Nvidia’s Kress Backs Open AI After Huang’s Exit | Zeniteq
Loading
Zeniteq is now available on mobile for iPhone, iPad and Apple Watch.Download nowLearn more
Nvidia’s Kress Backs Open AI After Huang’s Exit
Nvidia’s interim CEO is keeping open models and the pending Hugging Face acquisition central to its strategy, with clear commercial stakes for its chip business.
Nvidia interim CEO Colette Kress reaffirmed the company's open AI model strategy during its first earnings call since Jensen Huang's resignation, signaling continuity on an approach Huang had championed. She cited a 35-fold year-over-year increase in weekly token volume on OpenRouter and described Nvidia's Nemotron as the platform's most-used U.S. open model.
The commercial stakes are clear. With Data Center revenue at $107.6 billion, roughly 93% of Nvidia's $115.8 billion quarterly total, open models function as demand development for its hardware business. Kress also confirmed the pending $12.9 billion Hugging Face acquisition remains on track.
Nvidia interim CEO Colette Kress reaffirmed the company’s commitment to open AI models on November 17, 2026, during its first earnings call since Jensen Huang’s resignation. Her remarks gave developers and investors a specific assurance: the leadership change has not displaced an ecosystem strategy that Huang had strongly supported.
Kress said Nvidia remained “committed to fostering a vibrant open model ecosystem,” according to TechCrunch’s account of her opening remarks. She cited growing usage on OpenRouter, highlighted Nvidia’s Nemotron models and discussed the company’s plans for Hugging Face, which Nvidia agreed to acquire in September.
The commercial connection is direct. Open models give more organizations a starting point for building AI applications. Those applications need computing capacity for customization and operation, creating opportunities for Nvidia to sell the infrastructure underneath them.
My read is that Kress was defending a route to future hardware demand as much as a software philosophy. The earnings release provides useful evidence for that interpretation, although it does not quantify how much revenue open models generate.
Open Models Support Nvidia’s Hardware Business
Nvidia reported $115.8 billion in revenue for its third quarter of fiscal 2027, which ended October 25, 2026. Data Center revenue reached $107.6 billion, up 110% from a year earlier, according to the company’s November 17 earnings release.
Dividing those published figures puts Data Center at approximately 93% of quarterly revenue. That calculation explains why Nvidia has a strong interest in software that helps customers find more uses for AI infrastructure. It does not identify what proportion of that revenue came from open-model workloads.
The connection works through several stages:
Development: An available model gives a team a starting point without requiring it to train a foundation model from scratch.
Customization: Fine-tuning, evaluation and other adaptation work can require additional computing resources.
Inference: Once deployed, the model consumes compute as it processes requests and generates results.
Expansion: Successful applications can attract more users, handle more tasks or run more frequently.
Open models can therefore lower one barrier to adoption while leaving substantial demand for computing capacity. Nvidia can benefit from that expansion without charging for every model download.
The mechanism is especially relevant to agentic applications, which can call models repeatedly while completing a task. TechCrunch describes Nemotron as a family of open-weight models built for long-running agentic AI systems and reasoning workloads. A single user request in such a system can involve multiple inference steps, although the actual computing cost depends on the application.
I would treat the open-model strategy as demand development, not a direct sales forecast. More accessible models create opportunities for additional workloads. Whether those workloads become substantial Nvidia purchases depends on deployment choices, efficiency and customer economics.
What Kress’s OpenRouter Figures Actually Show
Kress cited a 35-fold year-over-year increase in weekly token volume on OpenRouter, including a 3.5-fold increase over the preceding three months. She also described Nemotron as the most widely used U.S. open model on that platform, TechCrunch reported.
Tokens are the units models process and generate, often words or pieces of words. Token volume is therefore a measure of activity, but interpreting it requires attention to scope.
Statement Cited by Kress
What It Tells Readers
Weekly OpenRouter token volume grew 35 times year over year
Nvidia is pointing to rapid growth in activity on a particular platform.
Volume grew 3.5 times in three months
The reported expansion also covers a more recent period.
Nemotron led U.S. open models on OpenRouter
Nvidia claims a strong position within a defined category on that platform, not across the entire AI market.
These figures were stated by Nvidia and were not independently verified in the supplied reporting. The platform-wide growth numbers also should not be read as Nemotron’s own growth rate.
The distinction between tokens and hardware demand is consequential. A token processed by a small model can require very different resources from one processed by a much larger model. Context length, numerical precision, batching and the software used to serve a model also affect the amount of work the hardware performs.
As a result, a 35-fold increase in tokens cannot be converted into a 35-fold increase in GPU demand. More efficient inference could support much more activity on the same installed equipment.
Even with those boundaries, Kress’s choice of metric is revealing. She emphasized developers using models, giving Nvidia a usage-based argument for continued investment in the ecosystem. That is more commercially meaningful than download counts alone, though it remains several steps removed from chip revenue.
Hugging Face Makes the Commitment More Concrete
Nvidia agreed in September to acquire Hugging Face for $12.9 billion, according to TechCrunch. The publication reported that the platform serves more than 18 million developers and 200,000 companies, and that the acquisition remained on track after Huang’s departure.
Nvidia’s official earnings release presents the agreement as a way to expand Hugging Face’s platform and make AI more open, capable and accessible.
That puts a specific transaction behind Kress’s statement. Nvidia is maintaining support for a large developer platform while its board searches for new permanent leadership.
Hugging Face’s strategic value follows from where it sits in the development process. A model platform can influence which models developers discover, evaluate and choose as the foundation for an application. That gives Nvidia a potential connection to projects well before their operators make large infrastructure purchases.
The commercial logic is straightforward:
Developers find models they can build on.
Some projects progress from experiments to deployed services.
Deployed services need training, customization or inference capacity.
Nvidia competes to supply that capacity through its hardware ecosystem.
This is an analysis of the platform’s potential value, not evidence that every Hugging Face project becomes an Nvidia customer. Developers can choose other hardware, hosted services or smaller local deployments.
The size of the proposed purchase makes continued support significant. A $12.9 billion acquisition is a substantial allocation of capital to the software and developer side of Nvidia’s business. Keeping it in the strategy after a founder’s departure sends a clearer signal than a broad promise to support innovation.
TechCrunch also reported that the acquisition was not directly addressed during the call’s question-and-answer session. The available evidence supports continuity of intent, not a completed transaction or a detailed integration plan.
For developers, I would watch the eventual operating commitments closely: platform access, support for models from different organizations and compatibility across deployment choices. Those details will determine what Nvidia’s promise means in daily use.
The Hardware Roadmap Gives the Strategy a Destination
Nvidia’s earnings release said it had begun shipping Vera Rubin NVL72 systems, with racks running at partners including CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius.
That creates a useful distinction between the two sides of Nvidia’s announcement. The hardware roadmap supplies computing capacity. The open-model strategy helps customers identify work to run on it.
A company that starts with an available model still has deployment decisions to make. It needs enough memory for the model and its workload, acceptable response times, and a serving cost that fits the application. A model’s popularity cannot answer those questions by itself.
Nor does open availability mean equal performance across hardware. Runtime support, numerical formats and serving software can affect which system is economical for a particular deployment. Those are practical reasons for Nvidia to invest in the software ecosystem around its chips.
The release also highlights Nvidia’s Open Agent Safety Platform and open-source OpenShell architecture. Their inclusion broadens the software commitment beyond model weights to supporting infrastructure for agents.
“Open model” and “open source” still deserve careful handling. Access to model weights is one dimension of openness; license permissions, training-code availability and access to training data are separate questions. Kress’s ecosystem commitment should not be treated as a blanket licensing guarantee for every model or tool.
For infrastructure planning, this software story belongs alongside Nvidia’s AI supply commitments into 2029. Model availability and hardware availability solve different parts of the deployment problem.
Permanent Leadership Will Set the Longer-Term Priorities
Kress is serving as interim CEO while remaining Nvidia’s chief financial officer. CNBC reported that the board had hired executive search firm Heidrick & Struggles to lead the search for a permanent CEO. Board chair Harvey Jones also said Nvidia did not intend to bring Huang back as chief executive.
That makes Kress’s remarks consequential but bounded. They establish the current leadership’s public position during an unsettled transition. Execution under a permanent CEO remains untested, and that leader could change priorities, spending or the way Nvidia manages its platforms.
CNBC also reported that Kress said Nvidia would rigorously evaluate future investments, commitments and guarantees. The open-model assurance therefore sits alongside a broader message of capital discipline. Support for the ecosystem does not imply that every associated project or investment will proceed unchanged.
The useful indicators now are concrete: progress on the Hugging Face transaction, continued Nemotron development and the terms under which developers can use the surrounding tools.
What Developers and Enterprises Should Do With the Signal
Kress’s statement gives teams a reason to continue evaluating Nvidia’s open-model ecosystem without assuming that Huang’s departure has ended the strategy. It is less useful as a reason to make an irreversible infrastructure commitment.
For near-term planning:
Separate the model from the deployment. Evaluate the model’s suitability and license, then compare the cost and performance of the available hosting options.
Use OpenRouter rankings as discovery signals. A model’s position can justify a closer look, but it cannot replace workload-specific evaluation.
Track the Hugging Face agreement separately from integration promises. An announced acquisition and an operational change to the platform are different milestones.
Preserve flexibility where practical. Clear records of model versions, licenses and serving requirements make future migration decisions easier.
I would give the most weight to whether Nvidia follows this statement with sustained model releases, useful deployment support and clear platform policies. Those actions would connect the interim CEO’s assurance to decisions developers can actually rely on.
Frequently Asked Questions
3 questions
1
Why Did Kress’s Open-Model Comments Matter?
They came during Nvidia’s first earnings call after Jensen Huang’s resignation and explicitly reaffirmed a strategy he had supported, including plans for Hugging Face.
2
Has Nvidia Completed the Hugging Face Acquisition?
The supplied reporting describes a pending $12.9 billion acquisition agreement. Kress’s remarks reaffirmed the plan; they were not an announcement that the transaction had closed.
3
Does More Token Usage Mean More Nvidia GPU Sales?
Not automatically. More activity can increase computing demand, but model size, serving efficiency, hardware choice and available capacity determine how that activity translates into equipment purchases.