Nvidia’s $25 Billion War Chest Fuels a Pivot from Training to Inference as Ethernet Switch Sales Surge 193%

Nvidia targets AI inference demand via Vultr partnership, unveils Vera CPU, issues $25B in bonds, and raises dividend to 25 cents per share.

The artificial intelligence arms race is entering a new phase, and Nvidia is retooling its strategy to capture the next wave of demand. For much of the past two years, hyperscalers and startups alike snapped up GPUs to train ever-larger models. Now the emphasis is shifting to inference — the deployment of those models in live business environments — and Nvidia is positioning itself to dominate that transition as well.

The clearest signal comes from a partnership with Vultr, a private cloud provider that has selected Nvidia and Hewlett Packard Enterprise to upgrade its data centers. Vultr will deploy Nvidia’s new GB300 systems, which are optimized for inference workloads. Vultr CEO J.J. Kardwell noted that demand has shifted: three years ago startups drove purchases, but today established companies running customer?facing applications are the main buyers. Financial terms of the deal were not disclosed.

Nvidia is also fleshing out its hardware roadmap for this new era. At a trade show in Las Vegas, the company unveiled the Vera CPU, a processor designed for autonomous AI agents. It handles tool calls and real?time data processing, with shipments scheduled to begin in 2027. The chip extends Nvidia’s reach beyond traditional accelerators into the control plane of AI systems.

To fund these ambitions, Nvidia has tapped the bond market for the first time in three years, issuing $25 billion of investment?grade notes. Investor demand hit roughly $85 billion, more than triple the supply. Goldman Sachs and J.P. Morgan led the syndication across seven tranches with maturities stretching to 2056. While the company officially cites general corporate purposes, analysts view the war chest as ammunition for the next chapter of the AI arms race, likely financing the Blackwell architecture and the subsequent Rubin platform.

Beyond its core GPU business, Nvidia scored a strategic milestone in data?center networking. In the first quarter of 2026, it became the global revenue leader in Ethernet switches for data centers, with sales in that segment soaring nearly 193% to $2.1 billion. That gave Nvidia a 21.5% market share, thanks to its Spectrum?X platform, which ties together massive GPU clusters. The move puts traditional networking giants like Cisco and Arista under pressure, as Nvidia now offers a fully integrated AI infrastructure stack.

Shareholders will see a direct benefit next week. Nvidia’s virtual annual general meeting is set for June 24, where routine board elections are expected. More notably, the company raised its quarterly dividend from one U.S. cent to 25 cents per share, payable at the end of June. The stock has been steady, trading around €180.54 — exactly at its 50?day moving average — and has gained roughly 14% year?to?date. The 200?day line at €162.86 provides solid support, and with a relative strength index of 52, there is room for further upside. The all?time high of €202.50 is about 9% above current levels.

Geographically, Nvidia is extending its reach into Europe through an investment in French AI startup Mistral, which is using fresh credit to build out computing capacity on the continent. CEO Jensen Huang has set a long?term revenue target of $1 trillion annually. The company’s confidence is bolstered by 85% year?over?year revenue growth and a forecast that the top cloud providers will spend $1 trillion on AI infrastructure by 2027. The upcoming shareholder meeting should offer more clarity on how Nvidia plans to allocate its expanding war chest.