NVIDIA Launches Revenue-Sharing Model With AI Cloud Partners to Build Always-On “AI Factories”

Picture a factory floor that never switches off. No shift changes, no downtime, no quiet Sundays. Just a continuous hum of servers generating billions of tokens — the building blocks of every AI-written sentence, every image conjured from a text prompt, every chatbot reply. That is the future NVIDIA is now openly building towards, and it has announced a new business model to match.

The company published details on its official blog of a plan to partner with what it calls “AI clouds” — cloud providers purpose-built for AI workloads — to deploy large-scale, multi-tenant AI infrastructure at a global scale. And NVIDIA won’t just be selling the hardware. It wants a cut of what those clouds earn.

From Training Runs to Always-On Inference

For years, the dominant image of AI computing was the training run: a vast, expensive, one-off process where a model like GPT-4 or Google’s Gemini was forged from mountains of data using thousands of GPUs over weeks or months. NVIDIA sold the chips. The cloud bought them. Job done.

But that picture is changing fast. Once a model is trained, it has to actually *run* — answering questions, generating images, powering autonomous AI agents — continuously, for millions of users, every hour of every day. This is inference, and it demands sustained, low-latency GPU compute on a scale that training alone never did.

NVIDIA’s blog frames this shift plainly: AI is moving from model development to “continuously operating AI factories that generate tokens at scale.” The company describes the exploding demand from generative and agentic AI applications — AI systems capable of acting autonomously across tools and services — as the engine driving a new phase of infrastructure build-out.

A New Deal: Revenue-Sharing and Credit Support

So what exactly is NVIDIA proposing? The new model works like this: AI cloud partners procure NVIDIA infrastructure, sell NVIDIA-powered services to their customers, and NVIDIA receives its standard product revenue — plus a share of the cloud’s ongoing revenue tied to supported capacity.

It’s a meaningful shift. Rather than a single hardware sale, NVIDIA gains a recurring, usage-linked income stream. The more tokens a cloud generates on NVIDIA silicon, the more NVIDIA earns. The company is also offering credit support to help partners finance the infrastructure build-out in the first place, lowering the upfront barrier for providers who want to enter the AI cloud market but lack the capital to do so alone.

The named partners in this ecosystem span the globe. They include Lambda and Yotta, GMI Cloud, Naver Cloud, Indosat Ooredoo Hutchison, YTL, Sharon AI, and Firebird — eight providers confirmed by NVIDIA as expanding AI factory capacity under this framework. Larger platforms such as Google Cloud and Oracle Cloud are also part of NVIDIA’s broader ecosystem, deploying hardware including the H100 and the newer Grace Blackwell GPU platform.

Opening the Door — Or Narrowing It?

NVIDIA says the model is designed to “open up compute access” to various organisations: startups, research bodies, enterprises, and regional AI players who might not have the time or money to build their own data centres. By plugging into an NVIDIA-backed AI cloud, they can skip the years-long process of site selection, power procurement, construction, and hardware setup.

Equinix and software firm Mirantis are among those working with NVIDIA on validated “AI Cloud Ready” designs — a kind of standardised blueprint for building a profitable AI cloud business around NVIDIA’s GPU stack.

Not everyone is entirely comfortable with the direction of travel, though. Some industry analysts have raised the question of whether tying so many cloud providers so closely to a single hardware vendor simply deepens NVIDIA’s already considerable grip on the AI compute market. If the infrastructure of the next generation of AI services runs overwhelmingly on NVIDIA silicon, the competitive effect on alternative chip makers — AMD, Intel, and a clutch of AI chip startups — are worth watching.

There are environmental questions too. AI factories, by definition, run continuously. Data centres already account for a big share of global electricity consumption, and always-on inference at scale will only add to that demand. NVIDIA has not, in the materials reviewed, addressed the specific energy footprint of this expanded model.

What the Numbers Don’t Yet Tell Us

It’s worth being clear about what NVIDIA has and hasn’t disclosed. The existence of the revenue-sharing arrangement is confirmed. The specific percentage split, the total financial volumes involved, and any precise breakdown of training versus inference compute demand are not publicly available. NVIDIA describes the inference shift in qualitative terms — “exploding token demand,” “accelerating compute needs” — but hard independent statistics on the training-to-inference ratio remain elusive.

NVIDIA is positioning itself not merely as a chip supplier but as a long-term economic partner in the AI cloud business. That’s a different kind of company from the one that simply shipped GPUs and moved on.

What This Means for Kent Residents

For businesses and organisations across Kent already using cloud platforms such as Google Cloud or Oracle Cloud for AI-powered tools — whether that’s data analysis, customer-facing chatbots, or document processing — NVIDIA’s expanded AI factory network could mean those services become more capable and more available over time, as the underlying infrastructure scales up. Tech startups and small firms in the county working with AI may also find it easier to access high-end GPU resources through cloud partners in this ecosystem, without needing to invest in costly local hardware. As with any expansion in large-scale data centre capacity, there will be broader questions for UK consumers and businesses around energy use and cost, though no specific impact on Kent’s power infrastructure has been identified in connection with this announcement.