Nvidia faces a new test as cloud giants push further into custom AI chips

Nvidia has long been the default supplier of AI compute chips, but a new market report argues the industry is entering a more segmented phase. As large cloud providers design an increasing share of their own AI chips, the report says some semiconductor firms that support or supply that ecosystem are quietly capturing a fast-growing slice of related spending.

The report frames the shift as a competitive and economic tradeoff. Cloud operators, which run the bulk of AI training and inference at scale, have incentives to optimize performance-per-watt, reduce unit costs, and differentiate their stacks for specific workloads. Custom chips can offer those advantages, but the work requires specialized tooling, design services, and manufacturing relationships across the semiconductor supply chain.

While Nvidia remains a foundational vendor for most AI systems, the report suggests the “share of wallet” problem can emerge as customers move from buying complete accelerators toward buying more integrated, provider-designed components. That does not necessarily eliminate Nvidia exposure, but it can change the cadence of new GPU procurement and influence which companies capture incremental revenue over time.

The Yahoo Finance piece highlights two semiconductor specialists it describes as benefiting from this trend. However, the article details and company names were not included in the material available for this draft, so those firms cannot be identified here without risking inaccuracies.

Nvidia itself has continued to publicize its broad AI platform approach, spanning GPUs, networking, and software for building and deploying AI workloads. The company’s newsroom and product communications emphasize “full-stack” deployment, reflecting the reality that AI performance depends not only on chips, but also on how systems are assembled and managed in data centers.

In sector terms, the story fits a broader pattern seen in other technology cycles, where dominant platform players face pressure as large customers invest directly in custom silicon. The outcome is often less about a single winner and more about how budgets are split between off-the-shelf compute and customized components, plus the intermediaries that help customers execute those designs.

What remains unclear from the information provided for this review is the scale and timing of the alleged shift. The market report does not, in the available excerpt, provide figures for the mix of custom versus vendor chips, nor does it quantify what share of any incremental spending is flowing to the two highlighted firms.

Next, investors and industry watchers will likely look for clearer indicates in company guidance, customer procurement patterns, and supply-chain disclosures. For Nvidia specifically, the key question is whether custom-chip programs replace Nvidia purchases in meaningful volume, or whether Nvidia’s role transitions toward software and networking components that remain difficult to replicate in-house.