Nvidia wants to dominate every chip in AI data centers

Nvidia, the semiconductor giant, has unveiled its new Vera Rubin platform, a system that integrates CPUs and GPUs into a single design. According to Wired, this move is part of Nvidia's strategy to dominate every chip within AI data centers. The company, which already controls approximately 80% of the AI chip market according to estimates from research firm Mercury Research, now seeks to expand its dominance to other key components, such as CPUs and network interconnects.

What happened?

Nvidia announced the Vera Rubin platform, which combines its powerful GPUs with custom CPUs based on the ARM architecture, eliminating the need for third-party processors. This system, which Nvidia says will deliver unprecedented performance for AI workloads from training to inference, is expected to hit the market in 2026. The platform also includes a new NVLink 6 interconnect and a Quantum 2 switching network, allowing Nvidia to offer a complete ecosystem. This move is not isolated: in 2020, Nvidia acquired Mellanox for $6.9 billion to strengthen its networking offerings, and in 2022 it attempted to buy ARM for $40 billion, though the deal fell through due to regulatory pressures. Vera Rubin represents the culmination of that strategy: controlling all critical components of an AI data center.

Why is it important?

Historically, data centers have used CPUs from Intel or AMD alongside Nvidia GPUs. With Vera Rubin, Nvidia aims to replace those CPUs with its own, potentially reducing costs and improving performance by eliminating communication bottlenecks between chips. This poses a direct threat to Intel and AMD, which rely heavily on data center CPU sales. Intel, for example, generated approximately 50% of its revenue in 2023 from its data center business, according to its annual report. Additionally, the Vera Rubin platform could compete with AMD's integrated solutions, such as its Instinct MI300 series, which combines CPUs and GPUs in a single module. However, Nvidia has a significant advantage: its CUDA software, which is the de facto standard for AI development, with over 4 million developers according to the company. This creates an ecosystem that is difficult for competitors to replicate.

Market consequences

If Nvidia succeeds, it could further consolidate its power in the semiconductor industry. Data center customers, such as hyperscalers (Google, Amazon, Microsoft), could benefit from more integrated and efficient systems, but would also face even greater dependence on a single supplier. This could raise regulatory concerns about monopoly. In fact, in 2023, the U.S. Federal Trade Commission (FTC) launched an investigation into Nvidia's practices in the AI market, according to Reuters. Moreover, reliance on a single supplier could increase supply chain risks, as seen during the chip shortage in 2021-2022. On the other hand, hyperscalers have begun developing their own AI chips, such as Google with its TPUs, Amazon with Trainium and Inferentia, and Microsoft with Maia. Vera Rubin could accelerate this trend, as large tech companies seek to diversify their supply sources. For smaller companies, adopting Vera Rubin could be a strategic decision: gaining superior performance in exchange for greater dependence on Nvidia.

What readers should know

Nvidia is not just selling chips, but complete platforms. Vera Rubin is part of a broader strategy to offer turnkey solutions for AI, similar to what Apple does with its devices: total control of hardware and software. Investors should watch for reactions from Intel and AMD, which have already announced their own integrated platforms. Intel, for example, unveiled its Falcon Shores chip for 2025, while AMD plans to launch its MI400 in 2026. Additionally, potential antitrust actions could affect Nvidia, especially if its market share exceeds 90% in certain segments. For businesses, the decision to adopt Vera Rubin will involve weighing performance against the risk of dependency. As Nvidia CEO Jensen Huang noted at the GTC 2024 conference: "The era of accelerated computing is here, and we are building the infrastructure to make it accessible to everyone." However, that accessibility may come at a price: Nvidia's control over the entire technology stack.