Nvidia Pours Capital Into Data Infrastructure as Cloud Giants Build Their Own Chips

The AI chip leader is placing bets on two fronts. On Tuesday, Nvidia and Google Cloud announced plans to deploy the Vera Rubin architecture at unprecedented scale, while separately the company has invested roughly $1 billion in data infrastructure specialist Vast Data. The moves underscore a strategy that extends far beyond selling silicon.

At the heart of the Google Cloud partnership are new A5X instances powered by Nvidia’s Vera Rubin technology. A single site can now link up to 80,000 GPUs, and the architecture supports as many as 960,000 Rubin chips across multiple locations. That kind of density slashes costs dramatically — inference expenses per token fall to a tenth of previous levels, while throughput per megawatt jumps tenfold. The efficiency gains are designed to meet surging demand for real-time AI agents.

Nvidia’s investment in Vast Data, confirmed on April 22, takes a different approach. The software company, now valued at $30 billion, has built an operating system for AI that combines storage, databases and computing power on a single platform. The funding round, which included Drive Capital and Fidelity, gives Nvidia a strategic foothold in the infrastructure layer. Vast Data’s revenue has tripled year-over-year, with cumulative orders exceeding $4 billion and committed annual recurring revenue above $500 million. An IPO is expected no earlier than the second half of 2026, according to insiders.

The logic behind these twin bets is clear. Amazon, Google and Microsoft are all developing their own AI chips, threatening Nvidia’s hardware dominance. By forging close ties with infrastructure providers like Vast Data and Google Cloud, Nvidia ensures its graphics cards remain the industry standard even as cloud giants explore alternatives. The company is effectively securing its position as both the dominant hardware supplier and a strategic investor in the next wave of AI.

Should investors sell immediately? Or is it worth buying Nvidia?

Hardware at this scale demands enormous manufacturing capacity. Nvidia primarily produces chips at Taiwan Semiconductor, but geopolitical risks in East Asia have prompted a backup plan. In December, the company invested $5 billion in Intel, acquiring roughly 4% of the chipmaker. Intel is developing a new fabrication technology for external clients, with mass production preparations expected to be complete by 2027. Binding decisions are likely in the second half of this year.

Nvidia’s transformation into a comprehensive infrastructure provider is reflected in its order book. At the end of January, the company held firm delivery commitments worth $95 billion. That backlog helps protect margins against rivals like AMD, which is ramping up its own capabilities and recently agreed to supply 50,000 GPUs to Oracle.

The stock market has taken note. Nvidia shares trade at around €172, roughly 90% above their level a year ago and just 4% below the 52-week high. Beyond hardware, the software ecosystem continues to expand. In mid-April, Nvidia unveiled “Ising,” an open-source model family for quantum AI. The next major milestone comes in June at Computex, where CEO Jensen Huang will present his “Five-Layer Cake” strategy covering everything from energy infrastructure to specialized AI.

Huang has been championing a shift toward agentic AI — systems that don’t just generate information but independently complete complex tasks. All of Nvidia’s software developers now work with such agents, and Huang says productivity has soared without any job losses. That transition is fueling demand for data centers and cloud capacity, with global investment in AI firms exceeding $280 billion this year alone. Nvidia is capitalizing on that boom, using its financial muscle to cement a dual role as hardware kingpin and strategic investor in the AI ecosystem’s future.