Analyst(s): Brendan Burke
Publication Date: June 2, 2026
NVIDIA has announced a coordinated product strategy at GTC Taipei that reframes its AI factory approach around software control and purpose-built CPU acceleration for agentic workloads. These launches position NVIDIA as an architect of programmable data centers where operations software, not silicon alone, determines throughput, efficiency, and security posture.
What is Covered in This Article:
- NVIDIA’s four-product GTC Taipei strategy for agentic data centers
- Vera CPU as an orchestration accelerator rather than a general-purpose processor
- DSX software control plane shifting competition to tokens per megawatt
- BlueField-4 STX enabling software-defined zero-trust agentic storage
- Implications of full-stack software integration for buyer lock-in dynamics
The News: At GTC Taipei on June 1, 2026, NVIDIA announced a coordinated set of products that together define how agentic AI data centers are engineered through software control and CPU acceleration. The company placed the Vera CPU into full production, declared the Vera Rubin platform ramping into full production, extended Vera BlueField-4 STX into secure-by-design storage for agentic AI, and introduced the DSX platform as a software framework for AI factory design, deployment, and operations.
Vera Rubin unifies Vera CPUs, Rubin GPUs, Groq 3 LPX, BlueField-4 STX storage, and Spectrum-6 SPX Ethernet into a single five-rack POD, while DSX adds open source operations software and a token-per-megawatt optimization layer. Dell, HPE, Lenovo, and Supermicro will offer Vera in standalone CPU servers, with NYSE, Anthropic, OpenAI, ByteDance, CoreWeave, and Oracle Cloud Infrastructure named as early adopters. NVIDIA claims Vera delivers 1.8x faster task completion versus x86 on agentic workloads, citing Phoronix benchmarks across code compilation, Python, Java, and database processing.
Jensen Huang stated that NVIDIA is not just shipping chips but giving every infrastructure builder a playbook to build AI factories, framing the launches as an integrated architecture where software control spans silicon through grid-responsive energy management.
NVIDIA Engineers the Agentic Data Center with Software Control and CPU Acceleration
Analyst Take: NVIDIA used GTC Taipei to make its vision of the software-defined agentic data center explicit, positioning four product launches as layers of a single programmable architecture rather than discrete hardware announcements. The Vera CPU, DSX operations platform, BlueField-4 STX security layer, and Vera Rubin system together constitute an end-to-end software control plane that governs everything from agent orchestration to utility-grid interaction. This is a departure from NVIDIA’s historical positioning as a GPU vendor, and it reframes the competitive question from ‘who has the fastest chip?’ to ‘who controls the operational software that binds the data center into a coherent system?’ NVIDIA’s entry with Vera is timed to capture this architectural transition while simultaneously establishing the software dependencies that make its CPU indispensable to its broader factory vision.
Vera Positions the CPU as Software Orchestration Layer
Vera is not designed as a general-purpose server processor competing across all enterprise workloads against Intel and AMD incumbents that retain 45.3% and 29.7% data center CPU market share, respectively. Its 88-core Olympus architecture with LPDDR5X delivering 1.2 TB/s of bandwidth and second-generation NVLink-C2C connecting to Rubin GPUs at 1.8 TB/s is purpose-built for the software orchestration patterns of agentic AI, specifically tool-use dispatching, sandbox execution, and multi-step reasoning coordination. The 1.8x task-completion advantage NVIDIA claims is workload-specific to this agent control path rather than general-purpose computing, targeting the code compilation, Python execution, and database processing that form the substrate of agent lifecycles.
Arm-based CPU revenue growth demonstrates that the architectural shift away from x86 in data center compute is already underway at scale. NVIDIA enters this transition not as another Arm server chip but as an Arm orchestration chip specifically designed to accelerate the software plane that manages agentic workloads before they reach the GPU. The implication is that Vera’s competitive significance lies not in displacing x86 broadly but in establishing NVIDIA’s ownership of the software execution environment that sits between agent frameworks and accelerator hardware.
DSX Transforms Operations Software into a Competitive Axis
DSX reframes AI factory competition from cores per dollar to tokens per megawatt, introducing a software control plane that spans reference designs, DSX Sim digital twin, DSX OS open source operations software, and the MaxLPS efficiency layer. MaxLPS combines liquid cooling with software-driven in-rack power tuning to run up to 40% more GPUs at their most efficient operating point inside a fixed power budget, making power management a software function rather than a hardware constraint. DSX Flex extends this control to the utility grid, with a multi-megawatt pilot alongside Emerald AI and Silicon Valley Power that adjusts factory consumption programmatically in response to grid signals.
The open sourcing of DSX OS is NVIDIA’s most strategically nuanced move, positioning the operations layer as an industry standard while engineering it to extract maximum performance from NVIDIA silicon specifically. DSX makes software-controlled operations the layer where NVIDIA’s integration advantage compounds, because optimizing tokens per megawatt across heterogeneous silicon is exponentially harder than optimizing for a known hardware stack.
BlueField-4 STX Extends Software Control into the Security Plane
BlueField-4 STX and DOCA transform the storage and network data path into a software-controlled security plane, enforcing zero-trust file access and runtime threat detection at line rates of 800 Gb/s, which NVIDIA claims is up to 1,000x faster than agentless security tools. In an agentic data center where autonomous agents read, write, and modify data without human supervision, the security enforcement point must exist in the data path itself rather than in separate appliances or host-based agents that introduce latency and blind spots. DOCA, as a software framework running on BlueField-4 STX hardware, means that security policies become programmable, updatable, and enforceable at wire speed without consuming host CPU cycles that Vera needs for agent orchestration.
This architectural choice connects directly to the broader software-control thesis because it removes security from the general-purpose compute domain and places it under a dedicated software runtime that NVIDIA controls end-to-end. The integration with Vera Rubin’s five-rack POD architecture means that storage security, network security, and compute orchestration are all governed by NVIDIA software frameworks operating on NVIDIA hardware across every tier. Buyers adopting BlueField-4 STX are not merely purchasing a storage controller but accepting NVIDIA’s security software framework as the enforcement layer for their entire agentic data path.
Full-Stack Software Integration Creates Both Moat and Target
The software layers spanning Vera CPU orchestration, DSX operations, and DOCA security create switching costs that compound across every tier simultaneously, making displacement a systems problem rather than a chip problem. However, the largest buyers, specifically hyperscalers and frontier labs that appear on Vera’s early adopter list, are precisely the organizations most capable of funding alternatives at individual layers, whether custom ASICs, in-house CPUs, open networking through OCP, or disaggregated inference architectures.
The fuller NVIDIA’s software stack becomes, the more valuable any single defective layer becomes to a competitor seeking to break the integration dependency. NVIDIA’s counter-strategy is the open sourcing of DSX OS, which allows the company to claim openness at the operations layer while engineering performance optimization that routes economic gravity back to NVIDIA hardware underneath. NVIDIA’s software-defined factory is simultaneously its deepest moat against chip-level competition and its largest surface area for coordinated ecosystem responses targeting individual software layers rather than silicon itself.
What to Watch:
- Whether DSX OS attracts genuine multi-vendor contributions or remains optimized primarily for NVIDIA silicon
- Independent third-party validation of Vera’s 1.8x task-completion claim across diverse agentic workloads, and the competitive response from Intel and AMD in purpose-built agent-orchestration CPU designs.
- Adoption velocity of BlueField-4 STX as the security enforcement layer versus incumbent storage security architectures, particularly among organizations already invested in alternative DPU ecosystems.
- Whether DSX Flex grid-responsive pilots expand beyond the initial Emerald AI partnership into broader utility-scale deployments
See the full press release on NVIDIA’s Vera Rubin production and DSX platform announcement on the company website.
Disclosure: Futurum is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.
Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of Futurum as a whole.
Other Insights From Futurum:
NVIDIA Q1 FY2027: Data Center Diversification, Blackwell Scale, CPU Upside
At GTC 2026, NVIDIA Stakes Its Claim on Autonomous Agent Infrastructure
NVIDIA GTC 2026 Day 1 – Can NVIDIA’s Ecosystem Accelerate the Inference Inflection?
Image Credit: NVIDIA
Author Information
Brendan is Research Director, Semiconductors, Supply Chain, and Emerging Tech. He advises clients on strategic initiatives and leads the Futurum Semiconductors Practice. He is an experienced tech industry analyst who has guided tech leaders in identifying market opportunities spanning edge processors, generative AI applications, and hyperscale data centers.
Before joining Futurum, Brendan consulted with global AI leaders and served as a Senior Analyst in Emerging Technology Research at PitchBook. At PitchBook, he developed market intelligence tools for AI, highlighted by one of the industry’s most comprehensive AI semiconductor market landscapes encompassing both public and private companies. He has advised Fortune 100 tech giants, growth-stage innovators, global investors, and leading market research firms. Before PitchBook, he led research teams in tech investment banking and market research.
Brendan is based in Seattle, Washington. He has a Bachelor of Arts Degree from Amherst College.
