Nvidia’s recent leadership change has created an opening for emerging chipmakers, particularly in Europe, to challenge the dominant position the company has long held in the artificial intelligence (AI) processor market. The departure of CEO Jensen Huang, who developed a vast ecosystem around Nvidia’s proprietary Cuda programming platform, is expected to slow the company’s aggressive customer investment strategy under interim CEO Colette Kress, while potentially eroding Nvidia’s market stronghold.
Nvidia’s near 90 percent market share in AI chips has been sustained through a combination of financial incentives to customers and its tightly integrated hardware-software approach via Cuda. However, with a less expansive funding approach now apparent, competitors and clients may find it easier to explore alternatives.
Industry observers note that some AI firms are actively diversifying their chip suppliers. For example, Anthropic indicated in its recent listing prospectus that it has significantly shortened the time required to adopt new chip platforms, now engaging multiple sources including Amazon Web Services, Google, and AMD alongside Nvidia.
Challenging the established players are numerous smaller startups, many of which are European. UK-based Fractile recently secured funding at a $7 billion valuation. Another notable player, Olix, founded by James Dacombe—the youngest self-made billionaire—has raised support from SoftBank’s Arm and is valued at $3.3 billion.
These newcomers are pursuing two strategic paths. One approach involves moving away from general-purpose processors towards specialized chips designed for specific AI tasks, akin to selecting different research labs depending on project requirements. The other bets on designing chips from the ground up, tailored to meet rapidly escalating computing demands that have surged over the past several years.
A key technical challenge lies in balancing memory capacity and processing speed. Current Nvidia chips combine two types of memory: one with high capacity but slower access, and another faster type integrated within the processor but limited in capacity. As AI models grow in complexity, overcoming this trade-off becomes increasingly vital. Fractile, for example, is innovating by placing dynamic random-access memory (DRAM) directly atop the processor to enhance performance.
Some startups, including those backed by the UK sovereign venture fund such as Callosum, are also focusing on systems that optimize how these specialized chips are orchestrated to work efficiently together.
Nevertheless, obstacles remain. Some customers express concerns about potential inefficiencies or "dead time" when specialized chips remain idle during certain operations. Additionally, many of these next-generation chips are not expected to reach production until around 2028, by which time the economics of AI may have evolved substantially.
Commercializing new chip designs can be challenging, as illustrated by the experience of Britain’s Graphcore. Once valued at $2.8 billion, the company was ultimately acquired by SoftBank for a fraction of its peak valuation.
In this context, Huang’s resignation unexpectedly removes a major barrier to more open competition in the AI chip industry. While Nvidia’s Cuda platform has long been considered a significant moat, its vulnerability appears greater than previously assumed, paving the way for a new generation of semiconductor innovators.
