NVIDIA Data Center revenue hits $89.02 billion as AI agents drive token demand

The $89.02 billion Data Center revenue figure for NVIDIA's second quarter of fiscal 2027 reflects a shift in how AI models are being used. According to OpenRouter data, agentic token usage rose from roughly 0.51 trillion tokens in February to about 7.3 trillion by August, a fourteenfold increase, while human usage grew only about 2.8 times over the same stretch. This means AI agents now consume roughly five times more tokens than people, and agents use about fifteen times more tokens per request than humans, according to the same data. CEO Jensen Huang described the trend on the earnings call, saying "When the world goes to agentic, fully agentic systems, you're going to have agents running all the time, working with other agents running all the time." The revenue growth is not limited to a handful of cloud giants: hyperscale customers contributed $49 billion, up 13% sequentially, while the group covering neoclouds, enterprises, sovereign customers and industrial buyers reached $40 billion, up 25% sequentially and 138% year over year. NVIDIA is also making more per dollar of sales, with non-GAAP gross margin expanding to 75.0% from 72.5% a year earlier. However, risks remain. About 85% of agentic tokens come from cached prompts charged at lower rates, and gross margin is expected to bottom at 71% to 72% in the fourth quarter as memory costs rise. Days sales outstanding extended to 60 days from 45, and online sentiment has cooled to 37.17. Whether NVIDIA can maintain its growth hinges partly on supply: management said it has supply for about 70% of expected demand, and it expects fiscal 2028 revenue to grow about 70%, limited by supply. The next checkpoint is NVIDIA's third-quarter guidance of $108.0 billion, plus or minus 2%, excluding China data center compute, with Vera Rubin expected to make up about 20% of data center revenue in Q3. For long-term holders, the bet is that agents running around the clock will keep adding to demand, though the margin pressure from memory costs and the reliance on cached tokens are worth watching.