NVIDIA shares have a market value of $5.27 trillion and trade at a price-to-earnings ratio of 27.6, yet CEO Jensen Huang told attendees at the Goldman Sachs Communacopia + Technology Conference on Sept. 10 that the company's growth trajectory has barely begun.
Unconstrained demand for NVIDIA's AI platforms is expanding at more than 100% annually, Huang said, while the company is projecting 70% year-over-year revenue growth. Over the trailing twelve months, NVIDIA reported $303 billion in revenue with gross profit margins approaching 75%. He described the company as "the world's first and only growth value stock."
Huang forecast that global AI infrastructure spending will reach between $3 trillion and $4 trillion by 2030. Venture capital flowing into AI-native companies totalled $400 billion in the prior six months alone, he noted. A 2-gigawatt data center commitment was announced for Australia, representing an $80 billion infrastructure investment scheduled to come online in 2027.
On pricing progression across architectures, Huang outlined that historic consumer GPUs began at $399, Hopper systems sell for roughly $18,000 each, Blackwell systems around $25,000, and Vera Rubin systems approximately $40,000. Integrated enterprise and data center GPU systems — connected through NVLink and containing about 2 million parts drawing 250 kilowatts — can cost up to $8.5 million. Each such system weighs roughly 2 tons.
Data center unit economics, evaluated over a six-year period, show revenue of about $50 billion annually per gigawatt of capacity. Huang said a 1-gigawatt facility costs roughly $60 billion, or $10 billion per year, and that systems are rented at about half of that revenue potential.
Growth in specific configurations remains steep: Grace, Blackwell, and NVLink systems in 72-rack arrangements posted a 27% month-over-month increase. Volta-class architecture continues to see rental demand a decade after its introduction, Huang added.
Application timelines were also addressed. Autonomous vehicles should show "really great progress" over the next two to three years, driven in part by a reasoning and training breakthrough called Alpamayo that reduces the need for massive raw mileage datasets. Manipulation systems targeting mid-market manufacturers could arrive in about two years, while broader physical-AI applications may take five years or more.
Huang pointed to what he called exceptional returns on platform investment, saying "we put in one and 100 comes back in," distinguishing the dynamic from concerns about circular financial flows within the sector.
NVIDIA named a broad set of enterprise and technology partners involved in its ecosystem, including OpenAI, Google, Meta, Anthropic, xAI, OpenRouter, and neocloud providers CoreWeave, Nebius, Lambda, Firmus and Nscale. OEM partners cited were Dell, Supermicro, Lenovo, HP and Cisco. Enterprise and quantitative firms on NVIDIA's platform include Jane Street, Hudson River Trading, Eli Lilly, Merck, Bristol Myers Squibb, Amazon, CrowdStrike, Palantir, Nokia, Waymo, Tesla and ASML.
