NVIDIA’s supply constraint is exposing a broader capacity-planning problem. Securing more semiconductor supply will not close the gap unless power, land, construction, cooling and customer infrastructure become available at the same rate.
The company expects revenue to grow approximately 70% in fiscal 2028, but customer forecasts indicate demand could support growth of around 100%. Its response shows how capacity planning changes when the constraint can move between tiers, functions and planning horizons.
In Brief
- NVIDIA’s fiscal 2028 outlook is supply constrained, but management has not identified one dominant bottleneck.
- The company is coordinating semiconductor and memory supply with land, power and data-centre capacity that can take two to three years to prepare.
- Inventory, multiyear supply agreements and infrastructure financing are being used together to support future capacity.
- The challenge is no longer securing individual inputs. It is synchronising the complete system required to convert demand into revenue.
One Forecast, Multiple Constraints
NVIDIA’s expected 70% revenue growth represents the output its supply network can currently support. Customer forecasts point to substantially higher demand, creating a gap that cannot be attributed to one component or supplier.
Management said the entire supply chain is operating under pressure. Semiconductor capacity, memory, yields and supplier output remain important, but they form only part of the constraint.
The infrastructure required to deploy NVIDIA’s systems has become equally critical. Land, power, data-centre shells, construction, cooling and labour must be available before additional technology can enter operation.
This changes the capacity question. Increasing output at one point in the network will not increase overall throughput if the next constraint is power availability or an unfinished customer site.
NVIDIA has therefore extended its involvement in both directions. It works upstream with memory and semiconductor suppliers while coordinating downstream with power generators and providers of land, power and data-centre capacity.
Planning Horizons Must Be Connected
The constraints also operate across different timescales.
NVIDIA began production shipments of Vera Rubin in August and expects the platform to account for approximately 20% of data-centre revenue in the third quarter. Supporting that ramp requires near-term inventory, production capacity and component availability.
Inventory increased to $32 billion ahead of the launch. NVIDIA has also secured multiyear supply commitments for Rubin and later products, with the largest requirements concentrated within the first three years.
Memory is already affecting the cost of that plan. Prices have increased more sharply than expected, prompting NVIDIA to work with all three major suppliers to expand the capacity required for its product roadmap. Higher component costs are expected to contribute to lower gross margins over the next two quarters.
Further downstream, the planning horizon is considerably longer. NVIDIA said land, power and shell capacity can require preparation two to three years before deployment.
At the Portsmouth campus, the company has secured infrastructure expected initially to support 4.25 gigawatts of AI capacity for OpenAI. The site could accommodate several generations of NVIDIA systems over 20 years, separating the lifecycle of the infrastructure from the technology installed within it.
The operational challenge is to connect these horizons. Component supply secured today must align with production ramps, infrastructure projects and customer deployments that mature at different speeds. A delay anywhere in that sequence can leave capacity stranded elsewhere.
Capacity Assurance Moves Onto the Balance Sheet
NVIDIA is also using financial commitments to bring infrastructure capacity online.
For selected NeoCloud projects, it provides take-or-pay commitments covering part of a facility’s capacity. The resulting minimum revenue guarantee helps infrastructure providers secure financing from independent lenders. NVIDIA then receives a share of revenue generated above that floor.
The company also plans to provide selective credit enhancement for nearly two gigawatts of compute for another AI business. Customers receiving balance-sheet support could represent roughly one-quarter of NVIDIA’s business next year.
These arrangements move capacity assurance beyond supplier contracts and purchase commitments. NVIDIA is helping to make downstream infrastructure financially viable so that future semiconductor supply has somewhere to be deployed.
That creates additional exposure. Management argues that its systems are sufficiently fungible to be redeployed if customer demand does not materialise as expected, but it has not disclosed the aggregate financial risk or the time and cost involved in moving equipment elsewhere.
Capacity Planning Follows the Constraint
NVIDIA has not identified whether memory, semiconductor production, power, land or customer infrastructure will be the most limiting factor in fiscal 2028. That uncertainty is itself important.
When demand is growing quickly, the primary constraint can move as capacity is added across the network. More semiconductor supply may expose a power shortage. Additional power may reveal a construction delay. Completed infrastructure may still depend on a customer’s ability to finance and operate it.
NVIDIA’s response is to plan capacity across the complete route to deployment. It is combining inventory, supplier commitments, infrastructure coordination and financial guarantees because none can close the supply gap independently.
For supply chain leaders, the case highlights a wider planning requirement. Capacity cannot be assessed solely through the output of direct suppliers. It must reflect every dependency that determines whether supply can be converted into usable capacity, customer service and revenue.
