PepsiCo has begun converting selected US manufacturing and warehouse sites into high-fidelity digital twins under a multi-year collaboration with Siemens and NVIDIA announced at CES 2026. The company reports a 20 percent throughput increase on initial deployment and capital expenditure reductions of 10 to 15 percent. Siemens is simultaneously turning the underlying tooling into a marketplace product.
PepsiCo has started converting selected United States manufacturing and warehouse facilities into high-fidelity three-dimensional digital replicas, under a multi-year collaboration with Siemens and NVIDIA that shifts digital twin technology from something manufacturers demonstrate to something they use to commit capital.
The collaboration, announced by PepsiCo at CES 2026, combines Siemens' digital twin and simulation software with simulation libraries from NVIDIA Omniverse and real-time engineering data drawn from the plants themselves. The twins model both individual plant operations and the end-to-end supply chain running between them, and are first used to establish a performance baseline against which proposed changes can be tested.
The numbers PepsiCo has attached to the early work are the reason the announcement matters more than the technology description suggests. The company reports a 20 percent increase in throughput on initial deployment, reductions of 10 to 15 percent in capital expenditure, faster design cycles, and near-complete validation of designs in simulation before physical work begins — with the large majority of potential problems identified before anything is installed on a factory floor.
That last point is the substantive change. Digital twins have existed as a concept in industrial software for well over a decade, and most manufacturers have built one at some point: a visually convincing model of a line or a cell, produced for a specific project, admired, and then left to go stale as the real plant drifted away from it. The reason those efforts rarely compounded is that the twin was never load-bearing. Nothing depended on it. Engineering decisions were made the way they had always been made, and the model was a parallel artefact.
What PepsiCo describes is different in kind: a twin that sits ahead of the capital approval, absorbing the risk that used to be discovered during commissioning. If a line reconfiguration can be validated in simulation to the point where the capital request shrinks by a tenth or more, the twin has stopped being a visualisation and become part of the investment process. Warehousing is included alongside manufacturing, which extends the same logic to material flow rather than only to machine layout.
Siemens is not treating this as a bespoke engagement. At CES 2026 the company also unveiled Digital Twin Composer, the tooling used to assemble these models, and said it would be available on the Siemens Xcelerator Marketplace in mid-2026. The distinction is commercially significant: what was previously delivered as a consulting-heavy integration project is being turned into a licensable product, which is how a capability moves from the few companies that can afford a custom programme to the many that cannot.
Behind it sits an expanded Siemens–NVIDIA partnership to build what the two companies call an industrial AI operating system, with NVIDIA supplying AI infrastructure, simulation libraries, models and frameworks, and Siemens contributing industrial software, hardware and domain knowledge. Siemens intends to demonstrate the result on itself: its own electronics factory in Erlangen, Germany, is the first site the two companies have named in an effort to build fully AI-driven, adaptive manufacturing plants, with work starting in 2026.
Some caution is due on the figures. They are the participating companies' own, drawn from early deployments at a selected group of facilities, and a 20 percent throughput gain measured at one site during a first implementation is not a rate that holds across a network of plants built at different times to different standards. PepsiCo has not published which facilities are involved or how many.
The direction, though, is not in dispute. A food and beverage manufacturer — not an aerospace or semiconductor firm with unusual tolerance for engineering overhead — is buying simulation as operating infrastructure, and the largest industrial automation vendor in Europe is selling the tooling off a marketplace shelf alongside a chipmaker's simulation stack. That combination says more about where digital twins have arrived than any single deployment figure does.