NVIDIA builds the AI future with an accelerated supply chain

Part of the NVIDIA culture is that we love challenges,” says Ye Zhao, NVIDIA’s Senior Director of Business Process Operations. “Jensen Huang, our CEO, says we get the greatest joy from solving difficult problems.” As the premier provider of products and platforms for accelerated computing — the brains of AI — NVIDIA’s road to a multitrillion-dollar market cap has been paved with joyful problem-solving.

NVIDIA’s evolving business requires a scalable solution to manage its increasingly complex supply chain. Transitioning from a chipmaker to a comprehensive platform provider introduced challenges in three key dimensions:

• Expanded offerings: NVIDIA evolved to deliver a full accelerated computing stack, including chips, systems, software, services, and data centers.
• Sophisticated sales motion: Selling directly and through partners, along with offering both products and services, made demand signals more difficult to interpret.
• Massive growth: The acquisition of Mellanox in 2019 tripled NVIDIA’s active SKUs, further straining traditional planning processes.

NVIDIA’s demand planning organization remained lean throughout this transformation.  “In the last couple of years, our demand planning organization has grown about 20% while NVIDIA’s revenue grew 550%,” says Zhao. “That's why we rely heavily on efficiency and automation.”

“The complexity of our business has grown exponentially — from being a chip company to a full platform provider, integrating hardware, software, and services,” says Angela Ying, NVIDIA’s Vice President of Business Operations. “Managing this scale requires a new level of planning and coordination across demand, supply, and business strategy. The ability to adapt quickly and efficiently is what drives our success.”

Anaplan is one of the AI-powered solutions NVIDIA uses to enable agile, real-time supply and demand planning at scale.

A new era of supply chain excellence

Since launching in 2020, the Anaplan footprint at NVIDIA has grown into an integrated collection of some 20 models that support supply planning, demand planning, and capacity planning at the chip, substrate, component, and board level. The Anaplan models generate critical data used to generate long-range plans as part of an integrated business planning process.

Key models in the NVIDIA environment illustrate the efficiency and automation improvements achieved with Anaplan:

• Country of Origin: This Anaplan model helps demand planners solve a multifaceted “where to manufacture” puzzle. The variables are many: Due to regulations, some customers can only purchase products made in certain countries (inclusion rules), and other customers can’t purchase products made in specific countries (exclusion rules). On top of these “hard” rules, price factors like tariffs are built into “soft” rules. Using Country of Origin, demand planners recommend where products should be made, in what percentages, for maximum availability and cost-effectiveness.
• Pack Out Allocation:  This Anaplan model compares NVIDIA’s mostly unrestrained request-to-build plan with the inventory of semi-finished goods (called “six-star supply”) to create a constrained signal of what needs to be built. This constrained signal enables the demand and operations teams to establish accurate priorities in case shortages arise. The Pack Out Allocation model replaced a process that was less than optimal, even when NVIDIA offered fewer, simpler products. “In today’s world, at our scale, there’s no way we could do that process manually,” Zhao says.
• Sales Set Planning: This Anaplan model supports NVIDIA’s complete data center solutions — GPUs, switches, cables, and other components. Because a single missing cable can halt an entire data center project, the ability to calculate demand for sales sets versus stand-alone sales — and prioritize production accordingly — means fewer delays and more completed projects.
• Demand segmentation: This Anaplan model slices demand by products that are forecasted, backlogged, and allocated. Demand segmentation used to take two or three hours to run, but in a pilot project the demand planning team applied cuOpt, NVIDIA’s GPU-accelerated solver, to the challenge — and reduced the time to as little as under 15 minutes. “It’s game-changing for us,” Zhao says.

Having the production planning process on the Anaplan platform enables powerful scenario modeling during key production meetings at NVIDIA. “One planner will say to another, ‘Why don’t you change this number,’ and they can see the impact in real time. In other tools you have to save and refresh to see the change,” Zhao says. “You can imagine the impact this has to increase and improve team collaboration. It sounds very simple, but it makes a huge difference.”

Driving innovation and collaboration

With the planning process now running on the Anaplan platform, NVIDIA’s planners have the tools they need to drive innovation. Real-time collaboration and scenario modeling have increased team efficiency, while the integration of AI continues to unlock new possibilities.

Zhao says the reimagined processes at NVIDIA benefit from user input. “I'm a firm believer that technology should serve the user journey, and the measure of effectiveness of a system is the final user experience,” she says. “When you understand and follow a team’s workflow, users are more willing to change, and adoption is better because the process is natural to them.”

Having skilled Anaplan super users and a centralized Anaplan Center of Excellence (CoE) in IT also ensures success. The super users, who work on the demand planning team and thoroughly understand its processes but don’t have day-to-day planning responsibilities, create new models in collaboration with their planner colleagues. “The planners have a business to run, so it’s very hard for them to keep projects going — especially around the end of the quarter,” explains Ying. “So now we have a group within the demand planning team who can focus on driving the projects forward.”