PNY brings a digital twin configurator for AI factories to LEAP

PNY Technologies is at LEAP 2026 with its accelerated computing portfolio and, more interestingly, a tool for deciding what to buy before buying it. The PNY AI Factory Digital Twin Configurator lets customers design and configure an AI factory as a digital twin, built on NVIDIA Omniverse, so that a deployment can be laid out and examined before hardware is ordered. The company is on stand D10 in Hall 3 for the duration of the show. The rest of what PNY is showing is the conventional part: its AI Enterprise Factory portfolio, NVIDIA RTX PRO and GeForce graphics products, and the networking and infrastructure pieces that go around them. Live demonstrations on the stand have been put together with the partners SOMOD, DDN, INFINIARC, Vertiv and F5 — storage, power and delivery, which is a fair map of what an AI deployment actually consists of once the accelerators are accounted for. The configurator is the part worth pausing on, because it addresses a problem that has become expensive in this market. An AI cluster is not a rack of servers; it is a tightly coupled system in which network topology, power delivery, cooling capacity and floor layout constrain each other. Get the interconnect wrong and expensive accelerators sit idle waiting on data. Get the thermal design wrong and the equipment throttles. These are mistakes discovered on commissioning, when the hardware has been paid for and the room has been built. Simulating that in advance is not a new idea — industrial engineering has done it for decades — but applying it to AI infrastructure procurement is comparatively recent, and it lands at a useful moment in Saudi Arabia. The Kingdom is building AI capacity at a rate that leaves little slack for redesign: HUMAIN and NVIDIA have set out plans running to 500 megawatts of AI factories , and the first AMD systems went live with Cisco and HUMAIN this week . The organisations buying into that build-out are frequently doing it for the first time. PNY's own position in the chain is that of a specialist distributor and systems supplier for NVIDIA products rather than a primary manufacturer, which shapes what the configurator is for. It is a pre-sales instrument as much as an engineering one: a way of turning a vague requirement into a specified bill of materials, and of doing it in a room with the customer. That is not a criticism. Most of the friction in AI infrastructure procurement sits precisely there, between a customer who knows what workload they want to run and a configuration nobody has yet drawn. What the company has not published is any Saudi customer, contract or deployment tied to this appearance, nor pricing or availability terms for the configurator itself. Other vendors have brought comparable planning and orchestration arguments to the same floor, among them Think with its bare-metal AI compute layer and BIOSTAR with edge systems at DeepFest . A digital twin of an AI factory is only as good as the assumptions loaded into it. The test is whether a customer who configured a cluster this way finds the real thing behaves as the model said it would. TechScoop is reporting from LEAP 2026 in Riyadh.