The Race to Put AI Agents Everywhere

What It Covers

The enterprise race to productize AI agents accelerates in Q2 2025, driven by NVIDIA's NemoClaw security toolkit, OpenAI's refocus on coding and enterprise productivity, and a wave of desktop agent launches from Manus, Adaptive, and Perplexity all competing to make agentic systems viable at scale.

Key Questions Answered

• •NemoClaw Enterprise Security: NVIDIA's NemoClaw toolkit adds policy-based access control and isolated sandboxing to OpenClaw, making it hardware and model agnostic. Enterprises blocked by security concerns now have a credible path to deployment. Jensen Huang compared OpenClaw's industry timing to Linux and Kubernetes — foundational infrastructure that the entire software industry can build on.
• •OpenAI Enterprise Pivot: Applications CEO Fiji Simo declared an internal code red, explicitly telling staff to abandon side-quest products and concentrate on enterprise coding productivity. GPT-5.4 reached 5 trillion tokens per day within one week of API launch, hitting an annualized $1 billion in net new revenue — the fastest model ramp in OpenAI's history.
• •Desktop Agent Design Pattern: Manus, Adaptive, and Perplexity all launched local desktop agent products within 24 hours, converging on the same architecture: a persistent AI agent bridging local files and cloud systems. Adaptive's "encoded memory" feature retains learned workflows — how specific apps work and user preferences — to automate recurring tasks without re-prompting.
• •Chinese Open Source Shift: Alibaba folded its Qwen research team into a revenue-focused division called Alibaba Token Hub, directly led by CEO Eddie Wu. Separately, Z.ai released its first closed-source model, GLM-5 Turbo. The emerging pattern: lightweight open-source models drive developer adoption while more capable, agent-optimized models are kept proprietary for enterprise monetization.
• •AI Infrastructure Capacity Crunch: Meta signed a $27 billion, five-year deal with NeoCloud provider Nebius — roughly ten times Nebius's entire prior revenue — to deploy NVIDIA Vero Rubin chips. The scale signals that hyperscalers are capacity-constrained enough to partner with smaller specialized data center operators, accelerating NeoCloud growth across the entire infrastructure tier.