Nvidia GTC 2026 underscored surging AI infrastructure demand and ecosystem growth. Tokenomics and inferencing are emerging as key drivers of future value.
Nvidia GTC 2026 was bigger and broader than ever, reflecting the continued, relentless growth of demand and rapid pace of change and innovation in the AI sector. Nvidia has placed itself at the center of it all, as the key supplier of a growing number and variety of xPUs and the related systems that underpin the most advanced AI factories.
Equally important is the size of the ecosystem that Nvidia has cultivated across the full AI stack, as reflected by the 400+ exhibitors on the exposition floor.
Nvidia GTC highlights
Nvidia CEO Jensen Huang highlighted several important developments from the Nvidia mainstage keynote, including the following:
NemoClaw overview. NemoClaw, a safe environment for OpenClaw built on OpenShell, offers enterprises a path to safely use the approach to building and running AI agents.
Token production efficiency. Incorporating Nvidia Groq 3 LPX into the AI factory substantially extends token production velocity at the high end of the range, overcoming the inherent limitations of NVL72.
Storage and performance optimization. STX, the new storage reference architecture enabled by BlueField-4, reinforced data storage as an essential part of the mix that must be optimized -- along with compute and networking -- to continue driving capacity up and incremental cost down.
Revenue growth. Nvidia predictsit could generate $1T in visible revenue potential by the end of 2027. Growth will partly be due to more AI Gigafactory builds, but increasingly it will be driven by the rapid expansion in infrastructure demand for distributed inferencing. Nvidia believes the long-anticipated pivot point has been reached, and training will continue to drive revenue; AI inferencing will drive them faster from this point forward.
Tokenomics as a driver for AI business cases and ROI
The AI sector is focused on tokenomics, a term with roots in blockchain and cryptocurrency but applicable to AI as well. In the world of AI, tokens are the unit of production. The capacity and performance of AI systems are measured in terms of time to first token (TTFT) and tokens per second.
Contextual data management is measured by token retention rate. There are even efforts underway to track value per token to understand and balance quality. And, increasingly, AI services are being metered and licensed in tokens.
According to Huang, every chief experience officer should be thinking about how much token capacity to grant each employee, and employees should be measured by how well they use their token budgets. This concept represents a means by which AI technology usage can be measured, managed and translated into business value.
Proper management and review of tokens, as with any other scarce resource, separate good use from bad. The bottom line is that everyone needs to start thinking in terms of tokenomics for business cases, KPIs and ROI across the ecosystem, from both supplier and user perspectives.
Ecosystem highlights
The conference featured many press releases and announcements from vendors in Nvidia's ecosystem. Findings from discussions related to enterprise networking infrastructure include the following:
Akamai
Akamai announced support for the Nvidia AI Grid reference architecture through its Akamai Inference Cloud, based on the deployment of thousands of Nvidia Blackwell GPUs and BlueField DPUs across its 4,400 global edge locations.
Arrcus
Arrcus promoted its recently announced Arrcus Inference Network Fabric, a purpose-built networking approach for optimizing the delivery of inferencing workloads across distributed environments, now integrated with Nvidia Dynamo, Nvidia BlueField-3 DPUs and Nvidia Spectrum-X Ethernet.
Cisco
Cisco announced several new products related to its Secure AI Factory initiative to simplify network design, deployment and operations. Products include the following:
UCS and Unified Edge support for Nvidia RTX Pro Blackwell GPUs.
Reference architecture support for AI Grid.
A new N9100 switch powered by Nvidia Spectrum-6 silicon.
Integrated support of that same N9100 switch series within Cisco Nexus Hyperfabric.
On the security side, Cisco announced the extension of its Hybrid Mesh Firewall for policy enforcement on Nvidia BlueField DPUs connected to Cisco Nexus ONE fabrics, as well as new purpose-built guardrails for AI agents within Cisco AI Defense, including support for Nvidia OpenShell for safer OpenClaw deployments.
Cloudflare
Cloudflare was highlighting its offerings to directly support the development of stateful serverless AI agents, using its global Connectivity Cloud network reach and Agent Developer Platform, which includes a deep model repository, durable objects, and egress-free R2 storage for optimizing both cost and performance.
Dell
Dell continued to advance its AI Factory, with over 4,000 customers now in deployment and early adopters seeing ROI increases of nearly three times in the first year. The product uses Dell's ability to bring together compute, storage and networking, including PowerEdge servers that use both Nvidia HGX Rubin NVL8 and Vera Rubin NVL72 GPUs for high-end training, as well as RTX Pro 4500 Blackwell GPUs and the new Vera CPU for enterprise workloads in the data center.
Equinix
Equinix highlighted its recently launched Distributed AI Hub, part of the Equinix Fabric Intelligence initiative, and its first direct partnership with Palo Alto Networks to simplify and secure agentic AI deployments.
The bottom line is that everyone needs to start thinking in terms of tokenomics for business cases, KPIs and ROI across the ecosystem.
F5
F5 has turned its focus onto improving tokenomics outcomes -- such as TTFT, token throughput and cost per token -- by integrating its BIG-IP Next for Kubernetes with Nvidia BlueField-3 DPUs. Given that most AI applications are built using cloud-native architectures, this approach acts as a control plane that optimizes results in a manner closely aligned with where workloads reside and execute.
HPE
HPE announced new Nvidia-related capabilities across compute, networking and software realms, set to be available in 2027. Examples include integration of Nvidia Vera CPU blades and Nvidia Quantum-X800 InfiniBand networking into the HPE Cray Supercomputing GX5000 platform.
HPE also updated its turnkey HPE Private Cloud AI offerings to expand network capacity and introduced ProLiant servers with Nvidia RTX Pro 6000 Blackwell GPU support. Additionally, the company also integrated Nvidia AI-Q and Omniverse into its design blueprints, offering self-contained options for sovereign AI factory deployments.
IBM
IBM relayed results of using Nvidia cuDF to accelerate its watsonx.data's SQL engine Presto to dramatically accelerate terabyte-scale data refresh cycles for a global foods supplier. This yields a remarkable 30x improvement in price-to-performance over prior processes. IBM also promoted its certification of IBM Storage Scale 6000 with Nvidia DGX platforms to address ongoing challenges with AI data management and storage.
Lenovo
Lenovo has been working to deliver AI infrastructure in multiple forms, from Gigafactory architectures to laptops and workstations powered by Nvidia RTX Pro Blackwell GPUs for local inferencing. It also launched an industry-vertical platform focused on improving fan experience, revenue growth, business performance and operational efficiency for global sports organizations.
Nokia
Nokia announced a slew of new optical components and amplifiers for use within and between data centers, coming available later this year and into 2027.
Nutanix
Nutanix focused on addressing challenges with deploying AI workloads in mixed environments, where both cloud-native and virtual compute components will be present. The company also announced its Nutanix Agentic AI platform, a full software stack designed to help accelerate agent development, security and operational deployment.
WEKA
WEKA announced the general availability of the NeuralMesh AI Data Platform, which aims to make scaling high-performance data pipelines seamless as AI workloads move from PoC to production. WEKA also announced the Augmented Memory Grid, which provides accelerated data pipelining for existing Nvidia-based AI factories, similar to how the new Nvidia STX architecture will work with future reference architectures.
The AI wave continues
When attending an event built around an individual technology vendor, it's usually easy to see through even the most carefully constructed frameworks of stories and promises. But when those stories come from a chorus of surrounding voices, it's much more convincing.
Getting the market timing right is one thing, and it has a lot to do with Nvidia's success, following years of tireless devotion to PC GPU cards. But one must also credit Nvidia's vision and execution in building and nurturing an ecosystem that can sustain and compound opportunities, sharing both the risks and the rewards.
Lingering questions about whether the massive AI buildout is a bubble or a wave remain. Bubbles can grow very large but fail to deliver and pop when overhyped based on hype rather than measurable value. Waves also start large but carry on due to a successful translation into outcomes.
AI is a big technology wave, and it's not showing any signs of slowing. In fact, quite the opposite is happening -- it's still building. The positive value outcomes are already being documented, and we now stand on the threshold of the era of AI inferencing, where the real and measurable values start rolling in, in ever-increasing volume.
Jim Frey covers networking as principal analyst at Omdia.
Omdia is a division of Informa TechTarget. Its analysts have business relationships with technology vendors.
