Envision Green will pay RMB380 million to Inventec subsidiary Shanghai Shunquan for more than 100 Nvidia-GPU servers, funding a cloud and intelligent-computing build-out whose revenue contribution has not yet been disclosed.
뒤로
Envision Green spends RMB380 million on 100 Nvidia servers
Envision Green (01783.HK) agreed to pay RMB380 million for more than 100 high-performance servers from Shanghai Shunquan Technology, an Inventec subsidiary, to build out cloud computing and intelligent computing services, the company said Wednesday. The servers carry Nvidia graphics processors, the same class of hardware that underpins the data-center build-outs of Oracle and Microsoft.
The order is a purchase agreement, not a customer contract. Envision Green did not disclose the delivery schedule, the specific Nvidia GPU model, or any revenue the servers are expected to generate, leaving the return on the RMB380 million outlay unquantified.
"Buying GPUs is the easy part," said Rachel Kim, an analyst covering AI infrastructure at Edgen. "The question every investor should ask is who is paying for the compute once it is racked — a signed customer contract is what converts capex into revenue."
What RMB380 million buys
At RMB380 million, the order implies an average cost of roughly RMB3.8 million per server, a figure consistent with eight-GPU configurations of Nvidia's data-center accelerators. For scale, Nvidia's H100 carries a list price near $30,000 per unit, and a single eight-GPU HGX H100 system typically prices between $250,000 and $400,000 depending on memory and interconnect options, according to published Nvidia specifications and server-vendor pricing.
Envision Green trades on the Hong Kong Stock Exchange and closed up 1.18%, or HK$0.05, on the day of the announcement, with short selling of HK$2.90 million representing 13.2% of turnover, according to AASTOCKS data. The modest move suggests the market treated the order as an incremental step rather than a re-rating event.
The supplier side carries more weight. Shanghai Shunquan is a unit of Taiwan-listed Inventec, one of the contract manufacturers that assemble AI servers for Nvidia's largest customers. Inventec's AI server business has been the primary driver of its revenue growth as hyperscalers expand capacity, and the Envision Green order adds a mainland Chinese buyer to a customer base historically dominated by US cloud providers.
The sector frame: backlog is converting
The order lands in a market where AI infrastructure spending has moved from promise to reported revenue. Oracle said last week it booked more than $30 billion of additional AI cloud contracts in its first fiscal quarter, lifting its revenue backlog to $664 billion, above the $639.89 billion analysts expected, according to Visible Alpha data. Oracle's quarterly revenue rose 30% to $19.3 billion, beating estimates of $19.14 billion, and the company said most of the new contracts were prepaid or brought their own hardware.
That distinction matters for Envision Green. Oracle's finance chief, Hilary Maxson, said the vast majority of new orders "were via prepay or bring-your-own-hardware or a similar mechanic, so they don't require incremental capital from Oracle." Envision Green is taking the opposite path — spending its own capital first and seeking customers afterward. Oracle expects roughly half its backlog to convert into sales within 36 months; Envision Green has given no equivalent timeline.
Microsoft's plan for 38 gigawatts of data-center capacity by 2032, reported by Bloomberg News, and the Pentagon's talks to lend $5 billion to AI cloud startup Fluidstack, reported by the Wall Street Journal, show the demand pool is widening beyond the largest hyperscalers. Smaller operators are entering the compute-rental market, which is where Envision Green's servers would compete.
What investors should watch
The near-term arithmetic is unfavorable. A RMB380 million outlay against an undisclosed revenue stream means depreciation and financing costs hit the income statement before any compute revenue arrives, and GPU economics deteriorate as newer generations ship. Nvidia's accelerator generations have historically repriced older inventory downward within 18 to 24 months of a successor launch.
The bull case rests on utilization. If Envision Green can lease the capacity at rates comparable to regional intelligent-computing providers, the payback period on a RMB3.8 million server is measured in quarters, not years. If it cannot, the order becomes a stranded asset on a Hong Kong-listed balance sheet with limited disclosure.
For Inventec, the order is small in absolute terms but confirms that AI server demand is broadening beyond the top four US cloud buyers — a diversification that supports the contract-manufacturing margin story. For Nvidia, a 100-server order is immaterial to revenue but adds to the count of non-hyperscale buyers, a category the company has cited as a growth vector.
Envision Green has not disclosed when the servers will be deployed or when intelligent-computing revenue will be reported separately. The next checkpoint is the company's next interim filing, where segment disclosure will show whether the RMB380 million is producing revenue or only depreciation.
This article is for informational purposes only and does not constitute investment advice.
