NVIDIA's BlueField-3 data processing units now run Fortinet's FortiGate VM directly on-chip, enabling line-rate security enforcement without impacting GPU workloads.1 Kevin Deierling of NVIDIA stated that "AI factories demand an entirely new class of secure, accelerated infrastructure," positioning DPUs as critical components in next-generation data center architectures.2
The integration allows organizations to deploy firewalling, network segmentation, and zero-trust policies at infrastructure level rather than through traditional appliances.3 This offloading model frees GPU resources for compute-intensive AI tasks while maintaining security isolation—a key requirement as enterprises scale training and inference workloads.
Hyperscale operators are converting existing facilities and expanding AI-specific infrastructure. Milton Ault III characterized 2026 as "an important year for Hyperscale Data," projecting growth from organic AI infrastructure expansion and digital platform operations.4 The company expects returns on multi-year capital investments in infrastructure, software platforms, and digital ecosystems to materialize through scaled revenue and improved operational efficiency.1
The buildout reflects broader enterprise adoption patterns. Adobe and UiPath are integrating AI capabilities into workflow automation tools, while agent-based systems like AgentMail are emerging as infrastructure consumers. GPU cloud platforms are scaling capacity to meet this demand, with providers prioritizing high-density compute configurations.
The architectural shift favors vendors offering integrated hardware-software stacks. NVIDIA's combination of GPUs, DPUs, and networking fabric positions it across multiple infrastructure layers. Security vendors like Fortinet gain access to new deployment models, while hyperscale operators face capital intensity trade-offs between retrofitting legacy sites and building greenfield AI facilities.
Market implications center on infrastructure spending mix. Traditional server and storage budgets are reallocating toward accelerated computing and specialized silicon. Companies with legacy data center exposure face margin pressure, while those aligned with AI-optimized architectures capture premium pricing. The transition timeline remains uncertain, but 2026 capital deployment patterns will signal whether current projections materialize into sustained revenue growth.
