Nvidia has paused parts of a finance scheme that swapped credit support for AI cloud providers’ future revenue.
According to the Wall Street Journal the programme offered credit backing to artificial intelligence cloud providers in return for a cut of their takings. Apparently, some Nvidia staff warned current and potential customers that the setup could attract antitrust scrutiny.
They raised concerns about how far the chip giant could go in telling customers how to run their businesses. Nvidia stepped back last week, less than two months after announcing the wheeze.
The precise reason for the retreat could not be learned, and Nvidia may revive the scheme or fold it into another programme. The move comes as Nvidia faces sharper scrutiny over using its balance sheet to prop up projects that then buy its chips.
The outfit recently trimmed a proposed financial backstop for OpenAI’s huge Ohio data centre plan, amid fears about how investors would view the liability.
“The new business model we introduced in July that opens up compute access to the fast-growing AI ecosystem is still in place and continues to evolve due to high demand,” an Nvidia spokeswoman said.
The arrangement, announced in July, gave Nvidia two bites at the cherry. It could sell the chips first, then collect part of the revenue when customers rented the same kit.
The first cloud outfits named in the finance scheme were Sharon AI and Firmus Technologies, according to Nvidia’s announcement. In the first few weeks, Nvidia irritated some potential partners with the amount of control it sought, the people said.
The company told some providers they could rent chips only to approved customers. It indicated it preferred capacity spread among several smaller AI companies rather than leased to one large customer, the people said.
Some cloud providers pushed back, saying they should be free to choose their own customers.
The programme, called the AI Compute Partnership, was designed to fix a financing headache for smaller cloud providers. Building an AI cloud means spending billions of dollars on Nvidia GPUs and data centres before enough firm customer contracts exist.
Nvidia tried to plug that gap by promising to rent the GPU capacity itself if the provider could not find another customer. That promise provided providers with guaranteed revenue and made it easier to borrow money for infrastructure.
Nvidia disclosed the size of the programme for the first time in its quarterly filing this week. The company said it had made $36 billion of commitments under agreements that usually lasted six years.
It noted the commitment would shrink as cloud providers sold capacity to other customers. Nvidia chief financial officer Colette Kress mentioned the revenue-share programme on Wednesday in a call with investors.
Kress said the new revenue stream could generate billions of dollars for the company over the medium- to long-term. Cloud providers generally rent capacity to customers for a few dollars an hour per GPU. Under the proposed deals, Nvidia and each provider would set a base hourly GPU rate covering the provider’s costs. Those costs included chip depreciation, data centre spending and staff, according to people familiar with the deals. Nvidia would receive 50 per cent of any revenue cloud providers earned above that threshold on the GPUs.
