Nvidia is starting a pivotal week with a technical milestone that underscores its dominance in the AI hardware race. A new industry-wide benchmark from Artificial Analysis shows that the company's Blackwell Ultra NVL72 platform can run up to 60,000 simultaneous AI agents per megawatt of power—a twentyfold increase over its predecessor Hopper architecture and a clear lead over AMD's MI355X in efficiency per watt.

The metric, dubbed AgentPerf, targets a critical shift in artificial intelligence. Unlike chatbots that simply answer questions, agentic AI systems break complex goals into multiple steps, call external tools, process intermediate results and act autonomously. That workload demands a different kind of hardware performance. Nvidia's integrated approach—72 GPUs in a single GB300 NVL72 rack, combined with custom CUDA kernels and TensorRT-LLM software—is already in production use by cloud providers such as Baseten, DeepInfra and Together AI. DeepInfra, for example, runs Pam.ai, a platform that handles service bookings, phone calls and sales campaigns for car dealerships.

On Wednesday, June 24, Nvidia holds its annual general meeting online. The formal agenda includes the election of ten directors, approval of executive compensation and the reappointment of PwC as auditor. Four shareholder proposals are also up for a vote, calling for simple majority voting, greater transparency on greenhouse gas emissions from sold products, and civil rights reporting in human resources. Preliminary results will be announced during the meeting, with the final tally filed with the SEC by June 30.

Investors will be watching for CEO Jensen Huang to offer revenue guidance for the upcoming chip generations. The stock has been hovering near €183 in recent trading, roughly 10% below its all-time high set in May, after climbing nearly 47% over the past twelve months.

Behind the scenes, Nvidia is already scaling its next architecture. The Vera-Rubin platform, powered by R200 graphics chips, has begun mass production across more than 350 factories worldwide, with 150 partners in Taiwan alone supporting the supply chain. The system is designed for autonomous AI agents and processes data ten times faster than the current Grace Blackwell generation. First deliveries are slated for autumn 2026.

The ramp-up is being financed in part by a $25 billion bond issue placed in mid-June, which drew demand of around $85 billion from investors. Chief Financial Officer Colette Kress has said the company intends to return at least half of its free cash flow to shareholders, with an $80 billion stock buyback program already underway.

The expansion extends beyond hardware. Nvidia has deepened ties with South Korea's LG Group. High-level executives from LG recently visited Nvidia's headquarters to discuss broad collaboration in artificial intelligence and robotics. Nvidia will supply its Jetson-Thor platform to automate LG's complex manufacturing plants. News of the talks sent LG Electronics shares up 12%.

The surge in activity follows stellar results. In the first quarter of fiscal 2027, Nvidia posted revenue of $81.6 billion, an increase of 85% year over year. The data-center segment alone grew 92% to $75.2 billion, and GAAP net income more than tripled to $58.3 billion. For the current quarter, management expects revenue of roughly $91 billion, though that forecast excludes any data-center sales in China.

Year to date, Nvidia's stock has gained about 13%, reflecting a short-term pullback from its May peak but still leaving the company firmly in the lead as AI workloads evolve from simple text generation to autonomous, multi-step agents.