Nvidia shares slide again as investors fret about customers building in-house AI chips

Nvidia’s stock fell again as investors weighed fresh signs that major AI customers may increasingly try to design their own chips. The move came as traders focused on a report suggesting that another customer is developing an in-house AI chip, a development that market participants often interpret as pressure on Nvidia’s long-running role as the primary supplier of accelerated computing for generative AI workloads.

According to the Yahoo Finance report, Nvidia shares were headed toward their lowest level since April. The decline reflected growing uncertainty about whether Nvidia will continue to capture the bulk of incremental spending on AI infrastructure, or whether more customers shift parts of their computing stacks in-house as their demand matures.

The specific concern, as framed in the report, is not just that alternative chips exist, but that a broader set of customers could follow the strategy of building custom silicon for at least some training or inference workloads. When that happens, investors often reassess revenue durability for the “picks-and-shovels” suppliers tied directly to those workloads, including chip vendors and the broader supply chain that depends on Nvidia’s platform.

Nvidia’s market position is closely tied to demand for data-center GPUs and the software ecosystem that helps customers deploy AI at scale. That ecosystem includes CUDA, Nvidia’s parallel computing platform, and a range of libraries and developer tools. Even when customers experiment with custom hardware, Nvidia’s installed base and software support can remain a competitive advantage, but the possibility of additional custom chip announcements can still weigh on sentiment.

The latest selloff fits a wider pattern in which investors swing between two narratives about AI infrastructure spending. One is that the buildout of data centers and accelerated compute continues to grow rapidly, lifting suppliers. The other is that costs are so large, and competition so intense, that some customers may seek to reduce dependency and margin leakage by bringing more of the stack internally, potentially slowing the rate of incremental demand for Nvidia-branded hardware.

A key issue for Nvidia is how much of customer demand is “portable” across chip designs, and how quickly custom chips can match performance, reliability, and integration requirements. Nvidia has historically benefited from tight coupling between its hardware and software tooling, but the market reaction to any report about additional in-house development suggests investors believe the barriers to entry may be falling for at least certain use cases.

Nvidia did not provide new disclosure in the Yahoo Finance post itself beyond the market reaction, and the report did not identify the company reportedly developing an in-house chip. That leaves multiple variables unclear: the timeline for the custom chip, where it would be used (training versus inference), and whether it would replace Nvidia GPUs entirely or supplement them.

For now, traders appear to be treating additional customer in-house chip development as a announcement that competition for AI compute could broaden beyond Nvidia’s most direct rivals. The next steps to watch are any follow-on reporting that names the customer, clarifies product readiness, or describes how custom silicon affects procurement plans for Nvidia’s data-center offerings.