Firms ramp up efforts to compete with Nvidia in AI chips, betting on alternative hardware stacks

Nvidia’s dominance in AI chips has made its GPUs the default choice for many large-scale machine learning projects, but a growing group of companies is trying to change that outcome. A July 1 report from Quartz, citing the intensifying race for AI hardware, describes how challengers are investing heavily to develop alternatives aimed at both training and inference.

The competition is not only about chip designs. To displace Nvidia in real deployments, rivals need workable systems that include hardware, interconnects, memory and networking approaches, and software tooling that can run popular AI workloads efficiently. That means competitors are increasingly building end-to-end “platform” stacks rather than single components, according to the report’s broad framing.

The report also points to a key commercial reality driving the investment. AI adoption is expanding beyond research labs into enterprise data centers, cloud providers, and industrial deployments. In those environments, buyers are often looking for performance-per-dollar, supply assurance, and integration with their existing infrastructure, creating openings for challengers that can deliver comparable results.

Nvidia itself has benefited from a flywheel that pairs its data center GPUs with a deep ecosystem of developer software and performance-optimized libraries. That ecosystem advantage makes switching harder, because changing chip hardware can require new compilers, runtime support, and retraining of deployment pipelines. The challengers’ goal, as framed by the report, is to reduce that switching friction and offer platforms that are simpler to deploy at scale.

While the Quartz report emphasizes the push for alternatives, it does not, in the material available here, provide detailed disclosure on which specific competitors are spending the most, what exact chip roadmaps they have published, or which customers have adopted their systems at scale. As a result, it is still unclear from this account alone how quickly rivals are translating engineering progress into meaningful market share.

The next phase of the contest is likely to hinge on tangible evidence: performance benchmarks on widely used AI models, availability of production hardware, and the maturity of software support that helps developers and enterprises run workloads with minimal friction. Buyers will also watch whether new chips can keep pace as model sizes grow and as AI inference becomes a larger portion of spending over time.