testify on December 1, accompanied by warnings of a potential subpoena, positioning the event as a voluntary request subject to compulsory legal measures.

Where will Crusoe manufacture its planned Spark modules?

Crusoe has targeted its manufacturing facility in Brighton, Colorado, to commence production runs in 2027. These sixth-generation modules are tailored for Nvidia GB300 NVL72 and Vera Rubin NVL72 server racks, delivering up to 3.5 megawatts of compute capacity per deployment unit.