On Wednesday in Berlin, NVIDIA set out what European businesses will be able to buy, rent and build with its technology over the next year. For anyone who’s been living under a rock for the last five years, NVIDIA makes the chips behind most of the world’s AI, so when it holds GTC Berlin, its big AI conference, it pays to listen. Its CEO, Jensen Huang, told the audience that Europe should be producing its own AI, not just consuming it. Sovereign AI, in other words (something I fundamentally agree with).
That’s a big ask for Britain, with some of the highest industrial electricity prices in the developed world and a queue for grid connections measured in years. But most of what Huang announced is about what businesses of any size can do with AI right now. Here are the five announcements that matter most and what each one could mean for your business.
01From next year, you can buy a ready‑built AI data centre and switch it on in weeks
What was announced
NVIDIA unveiled the DSX POD, a complete AI data centre built in a factory by partners including Dell and HPE and delivered in shipping containers. On a prepared site, you connect the power, water and network and switch it on. NVIDIA says that takes as little as two weeks, against around six months the usual way. The first ones arrive early next year.
It’s worth keeping in mind that nobody mentioned a price.
Why it matters to your business
If British operators buy them, it could mean more AI computing power in the UK, sooner. If you own a site with a good grid connection, that’s an opportunity. Speed matters because the sums are huge: Huang put a giant AI data centre at about $50 billion to build and used $40 billion a year as an example of what one might earn. That’s around $110 million for every day it sits unfinished.
The catch for Britain is power. This week Microsoft walked away from what was meant to be the UK’s biggest AI data centre because the grid connection won’t be ready in time.
02NVIDIA’s new Vera Rubin systems handle every AI job, from training a model to answering your customers
What was announced
Vera Rubin, NVIDIA’s newest generation of AI computers, is now in full production. Google Cloud announced the same day that businesses can rent it. Huang’s pitch is that the same machines can do every AI job, from preparing your data and training a model to answering your customers around the clock and running AI agents.
Why it matters to your business
Because one system covers every stage, you can go from a first trial to everyday use on the same kit. A machine that’s always busy also pays its way faster, which makes owning AI computing power, rather than renting it by the hour, easier to justify.
03NVIDIA offers free OpenShell software to keep AI agents safe and under your control
What was announced
An AI agent doesn’t just answer a question. It does the task. Ask a chatbot how to chase a late invoice and it will tell you. Ask an agent and it will send the email. To do that it needs access to your files, systems and tools. NVIDIA’s own video admitted that today the only thing stopping an agent misusing that access is “asking it nicely.”
OpenShell sets hard rules on what each agent can open, change and connect to. NVIDIA says those rules sit outside the AI, where the agent can’t talk its way round them. An optional layer called Sentry, which runs on a separate NVIDIA chip, can stop a misbehaving agent in milliseconds. Huang also said Anthropic, the company behind Claude, is working with NVIDIA on Claude agents that run inside OpenShell.
Why it matters to your business
Agents can take real work off your team’s hands, but only if you control what they can touch. The UK’s National Cyber Security Centre already tells firms to keep agents contained and always be able to stop them. OpenShell is young software too. A serious flaw was made public in August along with a fix.
So before you let an agent near your systems, ask the supplier what it can get into, who sets the rules and how you switch it off. If the answer is “trust us”, walk away.
04The rise of sovereign intelligence: free AI models, trained on your own data
What was announced
There are now several free, open AI models that anyone can download and adapt. NVIDIA’s own is called Nemotron, France has Mistral and there are plenty of others. Huang urged businesses to train one on their own contracts, drawings and order history. He said the result can be as good as the best AI at their particular jobs and called it “sovereign intelligence.”
To help, NVIDIA added new tools to its Agent Toolkit, which is open source, so free to use and change. Nemotron Retriever finds and reads the right documents. Kumo makes predictions from the numbers in your databases, from forecasts to fraud risk.
Why it matters to your business
Anyone can rent the same AI model you do. Nobody can rent your data, so that’s where your advantage now sits.
05NVIDIA wants one standard for robots and self-driving cars
What was announced
NVIDIA wants one standard design for self-driving cars and another for humanoid robots, so they improve quickly in the way phones and PCs did. Its car standard, called NVIDIA Hyperion, is a fixed set of cameras, radars and computers that car makers build in. Volkswagen, Mercedes-Benz and Volvo are among those building on it. With Munich’s Agile Robots, NVIDIA also announced Agile ONE SN, a standard humanoid robot for researchers, due early next year.
Why it matters to your business
Britain has far fewer robots in its factories than Germany. Huang’s aim is that once robots learn fast enough, a firm with no robotics team can simply show one what to do. That would put robots within reach of businesses that have never had one.
Our take
Britain will struggle to build big AI data centres until the grid catches up. But most of what Huang announced doesn’t need one. Any business can start now: get your data in order, pick one job worth handing to AI and make sure there’s an off switch.
Behind all five sit the same three decisions for any board: where AI should pay off first, what it should be allowed to touch and who owns what it learns from your data.
If you’re weighing up those decisions, talk to us.
Pictures: NVIDIA. NVIDIA’s own write-up of the keynote: blogs.nvidia.com.