Nvidia Ditches the One-Size-Fits-All Model: Equity Stakes, Desktop Supercomputers, and a New Breed of Cloud Partners

Nvidia shifts from chip sales to 'AI factories,' taking a recurring cut of cloud revenue. Early deals include massive data centers in Australia and Indonesia; stock dips 2% but up 5% YTD.

Nvidia is no longer content to simply sell shovels in the AI gold rush. In a sweeping reorganisation of its business, the chipmaker now demands a slice of its customers' future revenue in exchange for access to its prized hardware. The shift arrives as the market grows jittery — the stock slipped 2% on Thursday to €170.20, though it still clings to a year-to-date gain of around 5%.

The new playbook, dubbed “AI factories,” moves beyond the traditional hardware-sale model. Nvidia provides specialised chips and credit lines to cloud providers, then collects a recurring cut of the eventual cloud revenue. The strategy aims to lock in steady income from the worldwide infrastructure buildout and, crucially, to loosen the grip of hyperscalers like Amazon, Microsoft and Google that currently dominate the market.

Two early partners illustrate the ambition. Sharon AI is building a data centre in Australia that will house more than 55,000 of Nvidia's new Blackwell-generation graphics processors by mid-2027. Firmus Technologies, meanwhile, is constructing a massive campus in Indonesia capable of hosting up to 170,000 processors; the company expects offtake agreements worth as much as $30 billion over the first years of operation.

A separate initiative, the “AI Compute Partner” programme, provides small cloud providers with financial backing to buy Nvidia chips, along with buyback guarantees for unused capacity. In return, Nvidia shares in the operators' future earnings. The goal is to cultivate demand beyond the tech giants that have fuelled Nvidia's recent explosive growth.

On the desktop front, Nvidia is expanding its reach with the DGX Station for Windows, a local supercomputer powered by the GB300 Grace Blackwell Ultra Superchip. Delivering up to 20 petaflops of performance, the system lets sensitive industries in Europe and North America run AI models on-premises, bypassing the cloud. Dell and HP plan to begin shipments by late 2026.

Sovereign AI infrastructure is another pillar. The Australian partnership with Sharon AI is initially structured as a six-year arrangement using up to 40,000 Grace Blackwell chips — a slightly different configuration from the Blackwell-heavy Australian project mentioned above, reflecting Nvidia's willingness to tailor deals to local conditions.

The strategic pivot comes as the market digests potential competition from an unexpected quarter. Reports that Meta is creating a unit called “Meta Compute” to sell its spare data-centre capacity rattled Asian chipmakers on Thursday: SK Hynix plunged 14%, Samsung lost roughly 9%, and cloud provider CoreWeave also suffered steep losses.

Nvidia is backing its bet with muscle. Its portfolio of stakes in private AI companies has ballooned to $42.3 billion, up from just $3.4 billion a year ago. The company named Nicholas Parker as its new global sales chief, effective August. Parker arrives from Microsoft after 26 years and receives a stock package worth $35 million.

The financials remain formidable. In the last fiscal quarter, Nvidia posted record revenue of $81.6 billion, almost entirely from data-centre operations. For the current quarter, management targets $91 billion — a figure that explicitly excludes any revenue from China due to ongoing trade restrictions.

On the charts, the technical picture is mixed. The stock recently traded at €173.74, well off the record high of €202.50, and has slipped below its 50-day moving average. Yet the longer-term 200-day line at €164.11 continues to provide support, keeping the multi-year uptrend intact. Investors now face a question the data alone cannot answer: whether a company that once minted money selling picks and shovels can succeed by taking a seat at the miners' table.