NVIDIA Pledges 50% Cash Flow Return as Huang Maps Agentic AI Future

Key Points

• NVIDIA says it will return at least 50% of free cash flow to shareholders this year, next year and beyond, building on its existing buyback program and higher dividend plans.
• CEO Jensen Huang framed the company’s next growth phase around agentic AI, saying NVIDIA’s Vera Rubin and Grace Blackwell platforms are designed for AI agents that reason, use tools and run across data centers, PCs, cars and robots.
• Huang also highlighted a push into AI PCs with Microsoft and said NVIDIA’s Vera CPUs are expected to play a key role in AI data centers, though he stressed the CPU opportunity will remain smaller than the GPU business.

NVIDIA NASDAQ: NVDA founder and CEO Jensen Huang said the company plans to return at least half of its free cash flow to shareholders while outlining a broader strategy to reshape data centers, PCs and edge devices around “agentic” artificial intelligence.

Speaking at a company event that included analyst and investor questions, Huang opened with a capital return update. He reminded attendees that NVIDIA had previously announced an $80 billion share repurchase program and a 25-fold increase in its dividend. He then said the company now plans to return “50% or more of free cash flow to our shareholders this year, next year, and beyond.”

Huang added that NVIDIA plans to increase stock repurchases and its dividend over time.

Huang Frames AI Future Around Agents

Much of Huang’s presentation focused on what he described as a new computing pattern built around AI agents. He said agents can reason, use tools, access structured and unstructured long-term memory, and operate across cloud data centers, PCs, workstations, cars, robots and other edge devices.

Huang said NVIDIA’s recent product announcements were designed around that shift. He described the Vera Rubin platform as built not only for pre-training, post-training and inference, but also for running agentic workloads that are “distributed and disaggregated” across a data center.

He said token generation remains the core economic driver for AI infrastructure customers, arguing that customers should maximize the number of GPUs in a given power envelope because “what you sell is tokens.” He said Grace Blackwell was designed to generate tokens at the lowest cost, while Vera Rubin extends the architecture to support agentic AI workflows.

Vera CPU Positioned for Agentic Workloads

In response to a question from Morgan Stanley analyst Joe Moore about NVIDIA’s data center CPU opportunity, Huang said CPUs designed for agents will differ from prior CPUs built primarily for human-directed computing. He emphasized single-threaded performance, bandwidth, energy efficiency and CPU-to-CPU communication.

Huang said every data center using NVIDIA GPUs will “likely” use the company’s Vera CPUs, including head nodes, orchestration systems and storage servers. He also said Vera could sell beyond NVIDIA GPU systems through NVLink Fusion and in workloads such as data processing, electronic design automation and simulation.

However, Huang cautioned attendees not to equate the CPU opportunity with the GPU business in average selling price. He said NVIDIA’s GPU and networking businesses remain “quite large” and later added that while the CPU market may expand significantly as agents proliferate, it “can’t reasonably come close to a GPU.”

Huang also discussed Vera’s architecture, saying NVIDIA designed a custom Arm-based CPU core called Olympus rather than using an off-the-shelf Arm core. He said the chip is designed for high instructions per clock and high internal bandwidth, and that NVIDIA used Grace Blackwell to prepare data centers for the broader Vera transition.

Microsoft Partnership Targets AI PCs

Huang said NVIDIA and Microsoft have been working for several years on a new class of computers intended to make PCs function as personal AI assistants. He said the systems are designed to include tensor processing, model parameter compression and secure operating-system sandboxing so agents can be granted limited permissions to access files, tools and communications.

He characterized the effort as “the first reinvention of the PC” in decades, saying future PCs will not only run existing applications but also operate as assistants that can act on users’ behalf. Huang said agents should be able to use applications such as Adobe Photoshop, Adobe Premiere, Autodesk and other tools, rather than replace them.

In a lengthy exchange about why users would want local AI models rather than relying entirely on cloud-based models, Huang said both approaches are needed. He argued that local agents can run continuously, interact with files and tools on the device, and call more powerful cloud models when necessary.

“Why choose?” Huang said, describing a future in which power users rely on both local and cloud AI. He also said NVIDIA is not planning to re-enter smartphone chips, citing Apple and Qualcomm as strong players in mobile devices and saying NVIDIA’s software stack is more differentiated on PCs.

Networking, Supply Chain and Segments

Huang also discussed networking requirements for increasingly large AI data centers. He said NVIDIA will use copper “as long as we can” for short distances, while optics will be necessary for larger scale-out deployments. He said Spectrum-6 is designed for AI factories with hundreds of thousands to potentially a million systems, and noted NVIDIA’s partnerships and investments involving companies including Coherent, Lumentum, Corning and Marvell.

Asked about supply chain capacity, Huang said NVIDIA has support from its ecosystem for “very robust growth,” though he also said the world supply chain remains supply constrained. He said the company has enough support to back the guidance it has provided.

Huang also explained NVIDIA’s recent financial resegmentation, saying the goal was to help investors understand how the business works beyond a single large data center revenue number. He described three broad areas:

• Cloud service providers, where NVIDIA supports external customers, internal workloads and AI companies.
• AI clouds, enterprises and industrial customers, including regional and specialized infrastructure operators.
• Robotics and edge systems tied to physical AI.

Huang said the new segmentation is intended to give investors more granularity as AI expands across cloud, enterprise, industrial and edge markets.

In response to a shareholder question about public skepticism toward AI, Huang said the industry must build AI safely and securely while avoiding rhetoric that discourages people from using the technology. He said he advises his own children to use AI because he does not want them “to be left behind.”

About NVIDIA NASDAQ: NVDA

NVIDIA Corporation, founded in 1993 and headquartered in Santa Clara, California, is a global technology company that designs and develops graphics processing units (GPUs) and system-on-chip (SoC) technologies. Co-founded by Jensen Huang, who serves as president and chief executive officer, along with Chris Malachowsky and Curtis Priem, NVIDIA has grown from a graphics-focused chipmaker into a broad provider of accelerated computing hardware and software for multiple industries.

The company's product portfolio spans discrete GPUs for gaming and professional visualization (marketed under the GeForce and NVIDIA RTX lines), high-performance data center accelerators used for AI training and inference (including widely adopted platforms such as the A100 and H100 series), and Tegra SoCs for automotive and edge applications.

This instant news alert was generated by narrative science technology and financial data from MarketBeat in order to provide readers with the fastest reporting and unbiased coverage. Please send any questions or comments about this story to contact@marketbeat.com.