• Home
• News

NVIDIA heads toward GITEX GLOBAL 2026 at a very different point in the artificial intelligence cycle from the one that made GPUs the technology industry’s most sought-after hardware.

The question is no longer simply how many accelerators a company can deploy.

As AI moves from chatbots toward agents capable of reasoning, using software and carrying out multi-step tasks, NVIDIA is increasingly designing the entire environment around those systems — processors, networking, security, software, cooling and even the relationship between an AI data center and the electricity grid.

That transition is likely to define NVIDIA’s presence at GITEX GLOBAL 2026, where the company is a Badge Sponsor, particularly after a wave of October announcements that expanded its strategy well beyond GPUs.

At the center is Vera Rubin, NVIDIA’s new generation of AI infrastructure. But perhaps more important is what the company is building around it: AI factories designed to produce tokens at industrial scale, security architectures that place boundaries around autonomous agents and new systems intended to squeeze more computing from every available watt of electricity.

Why GITEX Comes at a Critical Moment for NVIDIA

GITEX arrives only weeks after NVIDIA used GTC Berlin to set out the next stage of its AI infrastructure strategy.

The timing matters.

Vera Rubin is moving into production, cloud providers are preparing deployments and NVIDIA has started talking increasingly about AI infrastructure in economic terms — cost per token, tokens per watt and how quickly an AI factory can begin generating production workloads.

This changes the conversation from raw GPU performance.

For enterprises and governments, building AI infrastructure now involves a much larger equation: how quickly a facility can be deployed, how much electricity it consumes, how securely autonomous agents can operate and whether sensitive workloads and data can remain within national or organizational boundaries.

Those questions are particularly relevant in the Gulf, where governments, sovereign technology companies and cloud providers are committing substantial capital to AI infrastructure.

NVIDIA therefore arrives in Dubai with an opportunity to position itself not merely as the company supplying the chips inside that expansion, but as the architecture sitting underneath it.

Vera Rubin Moves From Roadmap to Real Infrastructure

The technology to watch most closely is the NVIDIA Vera Rubin platform.

Vera Rubin is designed specifically for increasingly complex reasoning and agentic AI workloads and combines multiple technologies into a rack-scale computing platform rather than treating the GPU as an isolated component.

A Vera Rubin NVL72 system combines 72 Rubin GPUs and 36 Vera CPUs, alongside ConnectX-9 SuperNICs, BlueField-4 data processing units and sixth-generation NVLink connectivity.

NVIDIA describes the wider platform as a multi-rack, pod-scale AI supercomputer built from five purpose-designed rack-scale systems.

The architecture is intended to tackle one of the biggest problems facing advanced AI: moving enormous amounts of data quickly enough between processors and memory without allowing communication bottlenecks to erase the benefits of adding more computing power.

That becomes increasingly important for reasoning models that may spend significantly more computing time working through a problem before producing an answer.

The Numbers Behind Rubin Are Starting to Matter

Early benchmark results provide an indication of why NVIDIA sees Rubin as more than a routine GPU upgrade.

In MLPerf Inference 6.1 preview results released in September, Vera Rubin NVL72 delivered up to 3.7 times the throughput of GB300 NVL72 on Qwen3-VL workloads.

On DeepSeek-R1, NVIDIA reported throughput of up to 2.5 times higher than GB300 NVL72.

Those results matter because inference is becoming one of the biggest economic questions surrounding AI.

Once a model is trained, every question, image, video or task processed by that model consumes computing resources. As AI services reach millions of users and autonomous agents begin carrying out longer chains of actions, the amount of inference required can rise dramatically.

The commercial battle therefore shifts toward how many useful tokens an AI system can generate from a rack — and how much each of those tokens costs.

One of NVIDIA’s Most Interesting Innovations Is Not a GPU

The more surprising technology to watch around GITEX may be OpenShell.

Announced as part of NVIDIA’s latest agentic AI push, the OpenShell Reference Architecture is designed to address a problem that becomes more serious as AI gains greater autonomy: how do companies prevent an agent from accessing something it should not?

Traditional software generally executes predefined instructions.

AI agents are different. They can determine intermediate steps, call external tools, access company information and perform actions on behalf of users.

That flexibility also creates new security risks.

NVIDIA OpenShell treats agent workloads as untrusted during execution and gives them access only to the resources required for a particular task.

Its architecture combines Vera Rubin infrastructure for inference, Vera CPUs for agent orchestration and tool use, and BlueField-4 DPUs running NVIDIA Sentry for hardware-level enforcement.

At the software level, OpenShell uses gateways, supervisors, isolated sandboxes and policy controls to determine what an agent is allowed to access and execute.

AI Agents Need Boundaries, Not Just Intelligence

This could become one of the defining enterprise AI issues of the next few years.

A customer-service chatbot producing an inaccurate response is one type of risk.

An autonomous agent with permission to access corporate databases, execute code, use credentials or interact with financial and operational systems creates an entirely different security problem.

As enterprises move agents from demonstrations into production, governance could therefore become as important as intelligence.

NVIDIA is attempting to make those controls part of the underlying AI infrastructure rather than adding security after an agent has already been deployed.

Its wider Agent Toolkit now includes the Agent Knowledge Foundation, Nemotron models, OpenShell, NemoClaw blueprints and tools designed to help companies connect agents with enterprise information while monitoring and governing what they do.

That gives NVIDIA another route into enterprise AI beyond selling the hardware used to run the models.

The AI Factory Is Becoming a Product

Another technology unveiled just before GITEX illustrates how radically the data-center model is changing.

NVIDIA’s new DSX POD reference design effectively turns an AI factory into something that can be manufactured before arriving at the data-center site.

Instead of building every element of an AI facility independently on location, partners can integrate Vera Rubin computing, networking, power and cooling infrastructure inside modular units before delivery.

NVIDIA says the approach could reduce deployment at a prepared site from around six months to as little as 14 days.

Dell Technologies, HPE, Lenovo, Schneider Electric, Vertiv and other infrastructure companies are participating in the ecosystem, with deployments expected to begin in 2027.

That changes the economics of AI infrastructure.

If AI computing capacity can be assembled, tested and validated before reaching its destination, companies could potentially bring new capacity online much faster than through conventional data-center construction.

Electricity Is Now Part of NVIDIA’s Computing Strategy

Perhaps the clearest indication of how far NVIDIA has moved beyond GPUs is electricity.

The company increasingly treats power availability as part of the AI computing problem.

Its DSX AI Factory Platform is designed to coordinate the data center, computing infrastructure and electricity grid rather than managing them as entirely separate systems.

DSX MaxLPS, for example, is designed to maximize the number of computing workloads that can operate within an available power envelope.

Tests conducted by European cloud provider Nscale using GB300 NVL72 infrastructure showed that the technology enabled 37% more GPUs to operate within the same provisioned power budget while increasing aggregate inference throughput by nearly 50%.

NVIDIA is also working with energy and infrastructure companies including GE Vernova, Hitachi Energy and Schneider Electric on the systems surrounding AI factories.

This is an important shift.

The limiting factor for AI is increasingly not whether another GPU can be manufactured. It is whether the data center has enough electricity, cooling and grid capacity to switch it on.

Dubai Sits at the Center of NVIDIA’s Middle East Opportunity

That infrastructure question makes GITEX particularly relevant.

The UAE has become one of the world’s most ambitious markets for sovereign AI infrastructure, with large-scale investments designed to provide computing capacity inside the country.

NVIDIA is already part of that transformation through projects including Stargate UAE, which is being developed as part of a broader AI infrastructure initiative and is designed to use NVIDIA accelerated computing.

Regional technology groups are also expanding NVIDIA-powered infrastructure, while G42 is among the companies developing additional AI computing capacity.

The Gulf therefore offers NVIDIA something increasingly valuable: capital, energy infrastructure, government backing for AI development and demand for sovereign computing.

GITEX brings many of those buyers, policymakers, cloud operators and technology partners into one place.

Physical AI Could Be the Next Layer

There is another part of NVIDIA’s strategy likely to become increasingly visible in Dubai: physical AI.

The company is extending accelerated computing beyond language models into robots, autonomous machines, industrial systems and digital twins.

This is where technologies such as NVIDIA Omniverse, Cosmos and robotics platforms become important.

Instead of AI existing primarily inside software applications, physical AI allows machines to perceive their surroundings, understand instructions and act in the physical world.

For the Middle East, potential applications extend across logistics, manufacturing, energy, smart cities and autonomous transportation — sectors receiving significant investment across the region.

GITEX 2026 is therefore likely to provide NVIDIA with a platform to connect its AI factory infrastructure with what ultimately runs on it: enterprise agents, robots and increasingly autonomous industrial systems.

What to Expect From NVIDIA at GITEX 2026

NVIDIA has not disclosed a complete GITEX product-launch schedule, so it would be inaccurate to predict a specific new product announcement in Dubai.

But its latest technology roadmap provides a strong indication of what to watch.

Vera Rubin should be central to the infrastructure conversation as deployments move closer to commercial scale.

OpenShell could become particularly relevant to enterprises trying to move AI agents into production without giving them uncontrolled access to corporate systems.

DSX and DSX POD bring the discussion toward power and the speed of building AI factories, while NVIDIA’s physical AI portfolio extends the story beyond data centers into robots and industrial systems.

Taken together, these technologies reveal a company whose ambitions have expanded far beyond the component that made it famous.

NVIDIA spent the first phase of the AI boom convincing the world that accelerated computing was necessary to build artificial intelligence.

At GITEX 2026, the more interesting question will be whether it can convince governments and enterprises that the GPU is only one component of an AI factory — and NVIDIA should provide the architecture for everything around it.