According to INSIDE, the new cloud version of Hermes Agent compresses the launch flow into two clicks and about 60 seconds, and it emphasizes handing an agent to multiple team members at once 1.
The same report offers a caveat: "two clicks, 60 seconds" is the kind of phrasing that "usually carries a marketing element at product launch," and the real-world experience is still shaped by network conditions and the first-time authorization flow 3.
| Claim | Detail | Source |
|---|---|---|
| Launch flow | Two clicks, ~60 seconds | INSIDE 1 |
| Framing caveat | Marketing element; affected by network and first authorization | INSIDE 3 |
According to INSIDE, what truly separates this update from earlier ones is "team-scale deployment" — the ability to roll an agent out across a team rather than one machine at a time 2.
INSIDE also notes what is still missing: pricing, how data is handled, and which models are supported have not been disclosed, and these are precisely the questions teams should ask before adopting the tool 4.
According to Business Next, at the GTC Taipei keynote Jensen Huang declared "practical AI has arrived," presenting a full software-and-hardware agent infrastructure so that every enterprise, every employee, and even every individual can use AI agents 5.
The same report gives NVIDIA's definition of an AI agent: agent = LLM + harness, meaning a foundation model plus an orchestration backbone called a harness 6.
According to Business Next, on top of the open-source agent framework OpenClaw — which has faced security concerns — NVIDIA released an "enterprise security version" called NemoClaw, packaging the model and harness into a runnable environment, paired with a security sandbox "OpenShell" that sets action policies and privacy controls 8.
For those who want data kept on their own hardware, the same report says NVIDIA provides three local devices — DGX Spark, RTX Spark, and DGX Station — so users can place sensitive data and personal AI agents on self-owned hardware 11.
According to Business Next, on the model side, Nemotron 3 Ultra — a mixture-of-experts model with 550 billion parameters — went live on June 4. It was post-trained for mainstream harnesses such as OpenClaw and OpenHands, supporting long-running agent capabilities across multiple application scenarios 9.
On the hardware side, the same report says NVIDIA introduced a CPU "built for agents" named Vera, focused on high-speed, high-efficiency agent-type computation, data processing, and task orchestration 10.
| Product | Type | Detail | Source |
|---|---|---|---|
| Nemotron 3 Ultra | Model | 550B-parameter MoE, live June 4 | Business Next 9 |
| Vera | CPU | Built for agents | Business Next 10 |
| DGX Spark / RTX Spark / DGX Station | Local devices | Keep sensitive data on own hardware | Business Next 11 |
According to Business Next, companies including Cadence, Siemens, Synopsys, and Dassault Systèmes have already built autonomous AI agents for their workflows 12.
The same report adds that Foxconn announced it is driving its own Nurabot and CoDoctor platforms via the NemoClaw architecture, using multiple teams of AI agents to support care coordination, document processing, and related work 13.
According to Business Next, Nader Khalil described a personal case where an agent saved him about US$120 in medical checkup costs, joking, "I don't know how it did it. But the roughly $120 saved is just enough to pay a fine" 7.
The same report shares another example: Khalil set up an agent on an iPad for a colleague's daughter, and the ten-year-old built a children's social platform they called "KidsGram" 15.
According to Business Next, Justin Boitano said the software development process is being reshaped by AI agents, adding: "I think this is the transformation every enterprise is going through — that is, agents can do productive work" 14.
The two reports sit at different scales but point the same direction. INSIDE describes Hermes Agent narrowing the launch flow to two clicks and about 60 seconds while centering on team-scale deployment (E1, E2) — yet leaving pricing, data handling, and supported models undisclosed 4. Business Next, meanwhile, shows NVIDIA defining an agent as LLM + harness 6 and shipping that stack end to end: the 550B-parameter Nemotron 3 Ultra live June 4 9, the Vera CPU 10, the NemoClaw enterprise build with the OpenShell sandbox 8, and three local devices 11. The adoption evidence — Cadence, Siemens, Synopsys, Dassault Systèmes, and Foxconn's Nurabot and CoDoctor (E12, E13) — alongside the US$120 saving and the ten-year-old's KidsGram (E7, E15), shows the same theme INSIDE raises for Hermes: the friction of deployment, not the model, is what the products are competing to remove.
According to INSIDE, the cloud version compresses launch into two clicks and about 60 seconds, though INSIDE notes this phrasing carries a marketing element and real experience depends on network and first-time authorization (E1, E3).
According to INSIDE, pricing, how data is handled, and which models are supported have not been disclosed (E4).
According to Business Next, it is a mixture-of-experts model with 550 billion parameters, which went live on June 4 (E9).
NVIDIA reported Q2 FY2027 revenue of $96.2 billion, up 106% year-over-year and above the roughly $92 billion consensus, with Data Center revenue of $89.0 billion, up 117% and about 92% of total sales. Gross margin held at 75.0% and non-GAAP EPS of $2.22 beat the $2.10 estimate. Q3 guidance of $108.0 billion excludes any China Data Center compute revenue.
SpaceX plans to launch its first NVIDIA-chip-powered AI satellites, Starmind AI1, in the fourth quarter of 2027, reaching large-scale deployment by 2028 — at least a year ahead of the original schedule. The company has also filed with the FCC for a network of up to 1 million satellites, while Taiwanese suppliers Unitech and Sesoda report rising satellite-related revenue and order shares tied to the buildout.
NVIDIA announced the Jetson Orin Nano 2 robotics computer on August 25, 2026, packing 78 TOPS of AI compute, 8GB of memory, and an 8-core Arm CPU while doubling inference performance and cutting power draw 40% versus its predecessor. Modules and developer kits ship in the first half of 2027, targeting a robotics developer base NVIDIA says already tops 3 million.