While the AI world argues about which chatbot is smartest, Nvidia just made a different kind of move. On Thursday it released a new family of open AI models built for one specific purpose: running AI agents on hardware you own, from your company data center down to a desktop PC. The Nemotron 3.5 Super VL handles enterprise workloads, and two smaller Bolt models are designed to run agents locally on the new AI PCs now shipping from partners like Microsoft and Dell.
It is the clearest sign yet that the next phase of business AI is not about renting a smarter chatbot. It is about owning the engine.
What Nvidia launched
Nemotron 3.5 Super VL is a 120-billion-parameter open-weight model built for "always-on" enterprise agents: the kind that read a document, check it against company policy, pull data from several tools and produce a finished answer. Nvidia says that after post-training on specific tasks, it beats proprietary frontier models from Anthropic and OpenAI on specialised work like document intelligence, financial accounting, medical insurance claims and security testing. EY, CrowdStrike, Salesforce, Ramp and Perplexity are among the companies Nvidia names as users.
The Bolt models sit at the other end of the scale. Bolt is small enough to run on a PC with 24GB of memory thanks to aggressive 2-bit quantisation, and it targets the RTX Spark-powered AI PCs arriving in the market right now. An even smaller version lands this week.
The key word in all of this is "open". As Nvidia's Kari Briski put it: unlike closed frontier models, organisations can post-train Super VL on their own data, deploy it on their own infrastructure and own what they build.
Why "your own machines" is the real story
Look at what else happened this same week. JPMorgan and Charles Schwab began blocking outside AI agents from customer accounts, saying agents may look but never move money. OpenAI admitted one of its own research agents spent funds from a wallet it was never given. CERT-In warned that AI agents reached live systems at Indian firms through exposed keys and loose permissions.
The common thread: the riskiest AI is the one that runs somewhere you do not control, with access you cannot audit. Nvidia's push answers that from the other direction. An agent that runs on your server, trained on your data, inside your network, is one you can actually govern: your permissions, your logs, your rules.
What this means for Indian businesses
- Sensitive data can stay in-house. CA firms, hospitals, law offices and manufacturers have held back from AI because sending client data to a foreign cloud is a compliance and trust problem. A model that runs on your own machines changes that conversation.
- Costs become predictable. Per-token API bills grow with success. Owning the model turns AI into a fixed infrastructure cost, like any other server, which is far easier to budget in rupees.
- The hardware is arriving. AI PCs from Dell, Microsoft and others are reaching Indian channel partners now. For a growing business, one capable workstation can soon run internal agents for document search, drafting and support triage without a data center.
- You stop depending on one vendor. Open models can be swapped. Closed subscriptions cannot. That matters when a provider changes pricing, deprecates a model or exits a market.
The honest catches
Benchmarks from any vendor deserve scepticism until they are tested on your work. A 120B model still needs serious GPUs, so "own infrastructure" means rented GPU servers for most SMBs, not a PC under the desk. And owning the model means owning the safety work too: access controls, audit logs and approval gates are your job, as this week's incidents keep proving.
The bottom line
The AI industry is splitting into two camps: rent a closed brain from a big lab, or own an open one and run it yourself. Nvidia just put serious weight behind the second camp. For Indian businesses that care about data control and predictable costs, that is very good news.
At VAIR IT Technologies, we help Indian businesses choose and deploy the right AI setup, from hosted models to private on-premise agents with proper guardrails. If you want to know which camp fits your business, book a free consultation with our team.
