英伟达GTC柏林公布与Ibiden、三星电机、欣兴等六家基板/封装/测试供应商扩产协议并推出DGX POD

Nvidia on Wednesday announced new deals with chip substrate suppliers Ibiden, Samsung Electro-Mechanics and Unimicron, to ensure it has enough component materials to produce more of its AI server chips in the coming years. It said the deals, which include a multibillion-dollar reservation with Japan-based Ibiden, would grow substrate capacity for Nvidia products by two and a half times over three years. Substrates bridge compute dies with underlying circuit boards.

At its GTC Berlin event, Nvidia also announced similar deals with Amkor for advanced chip packaging, SPIL for components of co-packaged optics—a design that puts optical components in the same package as its chips to speed up data transfer—and KYEC for semiconductor testing. The KYEC deal comes with a $1.1 billion capacity reservation, in part to help the firm build new facilities in the U.S, Nvidia said.

Such deals will help Nvidia avoid a shortage of components, and show how it’s using its balance sheet heft to maintain its revenue lead over rivals. Nvidia had committed $279 billion to supply and chipmaking capacity reservations as of the end of July, more than double the previous quarter.

When Nvidia CFO Colette Kress previously said the company’s sales would grow 70% between the current fiscal year, which ends in January 2027, and the following fiscal year, she said that such growth forecasts were already being tempered by the lack of supplies for its chips.

Additionally at the event, Nvidia introduced DGX POD, a modular data center design. Partners will assemble these units—which can contain a varying number of AI server racks—inside containers to cut down the time a customer needs to spend to get a data center up and running after the components arrive. The effort comes as some data center developers run into labor, power and other physical challenges in setting up new facilities.
New multi-billion-dollar agreements ensure Nvidia locks down key supplier capacity to boost AI chip output.